Hg Capital Portfolio

Seven places where the published record says something other than the pitch. The pitch itself is in Source - Hg approach and thesis pages; each item below links to the evidence note that carries the source. Each links to the evidence note that carries the source.

1. The buy-and-build case is built on a study that excludes Hg's shape

BCG and HHL's 2016 work is the origin of nearly every "buy-and-build outperforms" line in private equity marketing. Its own table shows platforms with one or two add-ons returning 35.5% IRR and platforms with more than two returning 19.9% — below the 23.1% standalone benchmark. See Buy-and-build with three or more add-ons underperformed standalone buyouts. The only study that decomposed the premium found it vanishes once multiple arbitrage is stripped out: Buy-and-build outperformance disappears without multiple arbitrage.

2. Two headline figures exceed Hg's own audited accounts

Marketing says $195bn of aggregate enterprise value and 140,000 employees. The audited HgCapital Trust report says $185bn and above 130,000. Hg's marketed enterprise value runs ahead of its audited figure; Hg's portfolio headcount is above 130,000, not 140,000.

3. The returns headline has no primary source, and the independent numbers are lower

"3.0x gross MOIC, 31% gross IRR" appears on one marketing page, undated and unfootnoted, alongside a 3.3x on another page and 3.8x in a capital markets deck. Hg's 3.0x gross MOIC has no primary source. Three public pension funds publish net figures: 10–22% net IRR, 1.1–2.25x net TVPI. Independent LP data puts Hg net IRR between 10 and 22 percent.

4. The academic foundation of the moat argues against it

Farrell and Klemperer's handbook chapter is the standard citation for switching costs. Its conclusion is that switching costs need not generate supernormal profits, because competition to acquire the customer can dissipate the entire later rent. Switching costs need not generate supernormal profits. And "system of record", the phrase the whole category rests on, was coined by Geoffrey Moore to mean commoditised and finished as a source of differentiation — the opposite of how investors use it. Moore coined system of record to mean commoditised.

5. Regulatory demand has already been withdrawn once in this portfolio

Omnibus I, signed off in February 2026, removed roughly 80% of companies from CSRD scope. Omnibus I removed 80 percent of companies from CSRD scope. The US analogue is further along: the SEC proposed rescinding its climate disclosure rules in May 2026. Two other mandates in the portfolio are near completion rather than beginning — Medicaid electronic visit verification is roughly 90 percent complete and Italy's e-invoicing mandate is the completion-cliff exemplar.

6. The AI scoreboard is budgeted, self-reported, and one number is a misquote

The €260m is actually $260m, it is budgeted, it was collected by surveying portfolio company management, and Hg's own footnote says it is "not a forecast". Hg's AI EBITDA figure is budgeted and self-reported. The "2x uplift in AI-enabled exits" is a corruption of "two exits at ~100% uplift", and uplift to carrying value is not a return: Hg's two AI exits were uplifts to carrying value, not returns.

7. The metric yardsticks are softer than they sound

The Rule of 40 has no author and no derivation; Median private SaaS net revenue retention is 101 percent, not 120; and the revenue-per-employee model this vault itself uses for 31 companies carries ±40–50% error and is invalid for brokerage and services businesses altogether — Revenue per employee carries 40 to 50 percent error.

8. Almost none of this portfolio was ever funded by anyone else

Twenty-nine of the 59 took no external capital at all before the private equity deal, and only four have a fully disclosed funding history. Twenty-nine of the fifty-nine never raised external capital.

That is not a criticism — it is the strongest single piece of evidence for Hg, and it supports the sourcing explanation over the category one. It also means the commercial databases are close to useless here, and wrong in a specific, repeatable way: Data feeds report HgCapital Trust's tranche as company funding.

What this does not say

None of the above says Hg is a poor investor. Saturn 1 returned 21.2% net IRR over seven years, which is strong upper-quartile software private equity. What it says is that the explanation Hg gives for the returns — category selection into mission-critical regulated software — is the part with no evidence behind it, and that Hg's edge is sourcing and execution, not category selection is the better-supported reading.

Hg's website states "$195+ billion in aggregate enterprise value" across 60+ portfolio businesses.

HgCapital Trust plc's audited FY2025 Annual Report, published March 2026, states a portfolio of more than 60 companies "worth >$185 billion in aggregate enterprise value". The FY2025 results presentation says the same. The FY2024 audited figure was ">$160bn".

The marketing number runs roughly $10bn ahead of the audited one.

Why it matters here

The trend is real and strong — $160bn to $185bn in a year is a substantial increase. The gap is small in percentage terms. What it establishes is a pattern, and the pattern recurs: see Hg's portfolio headcount is above 130,000, not 140,000, where the same comparison runs about 8% the other side of the audited figure.

It is also worth remembering what aggregate enterprise value measures. It is not Hg's money. It includes portfolio company debt — HgT reports 7.4x net debt/EBITDA and a 72% equity / 28% debt structure — plus all co-investor and minority equity. A firm that owns half of a $10bn company reports $10bn.

Sources

HgCapital Trust FY2025 Annual Report (PDF) · Hg AI data source page

Hg's website states "140,000+ employees" across portfolio companies.

HgCapital Trust's audited FY2025 Annual Report states ">130,000 employees globally".

Why it matters here

An 8% overstatement on a soft metric is not, by itself, important. It matters because of what it sits next to: Hg's marketed enterprise value runs ahead of its audited figure on the same page, and Hg's 3.0x gross MOIC has no primary source two lines further down.

The test this vault applies to any single figure is whether the person quoting it can name the document it comes from. For the audited numbers that is straightforward. For the marketing page it is not — no footnotes, no as-at dates, no methodology.

Where a figure appears in both places, use the audited one.

Sources

HgCapital Trust FY2025 Annual Report (PDF) · Hg Approach

Hg's marketing states: "25-year track record; generating a gross MOIC of around 3.0x and a gross IRR of over 31%... $30bn of proceeds returned from all technology and services businesses since 2001."

The triplet appears on one marketing page. It has no as-of date, no footnote and no methodology. It does not appear in HgCapital Trust's audited accounts for FY2024 or FY2025, nor in the PSERS or New Jersey board papers.

Worse, Hg publishes at least three different headline multiples without reconciling them:

• 3.0x / 31% — the AI data source page.
• 3.3x — the investors page, "on all realised software and services investments", traceable to a 31 March 2023 datapoint and apparently not refreshed.
• ~3.8x / ~34% — a capital markets day deck, on 34 liquidity events since 2022.

Different populations, presented interchangeably.

The $30bn is directionally supportable: Hg's own 2023 piece said $17bn at 31 March 2023, and HgT confirms ">£12 billion of total realisation proceeds over the last two years" plus $3.4bn of Hg-wide proceeds in 2025.

All figures are gross. Hg's own deck carries the disclaimer: returns "are gross and do not include the effects of fees, commissions and other charges."

Why it matters here

Anyone quoting "Hg returns 3.0x at 31%" is quoting an undated, unaudited, unspecified-population gross figure that Hg's own website contradicts twice. The independently verified alternative is Independent LP data puts Hg net IRR between 10 and 22 percent.

Sources

Hg AI data source · Hg, Built to last · Capital markets day deck (PDF)

Three US public pension funds publish net-of-fee performance on Hg funds. These are the only independent numbers on Hg's record.

CalPERS, as at 31 December 2025:

New Jersey SIC, Saturn series as at 30 September 2024: Saturn 1 (2018) 21.2% net IRR / 2.25x net TVPI / 1.36x DPI; Saturn 2 (2020) 21.0% / 1.5x / 0.30x; Saturn 3 (2022) 20.1% / 1.14x / 0.0x DPI.

PSERS, as at 30 September 2021: blended Hg position 28.3% net IRR / 2.0x.

How to read this

The gross-to-net gap is large. Hg markets 31–33% gross IRR and 3.0–3.3x gross MOIC. The best-seasoned fund with independent net data — Saturn 1, seven years in — is 21.2% net and 2.25x. That is strong upper-quartile software private equity. It is roughly a third below the marketed headline.

Fund-level "net IRR" is not one number. Saturn 3 appears at 20.1% (New Jersey, Sept 2024) and 10.5% (CalPERS, Dec 2025), explained partly by commitment timing and vintage labelling.

Early-vintage figures are noise. PSERS recorded Genesis 9 at 334.5% and Saturn 2 at 176.7% in 2021 — J-curve artefacts on tiny called capital. Saturn 2 settled at 21.0% by 2024. Anyone citing the three-figure numbers is citing an accounting artefact.

DPI is the tell. Saturn 2 had returned 0.30x by late 2024, Saturn 3 0.0x; CalPERS shows zero cash returned on Genesis 10 and Saturn 3.

Why it matters here

This is the honest version of Hg markets gross returns, and the reason the vault's overall verdict is that Hg is a good investor whose explanation of why is the weak part — see Hg's edge is sourcing and execution, not category selection.

Sources

CalPERS PEP fund performance · NJ Division of Investment · PSERS board resolutions

Hg's page reads: "AI-Enabled Exits: 2x AI-enabled exits (GTreasury and Intelerad) at ~100% average uplift to book value."

The "2x" is the count of exits — two of them — not a 2x uplift. The popular restatement, "2x average uplift to book value", collapses the count into the multiple and double-counts. The claimed uplift is ~100%, unfootnoted.

The two transactions, from HgCapital Trust's regulated filings:

Two problems. First, 97% and 62% do not average to ~100% — they average about 80% simple, 74% weighted. Hg's fund-level carrying values may differ from HgT's, and nothing published lets you check.

Second, and more important: an uplift to carrying value is not a return. It measures the gap between a conservative internal mark and what a strategic buyer paid. A large uplift is as much evidence of conservative marking as of value creation — see Conservative carrying values and exit uplift. Neither announcement disclosed a MOIC or IRR, and a 97% uplift over 2.5 years and a 62% uplift over 5 years are very different annualised outcomes.

Neither announcement attributes any portion of the uplift to AI. The GTreasury release says Hg's AI expertise "helped accelerate" product development; GE valued Intelerad's InteleGence platform. Both buyers were strategics buying category position.

There is no independently verified case anywhere of AI materially raising an exit multiple in PE-owned software — no academic study, no auditor's opinion, no buyer disclosure isolating an AI component of price.

Sources

HgT on GTreasury · HgT on Intelerad · Hg, Driving AI transformation

This is the vault's own reading, offered as an inference and marked accordingly. It is not Hg's stated position.

The argument

Every element of Hg's stated thesis is either untested or contradicted. Mission-criticality has no published measure (No study operationalises mission-criticality); switching costs do not reliably produce rents (Switching costs need not generate supernormal profits); regulatory demand can be repealed (Omnibus I removed 80 percent of companies from CSRD scope); and buy-and-build past two add-ons underperformed in the study that founded the category (Buy-and-build with three or more add-ons underperformed standalone buyouts).

Yet the returns are real. Saturn 1 at 21.2% net over seven years is upper-quartile software private equity by any measure.

If the stated explanation does not hold and the results do, the explanation is somewhere else. The candidates the record supports:

A population nobody else had priced. Twenty-nine of the fifty-nine never raised external capital — twenty-nine holdings went from founder control straight to private equity with no venture round in between. You cannot buy at a discount in a market where every asset has already been bid, and this is a portfolio assembled almost entirely from assets that never were.

Sourcing density. Twenty-plus acquisitions in accounting software means Hg sees the twenty-first before anyone else and prices it better. This is the mechanism Multiple arbitrage describes, and Buy-and-build outperformance disappears without multiple arbitrage shows it is where the premium actually lives.

Cadence, not volume. Serial acquirer returns decay in blocks, not with count — steady programmatic acquisition does not degrade returns; bursts do. Hg's cadence is steady.

Hold length. Visma since 2006, IRIS since 2004. Twenty-year holds convert a moderate compounding rate into a large multiple, and avoid paying the transaction cost of an exit and re-entry.

Entry into cash-generative businesses at scale, which supports leverage, which amplifies a moderate operating result.

None of that is as attractive to write on a website as "mission-critical software in regulated markets". All of it is better supported.

What would change this view

A published decomposition of Hg's realised returns separating entry multiple, leverage, organic growth, acquired growth and exit multiple. It does not exist for Hg, and Hg does not disclose deal values means it probably cannot be assembled from outside.

Links: Multiple arbitrage, Programmatic M&A, The cluster strategy is concentration, not diversification.

Hg does not state this one as a principle on its approach page, which is itself telling — it is the mechanism behind almost every holding, and it appears in the marketing mainly as a description of individual companies.

The counts are extraordinary. Visma has made roughly a hundred acquisitions; The Access Group around sixty; team.blue around forty; insightsoftware around thirty; Septeo around twenty-five; Howden Group Holdings, IRIS and The Citation Group around twenty each.

Assessment

This is where the evidence disagrees most sharply with the practice.

The founding study of the whole category — BCG and HHL Leipzig, 2016 — is the source of the "buy-and-build outperforms" claim. Its own data shows platforms with one or two add-ons at 35.5% IRR and platforms with more than two at 19.9%, below the 23.1% standalone benchmark. See Buy-and-build with three or more add-ons underperformed standalone buyouts. Hg's largest holdings are ten to fifty times past the point where that study's support ends.

The one piece of work that decomposed the premium found it disappears entirely once multiple arbitrage is removed: Buy-and-build outperformance disappears without multiple arbitrage. And Software value creation is 42 percent multiple expansion and 6 percent margin says the same thing from the other direction — software buyout returns are not operational stories.

Two qualifications in Hg's favour, and they are substantial. Serial acquirer returns decay in blocks, not with count — the academic finding is that returns degrade when acquisitions come in bursts, not as the cumulative total rises, which is an argument for steady programmatic cadence of exactly the kind Platform and bolt-on describes. And The 70 percent M&A failure rate is the wrong sample for bolt-ons — the famous failure statistic is drawn from large public deals and does not describe small tuck-ins, where success rates are far higher.

The failure mode to watch is not the count but the integration. Upland Software shows what 31 unintegrated acquisitions produce: thirty-one technology stacks, shallow integration of billing and SSO only, and a 94% share price decline.

Who meets it

29 of 59 are platform acquirers or carry five or more documented bolt-ons.

Links: Programmatic M&A, Multiple arbitrage, Can a hundred-acquisition roll-up beat the add-on evidence, Platform and bolt-on.

This is the most important evidence note in the vault.

BCG and HHL Leipzig, The Power of Buy and Build (Brigl, Jansen, Schwetzler, Hammer, Hinrichs), February 2016. This study is the origin of essentially every "buy-and-build outperforms" claim in private equity marketing.

Measured: 2,372 deals exited 1998–2012 from 126 of the most active PE firms, narrowed to 121 deals with complete IRR and value-driver data.

The headline numbers, which get quoted:

• Buy-and-build 31.6% average IRR against standalone 23.1%.
• Small platforms under $70m EV: 52.4%. Large platforms over $290m: underperformed.

The qualification that gets dropped, and it is decisive:

• Deals with one or two add-ons returned 35.5% IRR.
• Deals with more than two returned 19.9% — below the 23.1% standalone benchmark.

A second dropped qualification: the 121-deal sample contains no US companies (disclosure rules), is ~90% European, includes only exited deals, and does not disclose whether IRRs were verified or self-reported.

Why it matters here

Visma has made roughly 100 acquisitions. The Access Group roughly 60. team.blue roughly 40. insightsoftware roughly 30. Septeo roughly 25.

The study that founded the strategy describes platforms with one or two add-ons, under $70m of enterprise value, in Europe, before 2013. It is not evidence for a hundred-acquisition software compounder — and on its own numbers, the population Hg's largest holdings belong to underperformed doing nothing.

Right mechanism, wrong population. See Can a hundred-acquisition roll-up beat the add-on evidence for the strongest counter-case, and Serial acquirer returns decay in blocks, not with count for the finding that is genuinely favourable to Hg.

Sources

BCG, The Power of Buy and Build (PDF)

160,000+ organisations across Europe, USA and APAC; 9,700 staff. Among the most prolific acquirers in UK software.

Hg cluster: Map - ERP and Payroll. SMB & mid-market business management, headquartered in United Kingdom, founded 1992, held since 2018 — 8 years. Scale implies the Hg's three fund families Saturn tier, though Hg does not disclose which fund holds what.

Revenue. €1,358.5m (2025), as reported, unconverted or a prior year.

Capital raised. Never raised external capital. Four successive buyouts since 2010. The GBP 9.2bn is enterprise value; the >GBP 1bn is M&A debt. That is the normal shape in this portfolio — see Twenty-nine of the fifty-nine never raised external capital. Source

What it sells

• Access Evo — AI layer spanning navigator, researcher, workflow and assistants
• Access HR and Payroll — Payroll, HRIS, recruitment, performance and learning
• Access Financials — Cloud accounting, expense, purchase order and group consolidation
• Access Legal — Case management and compliance for law firms
• Access Health and Social Care — Care planning, rostering and clinical records
• Access PaySuite — Direct Debit, card and open banking payments embedded in Access software

Bought by Finance, HR and operations. On the capability map that puts it in Offering - Payroll, Offering - HR and workforce management, Offering - ERP and business management, Offering - Legal practice software, Offering - Care operations, Offering - Embedded payments — see Map - The offering taxonomy.

The broadest overlap set in the portfolio: touches Visma, IRIS, Prophix, Septeo, Litera and Citation.

How it makes money

A saas business selling to CFO & Finance in the smb, mid-market segment, through inside sales. Regulatory driver scores 2 of 3 — compliance is a material but not dominant driver. Defensibility is coded as: Workflow lock-in, Switching cost, Product breadth. On the consolidation axis it is a platform acquirer with 60 documented acquisitions, and its reach is global.

AI posture is AI-embedded. That is Hg's own framing and should be read as positioning until the product evidence is independent — see Hg's AI EBITDA figure is budgeted and self-reported.

Where it sits in the portfolio

Sits in the SMB digital backbone stack alongside team.blue, Focus Group, JTL, CINC Systems, Parte, Athletic Sport Sponsoring — the same budget holder, adjacent product surface.

Closest on the encoded schema vector: insightsoftware (0.80), Sovos (0.72). That similarity is computed, not asserted — see Map - The quantitative schema.

It meets 7 of Hg's eight stated tests: Deterministic, repeatable workflows, Subscription revenue with high retention, White-collar professional end-markets, Buy-and-build consolidation, SMB and mid-market cloud migration, AI-native transformation, Cross-border scaling from a home market.

Leadership

Led by Chris Bayne, co-ceo, in post since 2026, alongside Jon Jorgensen under a co-chief-executive model.

What the record says

• Buy-and-build with three or more add-ons underperformed standalone buyouts
• The Access Group was valued at 9.2 billion pounds in 2022

Links: Map - ERP and Payroll, Deterministic, repeatable workflows, Chris Bayne, Map - The offering taxonomy.

Heisig, Kick & Schwetzler (HHL Leipzig), Buying Performance? The Impact of Multiple Arbitrage in Buy-and-Build Strategies, SSRN, 30 January 2022. 161 buy-and-build transactions with valuation detail on the add-ons.

The authors isolate an add-on sourcing effect — buying small companies at lower multiples and revaluing them at the platform's multiple.

• The sourcing effect contributes roughly 8% to equity value CAGR.
• "When this multiple arbitrage effect was removed, buy-and-build outperformance decreased significantly to the levels of their non-B&B peers."
• Preliminary evidence that exit buyers do not distinguish between organic, inorganic and sourcing-driven EBITDA growth.

Why it matters here

It relocates the explanation for buy-and-build returns from operations to arithmetic. If the premium is sourcing rather than synergy, then what produces the return is deal flow, pricing discipline and the spread between small-company and platform multiples — which is exactly the case made in Hg's edge is sourcing and execution, not category selection.

It also identifies the mechanism's expiry condition. The arbitrage persists because the next buyer does not price the difference. That works until buyers discriminate, or until the multiple spread closes — and HgT's NAV fell 4.9 percent in H1 2026 on multiple compression is what pressure on that spread looks like.

Sources

SSRN 4169612

Thibaut Morillon, Serial acquirers and decreasing returns: do bidders' acquisition patterns matter?, Financial Review 56(3), 2021, 407–432.

The finding: declining acquirer returns occur mostly within "blocks" of acquisitions — bidders who acquire "quickly in chunks". Active acquirers without that clustered pattern show no return deterioration. Proposed mechanisms: temporary overvaluation, agency costs, and bidder learning.

The break is cadence, not cumulative count.

Why it matters here

This is the single most favourable piece of independent evidence for Hg's model, and it is the right answer to Buy-and-build with three or more add-ons underperformed standalone buyouts.

If returns decayed with the number of acquisitions, Visma at ~100 and The Access Group at ~60 would be indefensible. Morillon's result says the count is not the variable — burstiness is. A platform acquiring steadily, several times a year, through a standing capability, sits on the right side of this finding. That is what Platform and bolt-on and Programmatic M&A describe, and it is consistent with McKinsey's 3.6-deals-a-year programmatic median.

What it does not excuse is integration failure, which is a separate mechanism — see Upland Software shows what 31 unintegrated acquisitions produce.

Sources

Financial Review 56(3)

Upland Software made 31 acquisitions since 2010. As of August 2025:

• Share price down 94% — $52 peak in 2019 to $2.55.
• $228m goodwill impairment in 2024.
• Revenue fell from $317m (2022) to $275m TTM.
• Net income −$113m in 2024, against $234m of debt.

The mechanism named in the analysis: 31 technology stacks (Java.NET, Ruby, Python, legacy COBOL), 31 user interfaces, 31 deployment models, with "shallow integration" of billing, back-office and single sign-on only — and customers migrated onto what the analysis calls inferior products.

Why it matters here

This is the failure mode for Platform and bolt-on, and it is invisible from outside. Nothing in Hg's disclosure distinguishes a genuinely integrated platform from a holding company with a shared logo. The counts at The Access Group, insightsoftware, Litera and Septeo are in the same range as Upland's.

The academic mechanism behind it is Buyouts concentrate R&D on focal technologies: private-equity-owned firms do not cut research and development, they concentrate it on the core — which is precisely what strands the fortieth acquisition's customers on a maintenance track.

The question to ask any of these platforms is not how many acquisitions, but how many codebases, how many UIs, and how many customers are on a product that no longer receives roadmap.

Sources

31 acquisitions, 31 tech stacks, 94% stock decline

The question this vault cannot settle, and the one that decides most of the portfolio's value.

Why it is open

BCG and HHL's founding study of buy-and-build shows platforms with more than two add-ons returning 19.9% IRR against 23.1% for standalone buyouts — see Buy-and-build with three or more add-ons underperformed standalone buyouts. Visma has made roughly a hundred acquisitions, The Access Group sixty, team.blue forty.

Either the study does not generalise, or a large part of this portfolio is built on a strategy that underperforms doing nothing.

What was offered in answer, and why it falls short

Hg does not address it. The industry answer is that the study is old, European, exit-only and small — 121 deals on the returns claim — which is true and does not produce a counter-finding. Absence of evidence against is not evidence for.

Candidate answers from outside

1. The count is the wrong variable. Serial acquirer returns decay in blocks, not with count finds decay occurs in bursts of acquisitions, not as the cumulative total rises. Steady programmatic cadence shows no deterioration. This is the strongest available defence, and Hg's cadence fits it.
2. The mechanism is arbitrage, and arbitrage does not exhaust. Buy-and-build outperformance disappears without multiple arbitrage says the premium is sourcing. If so, the hundredth acquisition is as good as the first provided the multiple spread holds — and the constraint becomes spread, not count.
3. The sample excludes the winners. Constellation Software, Roper, Vitec and Lifco are public serial acquirers with hundreds of deals and excellent long-run returns, none of them in a PE exit sample. The counter is Upland Software shows what 31 unintegrated acquisitions produce — the same strategy, executed badly.
4. Integration is the discriminating variable, and it is unobservable from outside. Buyouts concentrate R&D on focal technologies gives the mechanism by which acquired products get stranded.

What would settle it

A count, per platform, of live codebases, distinct UIs, and customers on products no longer receiving roadmap. That is knowable inside each company and published by none of them.

Links: Platform and bolt-on, Buy-and-build consolidation, Multiple arbitrage.

Across the portfolio, the single most repeated word in Hg's own copy is compliance. Thirty of 59 holdings score the maximum on the regulatory driver in this vault's schema.

The position: statutory obligation produces demand that is non-discretionary, recurring, and immune to the customer's budget cycle.

Assessment

Split the mandates into two kinds, because they behave completely differently — and Hg's copy does not distinguish them.

Annuity mandates. Transaction-level reporting and periodic recertification produce revenue forever. ViDA makes transaction-level reporting permanent from 2030; Making Tax Digital expands the paying population to 2028; SOC 2 annual Type II audits, ISO 27001 three-year cycles, FedRAMP continuous monitoring, and CMMC Level 2 certification becomes mandatory from November 2026 with annual affirmation. These are genuinely durable, and Sovos, IRIS, Bright and A-LIGN sit on them.

Spike mandates. Implementation revenue that ends when the population is compliant. Italy's e-invoicing mandate is the completion-cliff exemplar — live since 2019, exemptions closed 2024, work done. Medicaid electronic visit verification is roughly 90 percent complete — 43 states plus DC compliant, which is HHAeXchange's entire founding market, finished.

And then the third category Hg's thesis does not allow for: withdrawn mandates. Omnibus I removed 80 percent of companies from CSRD scope. The SEC proposed rescinding its climate disclosure rules in May 2026. HTI-5 proposes deleting 34 of 60 health IT certification criteria, with the regulator itself claiming $1.53bn of savings — savings that are, arithmetically, somebody's lost software revenue.

Regulation also arrives as a cost rather than a market: The EU AI Act high-risk tier was deferred to December 2027.

The theoretical frame is The economic theory of regulation, and the empirical warning is HITECH produced 800 EHR vendors, not a shakeout: a mandate reliably creates demand, and just as reliably creates the entrants who compete it away.

Links: Mission-critical systems in regulated markets, What replaces the ESG reporting line after Omnibus I, Where does HHAeXchange grow once EVV is complete.

George Stigler, The Theory of Economic Regulation, Bell Journal of Economics and Management Science, 1971 — work for which Stigler later received the Nobel Prize.

The central claim: regulation "is acquired by the industry and is designed and operated primarily for its benefit." Regulation functions as a barrier to entry, because compliance cost is largely fixed and therefore falls hardest on small entrants while being trivial for incumbents.

Direct evidence of regulation-generated spend exists in the Sarbanes-Oxley §404 cost studies — the SEC's 2009 study and GAO-25-107500 (2025).

Why it matters here

It is the correct theoretical frame for Regulation is the demand engine, and it is more precise than Hg's version. A mandate is a fixed cost imposed on everyone in a market, which is why compliance software is bought universally and price-insensitively, and why the vendor with the content library already built has a structural advantage over one building it.

It bears on Ncontracts, Cube Global, CTAIMA and Empyrean Solutions — all selling the discharge of an obligation rather than a productivity gain.

The limit of the theory for an investor: Stigler predicts protection for the regulated industry, not for its vendors. HITECH produced 800 EHR vendors, not a shakeout is what happens when that distinction is ignored.

Sources

Fifty-year assessment of Stigler 1971, Public Choice · SEC study of SOX 404 costs (PDF)

Italy's B2B e-invoicing mandate has been live since 1 January 2019 (Law 205/2017), clearing through the Sistema di Interscambio run by Agenzia delle Entrate. The forfettari exemption — the last carve-out — was fully removed on 1 January 2024. It operates under Council Implementing Decision (EU) 2024/3150, authorised to 31 December 2027.

Why it matters here

Italy is what a mandate looks like once it is finished. Every business is on it; the implementation and onboarding revenue is exhausted; what remains is per-transaction processing and maintenance at a much lower run rate. The next new work is the 2027 derogation renewal and then the ViDA makes transaction-level reporting permanent from 2030 re-platforming in 2035.

That is the shape every other national mandate in the portfolio is heading toward — France's e-invoicing mandate went live in September 2026 and Germany's e-invoicing issuance duty completes in 2028 each have a hard cliff at their final phase date. It is the reason Regulation is the demand engine has to distinguish annuity mandates from spike mandates, which Hg's own copy does not.

Sources

EU eInvoicing country sheet, Italy 2025

France's e-invoicing obligation took effect on 1 September 2026 — two weeks before this vault was built. From that date every business must be able to receive electronic invoices, and large and mid-sized companies must issue them and transmit e-reporting data. 1 September 2027 extends the issuing obligation to SMEs and micro-enterprises.

The timetable slipped from an original 2024/2026 plan. DGFiP has said it will apply the first phase with "bienveillance et tolérance".

Why it matters here

Directly relevant to MyUnisoft and Septeo, both selling into French accounting and legal practices at exactly the moment their clients must change how they invoice, and to Sovos as a cross-border compliance engine.

It is a spike rather than an annuity for the implementation layer: the onboarding revenue concentrates in 2026–27 and then falls to per-transaction processing. That timing is worth holding against MyUnisoft's 62% growth rate, which is measured off a small base in a year when the mandate was approaching.

Sources

economie.gouv.fr — facturation électronique · DGFiP practical guide

Section 12006 of the 21st Century Cures Act (P.L. 114-255) requires electronic visit verification for all Medicaid personal care services (deadline 1 January 2020) and home health care services (deadline 1 January 2023), with incremental FMAP reductions of up to 1% for non-compliance.

On CMS's own tracker, 43 states plus the District of Columbia were compliant for home health care services as of 1 January 2024. Six states and two territories were not.

Why it matters here

This mandate is HHAeXchange's entire founding market, and it is essentially finished. Remaining growth must come from displacing incumbent state aggregators, raising attach rates, and underlying home-and-community-based services volume — not from new mandate coverage.

It is the portfolio's clearest completion cliff, and the subject of Where does HHAeXchange grow once EVV is complete. HHAeXchange's 2024 acquisitions of Sandata, Cashé and Generations read differently in this light: consolidation of an addressable market that has stopped expanding.

Sources

CMS EVV guidance · CMS state-by-state HHCS compliance tracker

The Corporate Sustainability Reporting Directive (EU) 2022/2464 originally covered roughly 50,000 undertakings in three waves. Two cuts followed:

• Directive (EU) 2025/794 ("stop the clock") postponed waves two and three by two years.
• Omnibus I, signed off by Council on 24 February 2026, set the final scope at more than 1,000 employees and more than €450m turnover, removed listed SMEs and financial holdings, exempted wave-one companies for FY2025 and FY2026, and moved transposition to 26 July 2028 with full compliance by July 2029.

On the Commission's own numbers this removes "around 80% of companies" from scope, with roughly €6.3bn of annual administrative cost savings.

Why it matters here

This is the clearest case in the portfolio of regulatory demand being withdrawn, and it happened within four years of the directive's adoption. The addressable population for ESG reporting software shrank roughly fivefold and the deadline moved out two to three years.

It bears directly on Lucanet, which markets ESG reporting as a platform capability, and on the ESG modules at Ideagen and Optro. It is the empirical core of the objection in Mission-critical systems in regulated markets and the subject of What replaces the ESG reporting line after Omnibus I.

The US analogue runs the same way: the SEC ended its defence of the climate disclosure rules in March 2025 and formally proposed rescinding them in May 2026.

Sources

Council sign-off, 24 Feb 2026 · Commission on the 80% figure · SEC rescission proposal

Ebix Inc. was insurance-vertical software: regulated end market, mission-critical to its customers' operations, sticky, embedded in workflow. It is the thesis fully expressed.

It filed for Chapter 11 bankruptcy in December 2023, following short-seller attacks and leverage pressure.

Why it matters here

Every characteristic the category screens for was present, and none of them prevented the outcome. What failed was not the market position — it was leverage, accounting quality and acquisition integration, which are exactly the variables Mission-critical systems in regulated markets does not address.

It is the reference case for Which holding is the portfolio's Ebix, and the reason PE-backed companies defaulted at roughly twice the market rate belongs in the same reading. A regulated vertical software business with 7x leverage and an acquisition programme is not automatically safe; it is a specific risk profile with a specific failure mode.

Hg's insurance exposure is brokerage rather than software — Howden Group Holdings, GGW Group, Group Induver, Ascendia Gruppe, Fonds Finanz — so Ebix is not a direct comparable. It is a comparable for the thesis, not the sector.

Sources

Sidley on the Ebix Chapter 11 filing

Ebix was the regulated vertical archetype and filed Chapter 11 — insurance vertical software, regulated, mission-critical, sticky, embedded in workflow, bankrupt in December 2023. Every screen the category applies was satisfied.

Why the question is worth asking

Not because any Hg holding is in trouble. There is no public evidence of distress at any of the 59, and none should be inferred from this note. The question is which profile carries the Ebix failure mode, so that the right things get watched.

Ebix failed on leverage, accounting quality and acquisition integration — none of which are addressed by Mission-critical systems in regulated markets.

The risk profile, stated as a screen

1. High leverage against partly-acquired EBITDA. Portfolio-wide this is 6.9x net debt/EBITDA at H1 2026, and PE-backed companies defaulted at roughly twice the market rate.
2. A high acquisition count with unobservable integration — the Upland Software shows what 31 unintegrated acquisitions produce profile.
3. Demand resting on a mandate that is complete or repealed — Omnibus I removed 80 percent of companies from CSRD scope, Medicaid electronic visit verification is roughly 90 percent complete.
4. A metric set that looks strong because little is disclosed — see Is the schema measuring quality or disclosure.

What this vault can and cannot see

It can see acquisition counts, coded regulatory dependence and portfolio-level leverage. It cannot see company-level debt, customer concentration, covenant headroom, or how many codebases sit under a platform. Those are the four things that would actually answer the question, and none of them is public.

The honest answer is that the screen identifies a shape, not a name.

Links: Refinancing at the portfolio company, PE-backed companies defaulted at roughly twice the market rate, Map - Where the evidence contradicts the thesis.

Hg's own eight clusters are organised by end market — who the customer is. This layer is organised by capability — what the software actually does. The two cut across each other, and the gap between them is where the interesting links are.

How it was built

Every holding's product pages were read and its named products recorded — PestPac, octoplant, IrisX, Lito, Prophix One, LogaHR, Quantios Core, RegPlatform, not "field service software". 292 named products across 59 companies.

Each company was then tagged with two to five capability tags from a controlled list. A tag becomes an Offering note when two or more holdings share it: 37 of the 43 tags qualified. The six singletons — telecoms, life sciences data, post-trade, PLM and CPQ, social impact, subscription mobility — are recorded on their company notes but have no node, because a category with one member is not a category.

What it shows that the cluster map does not

The cluster map puts OneStream and Visma together under Tax & Accounting because both serve finance people. The capability map separates them — OneStream does Offering - Financial close and consolidation for group finance, Visma does Offering - Accounting practice software and Offering - Payroll for small businesses — and instead puts OneStream next to insightsoftware, Lucanet and Prophix, which sit in a different cluster.

That regrouping is the point. See The portfolio contains four competing CFO platforms.

The densest capabilities

What a shared capability does and does not mean

Two companies on the same Offering note are not automatically competitors. Several are separated by geography (Bright in the UK, Blinqx in the Netherlands, MyUnisoft in France), by customer size (OneStream enterprise, Prophix mid-market), or by whether they sell software or the service built on it (Quantios supplies the software, Waystone performs the administration). Where the collision is genuine, the company note says so.

Links: Map - The quantitative schema, Map - Where the evidence contradicts the thesis, Start Here.

Reading the product pages rather than the cluster labels turns up an overlap Hg's own taxonomy hides.

All four sell consolidation plus planning to the office of the CFO. Three of them sit in Hg's Tax & Accounting cluster; the fourth also does. The cluster label does not reveal the collision because it describes the buyer, not the product.

The sharpest pair is Lucanet against insightsoftware's IDL and Cubeware, which compete directly in the German mid-market — the same country, the same size of customer, the same product.

Serrala and Diamant Software sit adjacent to the same buyer without being direct competitors, and Empyrean Solutions sells planning and treasury capability that only banks buy.

Why this is not necessarily a problem

The defensible reading is deliberate tiering: OneStream at the top, Prophix in the middle, Lucanet in DACH, insightsoftware as the multi-brand aggregator underneath. Owning the whole ladder in a market you understand is a coherent strategy, and it is what Cluster specialisation and repeat-sector investing would predict.

Why it might be

Three things follow that the marketing does not address.

Shared fate. Four assets exposed to one buyer's budget cycle is concentration, not diversification — see The cluster strategy is concentration, not diversification. When software multiples compressed in the first half of 2026, they compressed together: HgT's NAV fell 4.9 percent in H1 2026 on multiple compression.

The exit problem. A strategic buyer for one is likely a competitor of the others, and a sponsor buying two would face the same question. What is the exit path for 185 billion dollars of enterprise value.

Internal rationalisation is the obvious value-creation move and the hardest to execute. Buyouts concentrate R&D on focal technologies is the mechanism by which the losing platform's customers quietly stop receiving roadmap — the Upland Software shows what 31 unintegrated acquisitions produce failure mode, applied across companies rather than inside one.

The same pattern, less acutely, runs through governance and risk — Optro, Mitratech, Ncontracts and Ideagen all sell risk registers, third-party risk and regulatory change, separated mainly by vertical.

Links: Map - The offering taxonomy, Offering - Financial close and consolidation, Offering - Planning and forecasting.

Hg's newest and largest 2026 entry (~$6.4bn). 1,700+ customers including 18% of the Fortune 500.

Hg cluster: Map - Tax and Accounting. Corporate performance management, headquartered in USA, founded 2012, held since 2026 — 0 years. Scale implies the Hg's three fund families Genesis tier, though Hg does not disclose which fund holds what.

Revenue. €601.9m (2025), as reported, unconverted or a prior year.

Capital raised. $561.0m across all disclosed rounds. 2021 Series B $200m (D1 Capital) plus ~$361m primary proceeds from the July 2024 Nasdaq IPO. The Nov 2024 follow-on was entirely selling shareholders. Source

What it sells

• Financial Close & Consolidation — Unified consolidation with journals and account reconciliation
• Financial Planning & Analysis — Budgeting, forecasting and scenario modelling
• Operational Planning — Driver-based workforce, revenue and cash flow plans
• Tax Provisioning — Tax provision calculation inside the close
• SensibleAI and Agentic Finance — Machine-learning forecasting, anomaly detection and finance agents

Bought by Office of the CFO. On the capability map that puts it in Offering - Financial close and consolidation, Offering - Planning and forecasting, Offering - Financial reporting, Offering - Tax compliance, Offering - Agentic AI products — see Map - The offering taxonomy.

The enterprise tier of a four-way internal collision with insightsoftware, Lucanet and Prophix.

How it makes money

A saas business selling to CFO & Finance in the enterprise, mid-market, public sector segment, through partner / channel. Regulatory driver scores 2 of 3 — compliance is a material but not dominant driver. Defensibility is coded as: Workflow lock-in, Ecosystem, Switching cost. On the consolidation axis it is a organic with 1 documented acquisitions, and its reach is global.

AI posture is AI-embedded. That is Hg's own framing and should be read as positioning until the product evidence is independent — see Hg's AI EBITDA figure is budgeted and self-reported.

Where it sits in the portfolio

Sits in the Office of the CFO stack alongside insightsoftware, Lucanet, Prophix, Serrala, Empyrean Solutions, Scopevisio, Diamant Software — the same budget holder, adjacent product surface.

Closest on the encoded schema vector: Serrala (0.95), Prophix (0.86), Lucanet (0.79). That similarity is computed, not asserted — see Map - The quantitative schema.

It meets 5 of Hg's eight stated tests: Deterministic, repeatable workflows, Subscription revenue with high retention, White-collar professional end-markets, AI-native transformation, Cross-border scaling from a home market.

Leadership

Led by Tom Shea, co-founder & chief executive officer, in post since 2010.

What the record says

• OneStream cost 6.4 billion dollars at 10.6 times trailing revenue

Links: Map - Tax and Accounting, Deterministic, repeatable workflows, Tom Shea, Map - The offering taxonomy.

Hg describes eight specialist verticals as a strength: depth of sector knowledge, repeat pattern recognition, cross-portfolio learning through HIVE and the portfolio community.

That is true, and it is worth stating the other half plainly: eight clusters of European regulated B2B software is one bet, not eight.

The argument

The clusters share a buyer type (a licensed white-collar professional), a demand driver (regulation), a business model (subscription workflow software), a value-creation engine (Platform and bolt-on), a currency exposure, an interest-rate exposure and an exit market. When software multiples move, they move together — which is exactly what happened in the first half of 2026: HgT's NAV fell 4.9 percent in H1 2026 on multiple compression, where +11% of trading performance was overwhelmed by −13% of multiple compression across the whole book.

The counter-argument, which is strong, and is set out in Cluster specialisation and repeat-sector investing: concentration is the source of the edge. A firm that has bought twenty accounting software businesses knows what the twenty-first is worth better than a generalist does, and can act faster. Hg's edge is sourcing and execution, not category selection makes that case, and the evidence supports it more than it supports the category thesis.

But an LP holding Hg alongside other software funds should price it as a single, highly correlated software position with leverage, not as a diversified portfolio of 59 companies. The 59 is a diversification of idiosyncratic risk, not of systematic risk.

Links: Map - Where the evidence contradicts the thesis, What is the exit path for 185 billion dollars of enterprise value, HgT's NAV fell 4.9 percent in H1 2026 on multiple compression.

HgCapital Trust H1 2026 results, published 14 September 2026 — two days before this vault was built.

• NAV 530.7p per share at 30 June 2026, down from 561.5p at FY2025.
• −4.9% NAV total return for the half.
• Trading contributed +11%. Multiple compression contributed −13%.
• Top-20 EV/EBITDA fell from 25.2x to 22.9x, attributed to public software comparables weakening on AI disruption fears.
• Net debt improved to 6.9x EBITDA from 7.4x.
• Underlying LTM revenue +16%, EBITDA +19%, 34% margins.
• H1 deployment £146m; H1 gross proceeds £134m.

Why it matters here

This is the whole thesis under stress in a single set of numbers. The portfolio traded well — 16% revenue growth, 19% EBITDA growth, margins holding — and the value fell anyway, because the multiple moved.

That is the practical meaning of Software value creation is 42 percent multiple expansion and 6 percent margin: if 42% of software buyout value creation is multiple expansion, a period of compression removes it regardless of operating performance.

It is also the correlation cost of The cluster strategy is concentration, not diversification. Fifty-nine companies did not diversify the exposure, because they all mark against the same public comparables.

And the stated cause — public software weakness on AI disruption fears — is the market pricing the risk in Does AI attach to seat pricing or erode it, against a portfolio whose owner is simultaneously marketing AI as its value-creation engine.

Sources

HgT H1 2026 results · HgT trading update

Lifetime disclosed funding was researched for all 59 current holdings in September 2026. The distribution is the finding.

Only 13 of 59 have any figure at all, totalling roughly $4.7bn — and $1bn of that is a single Hg equity cheque into insightsoftware whose primary/secondary split was never disclosed.

The thirteen with a number

The twenty-nine with nothing to report

Ascendia Gruppe, Athletic Sport Sponsoring, Azets, Blinqx, Bright, Caseware, Diamant Software, Fonds Finanz, Group Induver, IRIS, JTL, Litera, Lucanet, P&I, Parte, Payworks, Pirum, Prophix, Quantios, Revalize, Rhapsody, Septeo, Serrala, Sovos, The Access Group, The Citation Group, Trackunit, Workwave, team.blue.

These are founder-owned, family-owned or sponsor-created businesses. Caseware was founder-owned for 32 years. Bright for 27. Diamant Software for 45. Pirum, Prophix, Serrala, Lucanet, Payworks, Fonds Finanz and Group Induver all went from founder control straight to private equity without a venture round in between.

A further seventeen — AMDT, CINC Systems, CTAIMA, Cube Global, FE fundinfo, Focus Group, GGW Group, Gen II, HHAeXchange, IFS, Mitratech, Ncontracts, Nomadia, Norstella, Scopevisio, Visma, Waystone — took institutional capital where no round was ever sized.

Why this matters more than the number

It is the best available evidence for Hg's edge is sourcing and execution, not category selection. A business that never raised venture capital was never priced by a venture market, never ran a competitive process for growth capital, and in many cases had never been valued by an outsider before the sponsor arrived. That is precisely the population where Multiple arbitrage works — you cannot buy at a discount to a public comparable in a market where every asset has already been bid.

It also reframes what "growth equity" means here. Almost none of this portfolio's growth was funded by primary capital. It was funded by cash flow and by acquisition debt — which is consistent with Software value creation is 42 percent multiple expansion and 6 percent margin and with the leverage in Refinancing at the portfolio company.

And it is a warning about the data. Because so little is disclosed, the commercial databases fill the gap with the wrong number — see Data feeds report HgCapital Trust's tranche as company funding.

What is deliberately not counted

Acquisition prices, enterprise valuations, secondary stake purchases, LBO acquisition debt, and HgCapital Trust's own co-investment tranche. Each of those circulates as a "total raised" figure for at least one company in this portfolio, and none of them is capital raised. See Hg does not disclose deal values.

What not to repeat as fact, and where this vault is thin.

The confidence field

Every note carries confidence: high | medium | low. High means a primary or audited source. Medium means a single credible press source, a working paper, or a proposed rule not yet final. There are no low notes in this vault; where a claim would have been low, it was left out or written as an open question.

What is solid

• The roster, clusters, countries and partnership years. Taken from Hg's own portfolio page.
• Every HgCapital Trust figure. Audited or RNS-disclosed: NAV, carrying values, realisation proceeds and uplifts, EV/EBITDA, leverage. These are the most reliable numbers here.
• The regulatory instruments. Directive numbers, Federal Register citations and effective dates, from primary legal sources.
• The academic findings. Named, dated, with sample sizes.

What is soft, and how

• Six holdings returned no third-party match at all, so carry no KPI data: AMDT, Ascendia Gruppe, Athletic Sport Sponsoring, Cube Global, Fonds Finanz, Group Induver and Parte.

• Criticality and churn are one provider's banded tags, not disclosed company metrics — the most useful qualitative fields in the dataset and the least verifiable.

• Septeo and team.blue returned legal-entity figures far below group scale and were suppressed rather than published.

• 31 of 59 revenue figures are modelled from headcount at ±40–50% error. Flagged on every note. Revenue per employee carries 40 to 50 percent error.

• Currency is not normalised. Several figures are GBP or USD as reported. The basis flag says which.

• One figure is known to be distorted. P&I's €307.9m against 216 employees is a legal-entity filing; it is flagged rather than dropped.

• Capital raised is empty for 46 of 59. Twenty-nine never raised; seventeen raised amounts that were never published. Nine of the thirteen figures are floors. Twenty-nine of the fifty-nine never raised external capital.

• Ignore any funding figure from a commercial database for these companies. Nine of them carry HgCapital Trust's co-investment tranche as their "total funding" — Data feeds report HgCapital Trust's tranche as company funding.

• Bolt-on counts are floors, assembled from published announcements. Hg does not disclose deal values.

• Criticality and churn bands are one vendor's tags, not company disclosures.

• The qualitative coding is judgement, derived from Hg's own marketing copy — so it inherits Hg's framing of what each business does.

• Capital tier is inferred from revenue and headcount. Hg does not disclose fund allocation. Hg's three fund families.

• Thesis-twin edges are computed, not asserted. Two companies can be near-identical on the vector and unrelated in the world.

• Named products are current as of September 2026 and taken from each company's own product pages. Product lines change faster than anything else in this vault.

• Capability tags are judgement, applied from a controlled list against those product pages. A shared tag means an overlapping capability, not necessarily competition — Map - The offering taxonomy sets out the distinction.

• Chief executives were verified against company leadership pages, not taken from a provider. Eighteen contradicted what a provider would return. Three are low confidence: Norstella (the company site and trade press disagree), Parte and CTAIMA (neither publishes a leadership page).

Numbers that should not be repeated as fact

• "$195bn enterprise value" — audited figure is $185bn. Hg's marketed enterprise value runs ahead of its audited figure.
• "140,000 employees" — audited figure is above 130,000.
• "3.0x gross MOIC, 31% gross IRR" — no primary source, contradicted twice on Hg's own site, and gross.
• "Focus Group, £800m" — one pre-announcement press report.
• "Visma €19bn IPO" — an FT estimate, for a listing now deferred to 2027.
• "Howden £50bn" — a 2030 ambition, not a valuation.
• "2x uplift in AI-enabled exits" — a misreading of "two exits at ~100% uplift", and uplift is not a return.
• "€260m of AI EBITDA" — it is $260m, budgeted, survey-sourced, and disclaimed by Hg as not a forecast.

Where the research itself is thin

No published data exists on customer retention or NPS in software roll-ups; no study operationalises mission-criticality; no auditable quantification of seat-pricing erosion under AI; and no independently verified case of AI raising an exit multiple. Those are stated as gaps in the relevant notes rather than filled.

Links: How this vault was built, Is the schema measuring quality or disclosure, Where the roster disagrees with Hg's website.

This vault models revenue for companies that disclose none, using revenue per employee. This note is the error bar on that method, and the reason several companies are left blank instead.

Benchmarks. SaaS Capital (>1,000 private SaaS, 2026): overall median $141,125 ARR per FTE; $1–3m ARR → $109,644; $5–10m ARR → $152,295 equity-backed against $177,240 bootstrapped. Benchmarkit 2025: $50–100m ARR → ~$200k; above $100m → ~$300k. Public software spans roughly $100k–$400k.

This vault's portfolio median, from the 19 companies with reported figures, is €153k.

Why the spread is ~3x within "software", for reasons unrelated to quality:

1. Funding model — bootstrapped runs ~15% higher at identical ARR.
2. Scale — $110k at $1–3m ARR against $300k above $100m: a 2.7x swing from size alone.
3. Services mix — implementation revenue is recognised in full but carries headcount at 20–40% gross margin, so software-plus-services businesses show deflated RPE.
4. Gross versus net revenue — brokerage, marketplace and payments businesses booking gross rather than net under ASC 606 can show RPE ten times a SaaS peer with no efficiency difference at all.
5. Offshore and contractor headcount, and whether contractors count as FTEs.

Guidance: expect ±40–50% error at one sigma for pure software, and treat software-plus-services and brokerage/marketplace businesses as out of model.

The framework note Revenue per employee as a scale measure carries the full benchmark set.

Why it matters here

31 of 59 companies in this vault have no disclosed revenue. The correction this evidence forced: modelled revenue is now refused entirely for brokerage/MGA, tech-enabled services, software-plus-services and hosting businesses — so Howden Group Holdings, GGW Group, Group Induver, Ascendia Gruppe, Fonds Finanz, Parte, team.blue and Azets-type businesses show no revenue figure rather than a wrong one.

Every remaining modelled figure is flagged on its company note. None should ever be averaged with a disclosed figure.

Sources

SaaS Capital revenue-per-employee benchmarks · Benchmarkit 2025

Every one of the 59 company notes carries the same typed property set, so that two holdings can be compared without an argument about what the words mean. The schema is also the thing most likely to mislead you, which is why this note exists.

Eighteen quantitative fields

Capital raised — lifetime disclosed external funding in USD, with its own basis flag: disclosed, partial (a floor), none disclosed, or no external funding. Only 13 of 59 carry a figure and 29 never raised anything, which is itself the finding — Twenty-nine of the fifty-nine never raised external capital. Acquisition prices, valuations, secondary purchases, LBO debt and HgCapital Trust's own tranche are all excluded, because each of them circulates as a raise for at least one holding — Data feeds report HgCapital Trust's tranche as company funding.

Then: revenue and its year, with an explicit basis flag; year-on-year growth; EBITDA and gross margin; The Rule of 40; employees; customers; churn band; G2 rating and review count; documented bolt-ons; and four derived 0–100 scores — retention, scale, consolidation intensity and thesis fit.

The basis flag is the important one. It takes four values: reported actual, as reported (unconverted currency or a prior year), legal-entity filing (a statutory figure that overstates per-head economics — P&I is the clearest case), and modelled. Only 19 of 59 have a reported figure. The estimator behind the modelled ones is set out in Revenue per employee as a scale measure.

The offering and people layers

Two further layers sit alongside the schema. Offerings records each company's named products and tags it with two to five capabilities, producing 37 shared-capability nodes — Map - The offering taxonomy. People records the chief executive of each holding together with the source that confirmed the name, because commercial providers were wrong on a large minority of them.

Twelve qualitative fields

Business model, buyer persona, customer size, regulatory driver (0–3), criticality, moats, AI posture, go-to-market, consolidation role, geographic scope, sub-vertical and inferred capital tier. Each is drawn from a closed vocabulary, which is what makes the similarity computation below meaningful rather than decorative.

The derived scores, and what they hide

Retention score blends the churn band with the criticality tag. Both come from a third-party dataset with its own methodology, and 24 companies have neither.

Scale score is a log blend of revenue and headcount. Where revenue is modelled it is partly circular.

Consolidation intensity combines the coded consolidation role with documented bolt-on count. Bolt-on counts are floors, not totals — they come from published announcements, and Hg discloses very little (Hg does not disclose deal values).

Thesis fit is the count of Hg's eight tests a company passes, expressed out of 100. It measures conformity to a stated strategy, not quality, and it is the field most vulnerable to Campbell's law.

The similarity edge

Each company is encoded as a 60-dimension vector — one-hot on every qualitative field plus normalised quantitative fields — and the four nearest neighbours by cosine distance become thesis twin links. This is the one relationship in the vault that no human asserted. It surfaces genuine pairs the cluster structure hides, and it also happily links two companies that share a business model and nothing else, so read the similarity score rather than the line.

What the schema cannot see

Entry price — which is the number that most determines the return and is never disclosed. Also contract structure, pricing model, net revenue retention (nobody discloses it — Median private SaaS net revenue retention is 101 percent, not 120), debt at the company level, customer concentration, and the difference between recurring and merely repeating revenue (Quality of revenue). It also cannot see the thing that most determines the return, which is entry price.

Read The schema and its gameability before using any of this to rank anything.

The question this vault has to ask about itself.

Why it is open

Every company note carries a thesis fit score, a scale score, a retention score and a consolidation intensity. Those look like measurements of the business. They are partly measurements of how much the business discloses.

• 19 of 59 have a reported revenue figure. 31 are modelled from headcount, with ±40–50% error (Revenue per employee carries 40 to 50 percent error). Nine have neither.
• 35 of 59 have a retention score, because the rest have no churn band and no criticality tag.
• 11 of 59 have both inputs for The Rule of 40.
• Zero disclose net revenue retention, which is the metric that would matter most (Median private SaaS net revenue retention is 101 percent, not 120).

Campbell's law predicts exactly this: the axes most disclosed — growth, headcount — dominate the ranking over the axes that matter most but are least disclosed. The result is a ranking that substantially measures disclosure quality.

What has been done about it

Three mitigations, applied:

1. The basis flag. Every revenue figure records how it was derived, so a modelled number can never be mistaken for a reported one.
2. Refusal rather than estimation. Revenue is not modelled at all for brokerage, services and hosting businesses, where the estimator is invalid. Those notes show nothing instead of something wrong.
3. Missing means missing. No field is imputed to a sector average.

What has not been done

The re-scoring test. Campbell's law suggests re-running the composite under alternative weightings: if rank order moves materially, the schema is measuring noise. That has not been run, and it should be.

Nor has the validation holdout — deliberately keeping one metric out of the scored composite to test it against the rest.

The honest position

The qualitative coding is the more reliable half of this vault, because it is derived from descriptions every company publishes. The quantitative scores are useful for sorting and dangerous for ranking.

Links: Map - The quantitative schema, The schema and its gameability, Confidence and provenance.

The 59 companies in this vault come from Hg's own content API. Cross-checking against HgCapital Trust's regulated announcements shows the page is behind.

Listed as current, but sold

• Quantios — full exit to Vista Equity Partners announced 29 July 2026, HgT proceeds £13m at a 31% uplift, completing Q3 2026. Still listed as a current holding on Hg's portfolio page as at 16 September 2026. Its note in this vault is retained, with the exit recorded. See Quantios was sold to Vista Equity Partners in July 2026.

Sold, and correctly absent

• Geomatikk Group — full exit announced 30 March 2026, HgT proceeds £20.4m at only a 3% uplift, buyer not disclosed.
• GTreasury (Ripple, >$1bn, October 2025) and Intelerad (GE HealthCare, $2.3bn, November 2025) — both correctly moved to Exited.

Held, but not on the portfolio page

• Street Group — £7m HgT investment, 22 July 2026.
• Nourish Care — £17m HgT investment, 18 August 2026.

Neither appears in the CMS. They are therefore absent from this vault, and the true current count is likely 60, not 59, net of the Quantios exit.

Partial changes not reflected

• Septeo — partial exit in 2026.
• Refinancings at Lucanet, AMDT, Fonds Finanz and Ncontracts.

What to do about it

Treat the roster here as accurate to Hg's CMS as at 16 September 2026, not to the world. For anything that has moved, HgCapital Trust's RNS announcements are the authoritative and current source — Source - HgCapital Trust plc disclosures.

This is not carelessness on Hg's part so much as a structural feature: Hg does not disclose deal values and the marketing site is not a regulated disclosure surface. It is a reason not to use it as one.

The pipeline

1. Roster. Hg's portfolio page is a Next.js front end over a content API. Querying it for companies at stage Current returns 59 companies with cluster, country, founded year, partnership year and long descriptions. Two carry a second cluster: Bright and The Citation Group.
2. Quantitative layer. Company names were resolved and hydrated against a commercial private-company metrics platform in batches. 53 of 59 matched; the six that did not are named in Confidence and provenance.
3. Schema. Seventeen quantitative and twelve qualitative fields defined, with closed vocabularies for every qualitative field. See Map - The quantitative schema.
4. Qualitative coding. Each company coded against the schema from Hg's own descriptions. This is the step with the most judgement in it, and it inherits Hg's framing.
5. Research fan-out. Seven claim clusters, one research agent each, all dispatched at once: Hg's own claims; the vertical software canon; buy-and-build evidence; metric benchmarks; the specific regulations; AI value creation; and the transaction record. Each brief demanded a primary source, an explicit provenance verdict including the negative one, and a URL per claim.
6. Authoring. 59 portfolio notes generated from the structured data; the remaining ~120 written from the research.
7. Validation. vault_stats.py for link integrity, orphans, dead ends and density; check_links.py before the graph build.
8. Publication. Built to a single self-contained HTML file and deployed.

Later enrichment passes

Capital raised — lifetime disclosed external funding for all 59, which turned up Twenty-nine of the fifty-nine never raised external capital.

Commercial offerings — every company's product pages read and its named products recorded, then tagged by capability. 292 named products, 37 shared-capability nodes.

Chief executives — LinkedIn via the Apify harvestapi/linkedin-profile-search actor, which exhausted its free-tier run limit after nine runs; the remainder were resolved from company leadership pages and every name was verified there regardless.

What the research changed

Three findings altered the data rather than just commenting on it:

• Revenue modelling was narrowed. Revenue per employee carries 40 to 50 percent error established that the estimator is invalid for brokerage, services and hosting businesses. Modelled revenue was withdrawn from eight companies rather than published wrong.
• Two figures were suppressed entirely. Septeo and team.blue returned legal-entity filings far below group scale.
• The roster was found to be stale. Quantios was sold to Vista Equity Partners in July 2026, and two 2026 investments are missing from Hg's own page. See Where the roster disagrees with Hg's website.

What is missing

No interviews. Everything here is published material. A single conversation with an Hg operating partner or a portfolio CFO would be worth more than another week of desk research, particularly on integration depth — which is the unobservable variable in Can a hundred-acquisition roll-up beat the add-on evidence.

Links: Confidence and provenance, The schema and its gameability, Start Here.

What to pick up next, roughly in order of value.

Cheap and high value

1. Add Street Group and Nourish Care. Two 2026 holdings missing from Hg's own page — see Where the roster disagrees with Hg's website. Both have HgT announcements with investment size.
2. Run the reweighting test described in The schema and its gameability. One afternoon, and it tells you whether the scores mean anything.
3. Record the Quantios exit in its company note rather than only in the evidence note.
4. Pull HgT's H1 2026 interim report in full. It was published 14 September 2026, two days before this vault, and only the announcement summary has been read.

Worth real effort

1. Company-level debt. Companies House and national registries carry statutory accounts for the UK, German, French, Dutch, Belgian and Nordic holdings. Leverage per company is the largest missing field, and it is the one that would make Which holding is the portfolio's Ebix answerable rather than rhetorical.
2. Bolt-on counts from primary announcements. Current counts are floors from published deals. A systematic sweep would firm up Can a hundred-acquisition roll-up beat the add-on evidence.
3. The exited 64. Hg's CMS holds 64 exited companies with the same fields. Hold periods and outcomes for those would allow an actual base rate rather than a snapshot — and would be the single biggest improvement available to this vault.

Needs a person, not a search

1. Integration depth. How many codebases, how many UIs, how many customers on unroadmapped products, at The Access Group, insightsoftware, Litera and Septeo. Unobservable from outside and decisive. One conversation with a portfolio CTO would settle more than a month of desk work.
2. Net revenue retention. Nobody discloses it. An operating partner would know it for the top 20.
3. Whether the AI products renew. Hg's AI EBITDA figure is budgeted and self-reported resolves in the 2027–2029 exits, and earlier in renewal data nobody publishes.

Watch items with dates

• November 2026 — CMMC Phase 2 begins; A-LIGN's steepest demand curve.
• 2027 — Visma's London listing, if it holds.
• 1 September 2027 — French e-invoicing extends to SMEs.
• 1 January 2028 — German e-invoicing becomes universal.
• 2 December 2027 — EU AI Act high-risk obligations, as deferred.
• 26 July 2028 — CSRD transposition under Omnibus I.

Links: Start Here, How this vault was built, Confidence and provenance.