USD millions unless stated. Facts are in financials/facts.json; named arithmetic (prefixes snap, seg, ue) is in valuation/calculations.json as audit steps that do not feed the model. Sources: FY2025 shareholder report (csi-fy2025), FY2022 and FY2023 reports, Q2 2026 report (csi-q2-2026), the 2015 and 2021 letters (csi-2015-letter, csi-2021-capital).
Constellation (CSU) owns a large number of specialist software businesses (the reports read give no count) serving narrow workflows: municipalities, transit, utilities, hospitals, clubs, banks, media, construction and so on. Six operating groups (Volaris, Harris, Topicus, Vela, Jonas, Perseus) are aggregated into one reportable segment because each "operates essentially as a mini Constellation" (FY2025 note 23). Topicus (Europe) and Lumine (communications and media) are separately listed and consolidated through super-voting shares. Common shares trade in CAD; reports are in USD.
Value-creation mechanism
Buy, hold forever, and recycle the cash. A vertical-market software (VMS) business is bought for a price close to a small multiple of its after-tax cash; the customer keeps paying maintenance because replacing mission-critical software means data migration, retraining and operational risk; head office and the operating group impose cost discipline and best-practice benchmarking; the business's surplus cash is then sent back (or kept, under the "keep your capital" rule) and spent on the next purchase. The step where most value is created is the purchase price. CSU's own analysis of every acquisition since 2004 found that the revenue multiple paid was a far more obvious driver of IRR than post-acquisition organic growth (2015 letter, pp.5-6). Integration and margin repair matter (Altera below), but the edge is paying a price that a permanent, decentralized owner can justify and a leveraged buyer flipping in three to seven years usually cannot (2021 letter). The second step, recycling, works because the businesses need almost no tangible capital: customers prepay, so organic growth is funded by customers, and nearly all reinvestment goes into the next acquisition.
Business-model profile
Comparable firms are chosen by these traits, not by the industry label.
Growth vehicle: Acquisitions. Revenue rose from 3,969 in FY2020 to 11,623 in FY2025, about 24% a year, while FX-adjusted organic growth was 3% in FY2025 (5% for recurring revenue). Acquisition cash was 1,847, 1,683 and 1,513 in FY2023–FY2025, 16.8% of revenue. So roughly three points of growth a year come from the installed base and the rest is bought. The model therefore buys revenue every year at a measured price and grows the existing businesses separately at the installed-base rate; organic growth alone would describe a low-growth software company that CSU is not.
Capital intensity: 0.6% of sales, asset-light. Capital spending is about 0.6% of revenue: offices and hardware for staff who write and maintain vertical-market software. Tangible capital including leases is negative (−982 in FY2025) because customers prepay maintenance (deferred revenue of 70 days). The capital that drives growth is acquisition cash, about 17% of revenue a year, not capex.
Comparables. A comparable firm buys many small businesses every year with internally generated cash, needs almost no capital spending, and earns mostly recurring maintenance revenue (75% of FY2025) from customers who rarely replace mission-critical systems. Its constraint at this scale is finding enough deals at its price, so comparables are serial acquirers of small asset-light businesses, not software product companies. Not the label: "Software" suggests product-led organic growth valued on software multiples; CSU buys its growth and grows about 3% a year organically.
Declared in valuation/model.json (businessProfile). npm run check measures deal and capital intensity from the reported facts, and the outside view grades the forecast only against firms that share them.
Economic snapshot
FY2025, consolidated, reported definitions unless marked. EBITA here is revenue less operating expenses including depreciation (so leases are capitalized), before acquired-intangible amortization, FX, IRGA revaluation, finance items and impairments.
Normalization. Capex (68) roughly equals PP&E depreciation (total depreciation 201 less lease principal 131 ≈ 70), so no sustaining-capital adjustment is needed. Recently acquired businesses are only partly in FY2025: Cipal Schaubroeck alone was missing 49 of revenue (pro forma 11,672 vs 11,623). Minorities took 243 of the 1,926 pre-NCI FCFA2S (12.6%). The model's common-owner bridge (stage 7, research/07-assumption-register.md) handles these; this snapshot is the consolidated base it explains.
What the snapshot says. Return on tangible capital is meaningless for CSU: the business runs on customers' money, so the only capital that matters is the price paid for each business. Measured against that price (intangibles at cost), the portfolio earns about 19% after tax, well below the 37% on amortized book that flatters any accounting ROIC. That 19% blends decades-old purchases earning far above cost with recent ones earning less; the marginal return is measured separately (stage 6: 15.1%).
Economic segmentation
The reported segment table is one line. Re-cut by how value is created, using the supplemental Topicus, Lumine and Altera disclosures (FY2025 MD&A, PDF pp.15-17) and CSU's economic ownership (segLookThroughCfo, operating cash before interest, at accounting ownership):
So about four-fifths of CSU's owner cash comes from the small and mid VMS engine whose economics are described above. The large carve-outs (Altera, Optimal Blue, WideOrbit, Synchronoss) are a different mechanism: buy a neglected division cheaply relative to revenue, cut cost to group norms and accept shrinking revenue. They are a growing share of capital but not yet of cash. Profit by revenue type is even more concentrated (unit economics below): recurring maintenance earns roughly 85% of EBITA on 75% of revenue.
Revenue share is not profit share here: licenses and services were 22% of revenue but about 13% of EBITA, and they are the cyclical, discretionary part (licenses −8% and services −4% FX-adjusted organic in FY2025, while recurring grew 5%).
Unit economics
The unit through which CSU adds value is an acquired VMS business. Two views: a revenue-type P&L that shows where the profit sits inside every business, and observed deals that show what CSU pays and earns.
Inside the unit: profit by revenue type, FY2025 (allocations are estimates; ue* steps). Third-party costs (1,114) are allocated by software revenue; all R&D staff (1,615) is charged to recurring revenue, the adverse choice for that line; shared overhead (S&M 702, G&A 1,181, occupancy, travel/telecom/software, professional fees, other and depreciation: 3,175 in total) is allocated by revenue.
Capital per unit is the purchase price. Working capital is negative (deferred revenue exceeds receivables), PP&E is about 2% of revenue and capex matches depreciation. At the group's 2025 unlevered owner-cash margin (21.7% of revenue: operating cash after tax, leases and capex) and the measured 6.61x gross purchase multiple (stage 6; 6.57x at entry on net capital in the current model), a typical acquired business costs about 1.4x revenue and earns about 15% after tax on that price once integrated.
Observed units (FY2022, FY2023 and FY2025 acquisition notes; annual revenue from the issuer's pro forma figures):
Before and after the intervention. Every disclosed large deal arrives at roughly zero profit and is bought at 0.9-3.8x revenue; the group's acquired cash yield of about 15% means CSU lifts these units toward a cash margin of 15-25% of revenue. Altera is the cleanest case: a carve-out that earned nothing for Allscripts became a business producing about 13% FCFA2S on its price within a year, by cost restructuring (expenses were 558 on 712 of revenue in FY2025, 78%, against 73% for the group).
The failure, or the limit of the levers. The same Altera shows what cost discipline cannot do: revenue fell from about 846 to 712, professional services fell 16% organically in FY2025 and recurring revenue fell 19% in Q2 2026, partly against one-off prior-year contract revenue. A 12.7% levered yield on a shrinking base is below the group aggregate and below CSU's historical small-deal returns. The large-deal practice that the 2021 letter announced (lower hurdles for multi-hundred-million cheques) therefore produces lower and more fragile returns, as Leonard predicted.
Independent checks
- Group accounts against the unit model. Recurring revenue of 8,700 with maintenance and R&D staff of 2,838 implies a 33% staff ratio; group staff expense is 51% of revenue, and the gap is services, S&M and G&A, consistent with the allocation. The unit model's 30% recurring EBITA margin and 6% services margin sum exactly to the reported 3,077 EBITA (the allocation reconciles by construction; the test is that no line goes negative under reasonable keys).
- Adverse allocation. Charge all 3,175 of shared overhead to recurring revenue: recurring EBITA falls to 1,825, still 59% of the total, and services and licenses together would be 35%. Charging part of R&D to licenses (new products) would raise recurring's share. The conclusion that the installed-base maintenance stream is the profit pool survives every allocation tried.
- Implied price against observed prices. The aggregate 6.61x cash multiple on gross capital (6.57x at entry on net capital since the 3 November reconciliation) implies about 1.4x revenue at group margins. Observed large deals were 0.9-3.8x revenue at near-zero margins. CSU pays more per dollar of revenue for businesses whose margins it expects to raise, and less for shrinking ones (Altera). That is consistent with the 2015 letter's finding that the revenue multiple drives IRR.
- Impairment testing. Goodwill and intangibles are tested against revenue multiples of recurring revenue for each business unit (FY2025 note 10). FY2025 impairments were 43 on 15,157 of intangibles at cost (0.3%; 28 in FY2024). Few acquired units have failed badly enough to be written down, although this test is lenient by design.
- Group ROCE against marginal return. 19.3% on all acquisition cost against 15.1% measured on 2021-2025 deals (stage 6): the newer capital earns less than the stock, as expected when hurdles were lowered for larger deals.
Where the consensus disagrees. The market's live debate (2025-2026 commentary) is whether AI-assisted development lets cheap competitors replace these products, which would hit the recurring line that carries 85% of profit. The unit model says that is the right place to worry and that services and license weakness, which the market sees first, matters little to cash. The observation that would settle it is FX-adjusted recurring organic growth excluding Altera (4% in Q2 2026) and attrition commentary in the Q3 2026 MD&A on 6 November and following quarters; a move below about 0% for several quarters would mean the profit pool itself is shrinking.
Q3 2026 check — 6 November 2026
Q3 confirmed the economics above rather than changing them. Recurring revenue outside Altera grew 4% after FX (Q2: 4%), 5% outside US healthcare and Altera, 7% at Topicus and 3% in the CSU core. Licenses (−17%) and services (−5%) kept shrinking, so group organic growth was 2%. Altera's recurring decline narrowed to −8% (Q2 −19%), with sizable client departures expected to run until the end of 2027. CSU also bought two US healthcare businesses whose very high customer attrition it says it modeled into the price, a reminder that the purchase yield, not the target's growth, is the decision variable. Source: Q3 MD&A pp.4–5, 16–17.