Mark Leonard's Constellation Software: lessons from 27 years of shareholder letters

Executive overview

Mark Leonard took $25 million in 1995 and compounded it to a $40 billion business by acquiring hundreds of small vertical market software (VMS) businesses and never selling them. VMS customers rarely leave — 4% annual attrition implies a 26-year average relationship — and acquired businesses generate free cash flow that funds more acquisitions without reinvestment.

The operating architecture is radical decentralisation: autonomous units averaging 44 employees, no central command, and a systematic preference for small over large. The durable advantage is not the acquisition model — anyone with a checkbook can copy that — but a decade-deep culture of running experiments across hundreds of businesses and rapidly spreading what works.

Owning many small, permanent, cash-generative software businesses and compounding their best practices outperforms chasing scale.

The acquisition strategy

• Target: private VMS businesses, typically $2–4M, preferably still founder-run
• Founder-owned businesses have long-term orientation baked in — employee selection, customer relationships, and product quality all reflect it
• Second source: distressed assets spun out of large corporates after failed synergies — best opportunities arrive during recessions
• Objective is permanent ownership; minority stakes in public software companies are also acceptable when prices are attractive
• Tracks thousands of prospects; signals interest early so owners think of Constellation when they are ready
• Avoided debt for most of the company's history; used it tactically in 2008–2009 when competition evaporated and prices fell sharply

Why VMS businesses compound

• Organic revenue growth requires almost no incremental capital
• Annual software cost rarely exceeds 1% of a customer's revenues — low price, high dependency
• Low attrition (~4%/year) creates customer relationships lasting 26+ years
• Free cash flow funds acquisitions; no need to return capital or reinvest inside a single industry
• The longer Constellation owns a business, the larger and better it becomes

Decentralisation as competitive moat

• Leonard's personal distaste for centralised authority shaped the entire operating model
• Business units stay small by design — data showed almost zero correlation between size and performance
• Large units are split into smaller ones; original manager keeps the larger piece, a protege runs the spinoff
• Do not share sales, R&D, or HR across units — allocated costs are never controllable at the business level
• Highly talented people will not stay inside centralised command-and-control structures
• Illinois Tool Works under John Nichols ($369M to $4.2B in 14 years, 365 units by 1996) validated the model externally

Managing initiatives: fixing the R&D spending problem

• "Initiatives" (new products or new markets) consumed over half of all R&D and sales spending by 2005
• Problem: no clear hypothesis, no defined test, no single person accountable — spending dragged indefinitely
• Fix 1: require each initiative to state an explicit assumption and the conditions that would falsify it
• Fix 2: assign one dedicated initiative champion — part-time leadership killed accountability
• Result: number of new initiatives proposed dropped sharply; capital allocation improved markedly
• Took six years to change the mental models of a generation of managers

The learning machine advantage

• 199+ separately tracked business units provide a live laboratory for cheap, fast business-process experiments
• A practice that works in a handful of units is benchmarked and spread rapidly across the whole organisation
• Best practices tend to be counterintuitive — competitors without the same experimental base cannot replicate them
• Anyone can copy the acquisition model; nobody can copy a decade of compounding best-practice data

Small is beautiful: the philosophy

• In small teams everyone knows each other, trust is high, rules are few, focus stays on customers
• As organisations grow, policies multiply, talented people leave, innovation stalls
• Complexity and coordination costs increase faster than headcount — scale creates its own drag
• The goal: offer business managers autonomy, mastery of VMS skills, and a human-scale team
• Leonard's advice to his 35-year-old self: stay put, master your VMS business, it generates more than enough wealth

People and compensation

• Promoted from within almost exclusively — trust and loyalty take years to build; manipulative hires take years to identify
• In 2014, reduced his own salary to zero and stopped charging expenses — wanted to be a partner, not an employee
• Compensation tied solely to share ownership eliminates the principal-agent problem
• Wrote letters about the business, not the stock