Accel London AI Hackathon, 15 October 2026

Rollupa

We buy compliance firms from owners ready to retire, and make them worth more.

Taylan Durmus, founder

Problem

In London alone, 218 compliance firms are run only by directors over sixty.

218 of 722

Fire safety and security firms in Greater London trading 15+ years, from our run on Companies House data. They are too small for private equity, so the owner's options are to close, sell cheaply to a rival, or keep working.

Demand

Demand is set by law and barely moves with the economy.

JFMAMJJASONDFire alarm serviceBS 5839-1, at least every 6 monthsFire alarm service: month 6Fire alarm service: month 12Lift thorough examinationLOLER 1998, every 6 months for passenger liftsLift thorough examination: month 6Lift thorough examination: month 12Emergency lighting testBS 5266-1, monthly, plus an annual full testEmergency lighting test: month 1Emergency lighting test: month 2Emergency lighting test: month 3Emergency lighting test: month 4Emergency lighting test: month 5Emergency lighting test: month 6Emergency lighting test: month 7Emergency lighting test: month 8Emergency lighting test: month 9Emergency lighting test: month 10Emergency lighting test: month 11Emergency lighting test: month 12Fire extinguisher serviceBS 5306-3, every 12 monthsFire extinguisher service: month 12Landlord gas safety checkGas Safety Regulations 1998, every 12 monthsLandlord gas safety check: month 12

Under the Fire Safety Order, whoever runs a building must keep its fire systems maintained. Inspections recur on fixed intervals for landlords, schools, care homes and offices, whatever the economy does.

Why now

Software is largely solved. Businesses with vans and engineers are next.

The physical world is nextFirms that work in real buildings were too small and too varied for software vendors. Agentic workflows fit them as they are.
The back office is agent workInvoicing, chasing payment, scheduling, certificates and quotes are rules plus paperwork.
Owners are retiringRetirement is the most cited reason UK SME owners give for selling.
Investors are readyGeneral Catalyst, Thrive, Elad Gil and Khosla back AI-enabled rollups.
Product

An agent deal team inside a deterministic pipeline.

S0-S1ScreenCompanies House,directors 60+S2TriageReal firm,real websiteS3Deal teamLead, analyst,AI ops, structurerS4Grader25 checks,source re-fetchS5Committee3 personas,blind votesS6OutputsMemo, pipeline,draft letterfix loopsend-back with questionsPython owns the stages, retries, state and resume. Agents do the judgement inside each stage.Every stage checkpointed: a killed run resumes where it stopped

Agents write only through typed deal-file tools. Every claim has a source or a derivation. Valuations, structures and returns come from tested finance functions, never from the model.

Execution at scale

The risk in a rollup is pace. Agents make each deal fast and cheap.

What 50 completions a year asks of the engine

Deep dives: 334 a yearDeep dives334 a yearApproved offers: 100 a yearApproved offers100 a yearCompletions: 50 a yearCompletions50 a yearModel cost of the deep dives: about $782 a year
27 min

for 10 full deep dives in the recorded run, at $2.34 of model cost each.

50 completions a year needs about 334 deep dives, roughly 6 a week, at the run's 30% approval rate and an assumed half of offers completing.

Proof: one recorded run

From 4,226 firms to 3 approved offers in 27 minutes, for $24.61.

Active firms, 15+ years: 4,226Active firms, 15+ years4,226In Greater London: 722In Greater London722Every director 60+: 218Every director 60+218After exclusions: 53After exclusions53Shortlisted: 40Shortlisted40Kept by triage: 19Kept by triage19Deep dives: 10Deep dives10Approved: 3Approved3Bar length on a log scale
Deep dive 1: ApprovedDeep dive 2: ApprovedDeep dive 3: ApprovedDeep dive 4: RejectedDeep dive 5: RejectedDeep dive 6: RejectedDeep dive 7: RejectedDeep dive 8: RejectedDeep dive 9: RejectedDeep dive 10: Rejected3 approved7 rejected by the committee

Live Companies House data, Greater London. 7 committee send-backs answered, 50 challenges between agents resolved, one deliberate kill and resume. The narrated replay is at rollupa.com.

Evals

The grader catches planted errors. The revenue estimator needs owned-firm data.

Seeded fixtures correct: 3 of 3Seeded fixtures correct3 of 3Injected errors caught: 5 of 5Injected errors caught5 of 5Backtest firms inside revenue range: 18 of 30Backtest firms inside revenue range18 of 30

Two of the three fixtures are trap deals (a key-person dependency and inflated debtors) and both were rejected. On the backtest the median error of the mid estimate is 34%, because headcount undershoots revenue. Every completion recalibrates it.

Roadmap

The deal team is the first agent team. Every function in the firms gets one.

LIVE TODAYDeal teamSourcing from Companies HouseTriageUnderwriting and revenue estimateDeal structuringGradingInvestment committeeNEXTIntegration teamOnboarding to the shared back officeInvoicingChasing unpaid invoicesCertificates and logbooksTHENOperations teamsInspection scheduling and bookingsCustomer supportRemedial quotesOutbound salesEngineer routing

Each team moves work off the owner-director and onto the shared back office. That is where the margin uplift in the model comes from.

Deal structure

Banks lend against cash flow. Venture equity sits underneath.

Senior debt from banks: £506k (64%)64%Venture equity: £163k (21%)21%Retention earn-out, paid from cash flow: £118k (15%)15%Senior debt from banks: £506kVenture equity: £163kRetention earn-out, paid from cash flow: £118kPrice £788k: 3.5x EBITDA of £225k

3 to 4x EBITDA. Aggressive on price, but every offer must still beat the owner's best real alternative on cash, certainty or legacy.

85% in cash at completion, the rest on customer retention over two years. Owners are paid out and lend nothing.

Fair and fast gets referred. Owners talk to each other and to trade bodies.

Unit economics

Each firm is worth about 3.6x its price inside the platform.

Price paid, 3.5x EBITDA: £788k£788kPrice paid, 3.5x EBITDASame EBITDA at a 7.5x platform multiple: £900k+£900kSame EBITDA at a 7.5xplatform multipleMargin 15% to 25% on the shared back office: £1.1m+£1.1mMargin 15% to 25% on theshared back officeWorth £2.8m inside the platform, 3.6x the price paid

One standard firm: £1.5m revenue at a 15% EBITDA margin. Assumptions to test, not facts.

Trajectory, base case

264 firms and £87.5m of EBITDA by year seven at 50 deals a year.

Firms ownedYear 1: 6Y1Year 2: 26Y2Year 3: 64Y364Year 4: 114Y4Year 5: 164Y5Year 6: 214Y6Year 7: 264Y7264Group EBITDA, run rateYear 1: £-410kY1Year 2: £5.4mY2Year 3: £17mY3£17mYear 4: £33.5mY4Year 5: £51.1mY5Year 6: £69.1mY6Year 7: £87.5mY7£87.5m

Owned firms' free cash flow funds most later deals, so total equity raised is £26.5m. Equity value at a 7.5x exit multiple: £591.4m.

Illustrative exit

Year seven: sell the group to a PE buyer for £656.3m.

Sale at 7.5x EBITDA: £656.3m£656.3mSale at 7.5x EBITDABank debt repaid: £91.1m−£91.1mBank debt repaidHoldco cash: £26.3m+£26.3mHoldco cashEquity value: £591.4m£591.4mEquity valueWho receives the equitySeed: 11%, £66.8mSeed: £66.8m11%, 22x on £3mLater rounds: 55%, £324.3mLater rounds: £324.3m55%, 14x on £23.5mFounders and team: 34%, £200.3mFounders and team: £200.3m34%

Firms bought one at a time at 3.5x EBITDA are sold together at 7.5x, a size PE can buy. Selling regional clusters earlier is the alternative route. Base case of the illustrative model.

Returns

A seed investor makes 9x to 46x, and 22x in the base case.

Exit multiple, EV / EBITDA6x7x7.5x8x9x25 deals a year25 deals a year, 6x exit: seed returns 9x9x25 deals a year, 7x exit: seed returns 11x11x25 deals a year, 7.5x exit: seed returns 12x12x25 deals a year, 8x exit: seed returns 13x13x25 deals a year, 9x exit: seed returns 14x14x50 deals a year50 deals a year, 6x exit: seed returns 17x17x50 deals a year, 7x exit: seed returns 21x21x50 deals a year, 7.5x exit: seed returns 22x22x50 deals a year, 8x exit: seed returns 24x24x50 deals a year, 9x exit: seed returns 27x27x75 deals a year75 deals a year, 6x exit: seed returns 24x24x75 deals a year, 7x exit: seed returns 29x29x75 deals a year, 7.5x exit: seed returns 31x31x75 deals a year, 8x exit: seed returns 33x33x75 deals a year, 9x exit: seed returns 38x38x100 deals a year100 deals a year, 6x exit: seed returns 29x29x100 deals a year, 7x exit: seed returns 35x35x100 deals a year, 7.5x exit: seed returns 38x38x100 deals a year, 8x exit: seed returns 41x41x100 deals a year, 9x exit: seed returns 46x46xOutlined: base case. Seed multiple after dilution from later rounds, year seven.

Deal velocity moves the return more than any other lever. The engine exists to raise it.

Standard firm £1.5m revenue at 15% margin, bought at 3.5x EBITDA with 2.25x senior debt at 8%. Margin +10 points over two years, 5% revenue lost after completion, £75k integration per deal, central cost £2m plus £20k per firm. Seed £3m at £12m post-money, later rounds priced at 6x EBITDA less net debt. Exit in year seven.

Sensitivities

Velocity and the back-office uplift drive the outcome.

base case 22xDeals a yearDeals a year: 12x to 38x12x38x25 a year100 a yearEntry priceEntry price: 12x to 31x12x31x4.5x EBITDA3.0x EBITDAMargin upliftMargin uplift: 6x to 22x6x22x+0 pts+10 ptsExit multipleExit multiple: 17x to 27x17x27x6x9xBank debtBank debt: 18x to 25x18x25x1.5x EBITDA2.5x EBITDA

With no margin uplift at all, seed still returns 6x on multiple arbitrage alone.

Paying 4.5x instead of 3.5x cuts the return from 22x to 12x. Price discipline is worth protecting.

Moat

Owning the firms is what a software rival cannot copy.

Operating dataTrue revenue at every completion recalibrates the estimator.
Seller reputationRetiring owners talk to each other and to trade bodies.
Route densityEngineers share routes across neighbouring acquired firms.
Falling unit costBack-office cost per firm drops as the platform grows.
Team

Taylan Durmus, founder and operator.

BringsAI and ML (builds the engine), aerospace (safety-critical, compliance-heavy work) and venture capital.
Hiring forA co-founder who has run or sold a field-service firm, and an industry president per vertical.
The ask

Venture equity for the engine and the first deals. Banks fund the rest.

Use of fundsThe agent engine and evals, a central team, the shared back office, and equity for the first acquisitions.
MilestonesFalling cost per closed deal, shorter time to completion, and estimate error shrinking on owned firms.

Every deal makes the next one cheaper and better priced.

Taylan Durmus, linkedin.com/in/tayland · rollupa.com

PDFSite