Skip to content

Engraved illustration of a rough sea. The picture is decorative; the site works without it.

Editorial analysisChapter 4 of 9lesson

How it works: sourcing, acquiring, integrating, building technology and operating

The operating loop as the companies describe it, step by step, with the specific practices the research has been able to document and the ones that remain undisclosed.

1. Sourcing and selecting targets

The companies describe looking for recurring, labour-intensive, measurable work in fragmented industries with ageing owners. Titan's founders say they examine ticket-level work when evaluating MSP targets 1Company reported. American Growth Insurance ran ten pilot agencies before its first completed acquisition 1Company reported — and the research is explicit that ten pilots are not ten acquisitions. Sequence's Deployment Strategists are described as designing operating systems and workflows inside partner businesses 1Company reported, which implies diligence on the work itself rather than only on financials.

What is not disclosed: acquisition multiples, deal structures, earn-outs and seller rollover. Titan's consolidated revenue, EBITDA, valuation and multiples were not established anywhere in the corpus No source linkedAnalysis.

2. Acquiring — and the words that matter

The research separates signed, announced, closed and pilot at every step. General Catalyst reported more than 15 signed agencies for Integral where six had been acquired earlier 1Company reported. Baldwin's go-private is announced with closing expected in Q1 2027 1Company reported. Long Lake's Amex GBT agreement had shareholder approval but no confirmed completion at the cutoff 1Company reported.

Structures vary: a minority investment (Sequence in BankSouth 1Company reported), a majority go-private with sponsor equity and assumed debt (Baldwin: about $4.6bn equity, about $3.1bn net debt 1Company reported 1Company reported), a product carve-out (Accrual and Puzzle 1Company reported), and a fully owned launch (Percepta 1Company reported).

3. Integrating and standardising

Two philosophies appear. Thrive's Current keeps acquired firms' brands and describes partner-firm autonomy; Multiplier describes "boutique firm autonomy" alongside a permanent structure 1Company reported. Titan's CTO described flexible integrations with some stack standardisation 1Company reported. Long Lake's CEO is attributed with roughly 80% shared infrastructure across verticals 1Company reported.

The failure mode is documented too. Thrive's "The Pioneers" recounts a tax-software initiative that stalled because scanning work never became part of the admin team's routine, and RRBB's practitioners describe early pilot inaccuracies before adoption improved 1Testimonial.

4. Building the technology

Every substantive company names a platform: Sequence's Atlas (agents, orchestration, data systems, developer tools, reusable primitives 1Company reported), Long Lake's Nexus (which keeps environment, context and UX while an external Agents API supplies the harness 1Company reported), Eudia's Enterprise Brain and MINDs 1Company reported, Titan's ticket agents with human approvals and logged actions 1Company reported, Percepta's Mosaic orchestration 12Company reported, and Thrive's tax agents built with OpenAI's Codex 1Company reported.

The common organisational pattern is the forward-deployed engineer: engineers embedded with operators whose work is fed back into the platform 1Company reported. Long Lake's CTO reports standing up new agents in hours 1Company reported; Long Lake also describes evaluations from work traces and internal model post-training 12Company reported.

What is positioning versus demonstrated capability: platform names and architecture diagrams are company descriptions. Demonstrated capability is the operating evidence in step 5, and it is thin.

5. Operating and measuring

The measured results in the corpus are participant reports on different denominators:

CompanyReported measurementDenominator
Thrive / Current7,000 returns in pilot; ~31% less preparation time; ~50% more throughput 1Company reported 1Company reported 1Company reportedPreparation effort per return
Dwelly>1,000 August tasks; ~3 hours to 15–20 minutes of human effort per task 1Company reported 1Company reportedHuman maintenance time
Sequence / BankSouthStages 9→3; ~52% fewer days application-to-close; loan volume doubled with the same team 1Company reported 1Company reported 1Company reportedElapsed loan turnaround
Titan / RFA85%+ acceptance of recommended ticket actions; 50% faster AI-assisted resolution 1Company reported 1Company reportedTicket handling
Shield>50% of tickets resolved autonomously; median <15 minutes vs 5.5-hour baseline 1Company reported 1Company reportedTicket resolution time

Combining these into one "average productivity gain" would be misleading, and none of them yet shows the bridge from saved time to cash flow. The economics chapter sets out what that bridge requires.

How it works: sourcing, acquiring, integrating, building technology and operating · AI Holding Company Research