How a sourcing pipeline we built in 8 weeks found a ~$22M deal for a private equity firm.

Their capital was ready. Their pipeline kept showing the same names.

The firm bought control positions in franchisable, multi-location businesses. Their partners knew their niche and had a seven-person analyst team to evaluate opportunities.

After two exits, they had capital to deploy. But their usual relationship channels weren't producing the flow they needed. The pipeline sheet carried familiar names from one month to the next, while new entries often came through banker-led processes they expected to lose on price.

Meanwhile, analysts spent weeks collecting data-room material, checking management and commercial information, and drafting investment memos. That was the visible grind. And so, when they came to us, their request was to make diligence faster with AI.

Their deal log showed what faster diligence couldn't solve.

Before speeding up diligence, we spent a week studying their existing deal records. If analyst capacity was holding back investment, attractive deals should have been waiting for busy analysts, with opportunities lost because the team couldn't finish in time.

But their records showed a different pattern. Analysts had spare capacity. The firm was passing on the businesses it saw, and losing on timing rarely appeared as the reason. There weren't enough suitable targets entering the pipeline.

We laid out the two patterns side by side. Their partners initially pointed to a dry market and the relationship-driven nature of deal flow. But bidding faster wouldn't make a poor fit worth buying.

And so our recommendation kept the diligence work they wanted, but added what had to come first: a way to find businesses their existing channels weren't surfacing.

We built the search around their partners' judgment.

Finding more names wouldn't help if the partners still passed on them. So we interviewed their partners, studied their successful investments, and gathered feedback from founders whose companies they'd acquired. That gave us a practical definition of the businesses this firm wanted to examine.

We built a sourcing engine that searched the market for companies matching that profile. Its job was to rank relevant candidates and show the evidence behind the match.

Finding a target was the first step. For targets entering diligence, our co-pilot assembled the available data-room evidence and commercial research into a draft investment-committee memo. Claims linked back to sources. When the evidence didn't support an answer, the system could decline to provide one and send the question for review.

Their analysts and partners kept ownership of valuation, financial modeling, and investment decisions. We reduced the work of gathering and assembling evidence so they could spend more of their attention checking it.

The engine had to find investments they had already made.

A plausible-looking target list wasn't enough to justify changing their process, so we tested the sourcing model against their own history. We chose a cutoff, trained on earlier investments, and ranked ~5,000 businesses using information available at that time.

Six of the eight businesses they later acquired appeared in the top 200. All eight ranked in the top quarter. We also compared businesses they'd bought with ones they'd passed on. The model ranked the acquired company higher in 82% of those comparisons.

That showed the sourcing model could reproduce their investment preferences. But a good match didn't establish that a diligence memo was reliable, so we separately tested the co-pilot's retrieval, evidence support, red flags, and draft memos before the team relied on it.

A correct-looking name produced the wrong red flag.

One memo surfaced an adverse item about a target's executive. A partner caught the mistake: the item concerned a different person with the same name.

That was our matching failure, and it put trust in the whole memo at risk. So we tightened the identity checks that week and required a minimum match-confidence threshold for person-level claims. We also added mandatory human sign-off before a management red flag could enter a memo.

A source link wasn't enough. We had to establish that the source concerned the right person.

A business they had never seen became a closed investment.

The pipeline went into production in eight weeks. To make it part of real decisions, we appointed a champion to run new targets through it and bring the output to the weekly pipeline meeting. The team adopted it as the normal starting point for an evaluation.

One market scan surfaced a business in its top twenty that no one on their deal team recognized. It fit their niche and resembled businesses they already owned, but it hadn't crossed their desks through the usual channels. Our engine brought it into the pipeline, and the co-pilot supported the diligence that followed. The firm closed the deal at ~$22M.

By assembling the research and draft memo, the diligence workflow let each analyst complete 4× as many evaluations in the same period. That same productivity gain meant total diligence throughput more than doubled even as the team moved from seven analysts to four.

The added capacity came first; the managing partner then chose to reduce the deal team. The firm reported that the three analysts moved into similar roles at its portfolio companies. Together with other operating and overhead reductions, that decision removed ~$665K in costs.

We taught the team to check the claim behind the memo.

The same-name mistake made the handover requirement concrete: a convincing memo still needed scrutiny. Our training therefore centered on verifying claims against their sources, including answers that sounded convincing. The source links and escalation rules made that review part of the workflow.

The partners helped define the system's investment preferences from the start. Their team learned to use its output in real pipeline reviews, with partner judgment remaining the final step.

How we built and measured it

Read the technical details for the method, evaluation, and handover.

We gave faster diligence something worth working on.

Their original request was valid: diligence needed to improve. But our diagnosis showed why solving only that problem would leave their capital waiting. And so we built the sourcing engine that found new targets and the diligence workflow that helped their team evaluate them. The engine surfaced a business their usual channels hadn't brought them, and their team evaluated it and closed the ~$22M investment. The firm could put its available capital into a new opportunity instead of only processing the same pipeline faster.