Quick note on what's new. Just Curious: Applied AI Case Studies is the newsletter of Pluris, and we've spent the last few months heads-down on the AI Use Case Library, our open catalog of applied AI case studies, each one a first-hand account from the AI expert who built it, written by the team that built it so you can search by industry, function, AI pattern, and value type. It's now deep enough to mine, so the newsletter is changing shape to match.

Going forward, each issue will feature a handful of new case studies from the Library: two to four, written tight, with a one-click path to talk to the expert. The goal is simple: every Friday, we give you a couple of patterns you could borrow this month and a couple of people you could call this week. If you've been reading for a while, the deep single-expert Q&A format isn't going away; it'll show up alongside the case study issues when an interview is worth a full sit-down.

Ok, let’s get into it.

Chris Taylor's Fractional AI was acquired last Wednesday by the new Anthropic-, Blackstone-, and Hellman & Friedman-backed enterprise services venture, which makes the case study below a real-time look at the kind of build that team is going to be scaling. And Osman Ghandour of Soal Labs walks through what it took to encode six senior MDs as queryable personas inside a mid-market PE firm's IC process.

Key takeaways from this week's two cases:

My thoughts: I featured Chris on the podcast back in October. Last Wednesday, the new Anthropic-backed enterprise services venture announced his company as its operational centerpiece, so the case study / build below is also a preview of what that team is about to be doing at scale.

The setup. A PE-owned e-commerce business had outsourced a core back-office task to a third-party BPO: matching inbound shopping lists, arriving as images, PDFs, spreadsheets, and other unstructured formats, to its internal product taxonomy. The offshore team held all the institutional knowledge. Turnaround was 24+ hours per batch, and the business was hostage to a vendor it couldn't replace.

What Fractional built. A multi-model AI pipeline on AWS. Google Gemini handles text extraction across the messy input formats, OpenAI does the core taxonomy classification, and Claude handles select intermediate steps. A confidence-scoring layer routes low-certainty outputs to a human review queue, and a feedback loop writes corrections back to the taxonomy database so the system improves over time.

The part most teams skip. Fractional built the evaluation infrastructure before shipping the pipeline. That meant they could benchmark the AI against the BPO baseline objectively, in blind side-by-side comparisons, before go-live. Two things came out of that. First, they could prove the system worked on the metric that mattered (accuracy), which gave the client the confidence to decommission the BPO. Second, they discovered the BPO had been performing worse than anyone assumed. Invisible until somebody finally measured it.

Best fit for: Operators (PE-owned or otherwise) with a meaningful BPO or offshore cost center doing repetitive, high-volume document processing, who want to take a recurring cost line down and bring critical IP back in-house inside a fiscal year.

→ Want to talk to Chris? Reply to this email and I'll make the intro. Or request one directly on the case study page.

My thoughts: Osman is one of the applied AI voices I've come back to a few times; this is one of the approaches I find myself describing to other PE firms when they ask where to start.

The setup. A mid-market PE firm with 20+ years of deal history and a strong IC process wanted to push prep depth past what any single reviewer could realistically deliver. The work was rigorous, but it was bounded by one person's bandwidth, one person's recall of past deals, and one person's view of which angles mattered most.

What Soal built. An IC Insights Engine that does two things at once. First, it encodes six senior MDs as queryable AI personas, built through structured, recorded interviews that walk through past deals to surface how each MD actually reasons (not just what they say they look at). Second, it indexes the firm's 1,000+ opportunity history, IC decks, and deal metadata, so every new deal can be cross-referenced against the firm's collective memory. On any deal, the system generates firm-specific devil's advocate challenges, pulls comparable past opportunities, and benchmarks the deal's metrics against the firm's own track record, all before the deal walks into committee.

The methodology that's most replicable. Soal didn't try to encode all six MDs at once. They started with one partner. Two weeks of structured interviews, two months of iteration. The naive v0 persona surfaced useful angles but also exposed gaps, including heuristics the MD stated in interview that diverged from what they actually did when shown a real deal. Each round of testing fed back into both the interview methodology and how the personas were structured. Once it worked on one MD, they scaled to five more, deliberately chosen for sector coverage so every vertical had a senior voice from day one.

Best fit for: Investors with significant deal history (hundreds of opportunities with outcomes), multiple sectors or strategies where no single MD covers everything, and senior partners whose judgment is distinctive but currently lives only in their heads.

→ Want to talk to Osman? Reply to this email and I'll make the intro. Or request one directly on the case study page.

Latest podcast: Tom Scott (Wrike) on why most SaaS CEOs are sequencing AI wrong. Tom runs a $250M+ ARR work management platform owned by Symphony Technology Group. He's three years into PE ownership and inside what he calls the AI "rebuild moment." If you're a SaaS CEO trying to figure out the order of operations (product, GTM, internal ops), this is the one to listen to.

Background reading on the experts featured above: Chris Taylor's full Just Curious profile · Osman Ghandour's full Just Curious profile.

Working on something where one of these approaches might fit? Reply to this email and I'll either make the intro to the right expert directly, or open a Mini RFP brief and come back with 4-5 expert perspectives and pricing in a few days. Both are free, and I read every reply.

Stu

Just Curious: Applied AI Case Studies. The newsletter of Pluris.