Proof of Work is the newsletter of Pluris, a network of pre-vetted applied AI experts and AI-native service organizations that specialize in helping businesses design and implement AI solutions that drive value.

Pluris is free for our clients. Two ways to use us. Browse the Pluris AI Use Case Library, hundreds of real AI implementations from our network: what got built, what it cost, and what changed once it shipped, organized around the problem you're solving. Or submit an AI need through an Expert Review: we'll anonymize it, match it to experts with relevant experience, and return 3–4 perspectives with pricing.

Most firms that buy ChatGPT Enterprise never get past email rewrites. A $10 billion PE firm had bought it, watched an earlier CIM analysis project fail, and had a leadership team that couldn't agree on what to point it at next. Eliza, one of only two boutique AI-native firms at OpenAI's Advanced Partner tier, came in, rebuilt the case for the tool from the workflow up, and eight weeks later 92% of the firm was using it weekly and CIM write-ups that took four hours were coming back in under two minutes. "The biggest shift," Brian says, "was helping leadership stop treating ChatGPT like a magic box."

Eliza has a second case in the library, a smaller one: a PE-backed operator with thousands of receipts a week going into the ERP by hand, and batches that lagged for days. Under four weeks later it was running same-day.

Both are below. I sent Eliza five questions about how they did it, so a lot of what follows is in their words.

At the bottom, three things I wrote this month: what you should own, what you should borrow, and who pays for it.

Key takeaways from this week's cases:

If your firm has ChatGPT Enterprise, a previous AI project that didn't land, and a leadership team that can't agree on what to do next, this is the case for you.

A $10 billion PE firm, 150 people, was exactly there. The COO, head of AI, head of due diligence, CFO, and deal partners were not aligned on AI priorities and were weighing competing vertical tools with no shared framework for choosing between them. An earlier CIM analysis project had failed and taken some trust with it. Usage across the firm stayed shallow, mostly basic email rewrites, nowhere near the deal workflow where the hours actually are.

Getting leadership to try again started with resetting the expectation. "A lot of companies assume they can buy the software, run a couple of training sessions, and employees will naturally figure out how to get value from it. But ChatGPT is still enterprise software, and enterprise software needs an implementation cycle." That means connecting the right data sources, building the most valuable use cases into the tool, and making those use cases easy to run. Push that work onto the employee and adoption falls off.

The build ran in three phases. Assess, weeks one to three: structured leadership discovery and ROI-based prioritization of use cases. Build and activate, weeks three to eight: GPT development against a real evaluation framework, including human evals on actual CIMs and LLM-based regression testing. Then scale, expanding usage and consolidating onto the platform. First value landed in six weeks.

The flagship is a CIM Write Up GPT that turns a 100-plus-page confidential information memorandum into an investment-committee-ready draft in under two minutes, against a manual process that ate four hours or more. It shipped with a Board Deck Summary GPT, an NDA Review GPT, a Financial Summary GPT, and a DISC Profile GPT as an early proof point, plus a DealCloud connector and compliance workflows through the API. Development ran on ChatGPT Enterprise with Outlook and SharePoint connectors, so the firm's own documents were in scope from the start.

Then the adoption number: 92% weekly active across all 150 people. Brian credits three things. First, they connected the company's data sources, because without that context the tool only gives generic answers. Second, they built the highest-value use cases so they worked out of the box, rather than expecting every employee to start from a blank prompt. Third, they went past basic training and taught people to build and extend the use cases themselves, so the firm could keep developing the tool after Eliza left. Brian's summary: "Most rollouts focus on licenses and training. We focused on implementation, usability, and internal capability."

Accuracy is the other half of it. The CIM write-up runs at roughly 95%, and getting there meant building something that fit how the analysts already worked. Eliza embedded alongside them, learned the workflow end to end, and kept them in the build so the system reflected their process instead of an outsider's assumptions about it. They also built evaluations around the workflow and designed the system to recognize its own uncertainty. If it doesn't have enough information or confidence, it says so instead of inventing something. In Brian's words: "Trust came from building it with the analysts, testing it against their standards, and giving it permission to say no."

The playbook: run it as an implementation cycle. Four steps, portable to any firm sitting on an underused enterprise seat.

An idea. Pull your ChatGPT usage report and look at what people are actually doing with it. If the top use case is rewriting emails, the implementation never happened. Eliza fixed that here in about eight weeks.

Best fit for: Private equity and investment firms that have ChatGPT Enterprise but want to turn it into core deal workflows, starting with CIM analysis and diligence, backed by real evaluation and change management.

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

A PE-backed company was processing thousands of receipts a week from field operations and vendors, with a ten-person offshore team keying the required fields into the ERP. Batches lagged by days, which delayed close and cash visibility. Manual entry put errors into the system that someone had to reconcile later, and the cost line scaled with volume, so growing meant hiring.

Eliza built a vision-and-language pipeline: OCR and layout parsing feeding LLM extraction, with confidence-based QA that routes low-confidence receipts to the existing human team and lets routine ones process straight through. Schema-aware validation enforces the business rules for dates, taxes, GL mappings, currency, and line-item reconciliation. PII is redacted, data is encrypted in transit and at rest with full audit logging, and the whole thing drops into the existing stack through SFTP or a watch folder plus an API connector to the database. It shipped in under four weeks and first value came in three.

Speed came from doing the unglamorous part first. Eliza mapped the workflow end to end, found the real bottlenecks, and designed the future-state process around what AI made possible, rather than pointing a model at the process as it stood. That, in Brian's telling, is where most of these projects stall: "These projects usually get stuck when companies bolt AI onto the way the business already works. That may improve one step, but it preserves all the friction around it."

Best fit for: Finance and operations teams processing high volumes of receipts or invoices manually, especially PE-backed companies looking for fast cost takeout and a cleaner, faster close.

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

What are people looking to do with AI? These are the most-read use cases in our AI Use Case Library this week. Real problems, real numbers, real companies.

If one of these is your problem too, the library is free, and every case has a button that goes straight to the person who built it. Browse the full AI Use Case Library →

Three things I wrote on LinkedIn this month, all circling the same question: what you should own, what you should borrow, and who pays for it.

💬 You don't need to invent an AI capability across your portfolio. You need to borrow one that already exists. Eliza is one of the firms whose delivered work sits in the library. Send me your portfolio page and you get back, same day, the plays worth doing for every company you own, sized, with the firm that already built that exact thing. → Read it

💬 Renting a platform made sense when building one was out of reach. Hundreds of thousands of people rented their telephone from Bell for twenty years after they were allowed to buy one, and most never knew they had a choice. Why enterprise data sits in the same spot now, and what that trade looks like inside a five-year hold. → Read it

💬 Pluris is free for our clients, and somebody is still carrying the cost. How the network actually runs, why I cut briefs from twenty firms down to three or four after burning people's time, and why we put our experts' names out front instead of behind a wall. → Read it

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

— Stu

Proof of Work: Applied AI Case Studies. The newsletter of Pluris.