AI That Gets Used: Investor Relations at a Global Private Markets Firm

Private markets · investor relations & product teams · An illustrative engagement profile drawn from our work with private equity firms and their portfolio companies; details are representative and not attributable to a single client.

Situation

The firm is a global private markets investor, managing capital in the tens of billions on behalf of institutions, pension funds and private clients across Europe, the Americas and Asia. Like every firm in the asset class, it runs on documents: fund reports, due diligence questionnaires, market analyses, and a continuous flow of detailed questions from investors, each expecting a precise, consistent, and fast answer.

The operational reality behind that flow is less visible from the outside. Investor relations and product teams spend a substantial share of their week retrieving information the firm has already produced: locating how a similar question was answered eighteen months ago, in which letter, approved by whom. The knowledge exists; it is buried in past communications. Senior professionals hired for judgment were spending their hours on retrieval.

Leadership's framing of the problem was notably mature, and it shaped the engagement. The challenge, as they saw it, was never to find the best model on the market. It was threefold: identify the use cases actually worth building, build them inside strict confidentiality and European regulatory constraints, and, hardest of all, ensure the teams would fold the tools into their daily work. In their words, AI was a change-management problem before it was a technology problem.

We share that view. It is, in our experience, the difference between the minority of financial-sector AI programs that reach production and the majority that remain pilots.

Approach

The engagement followed our standard sequence: see the work as it is actually done, surface the use cases with the people who do it, then build narrow and secure.

Mapping the work. Structured interviews across levels and functions, from junior analysts to partners, across investor relations, product, and the data function, crossed with the systems record: where requests arrive, where answers live, where hours go. As always, we map friction, never individuals. The map confirmed the bottleneck leadership suspected and sized it: the single most repeated, most automatable, highest-volume task in the perimeter was matching incoming investor questions to the firm's own history of validated answers.

Co-creating the use cases. Rather than delivering a recommendation deck, we ran a working session that brought together staff from every level and the functions concerned, to shape candidate use cases against two filters: measurable value, and willingness of the team itself to adopt. This step looks soft; it is not. The people who will use a tool can tell you in an afternoon what a specification review will not reveal in a quarter, and involvement at this stage is what later shows up as adoption.

Building inside the firm's walls. The selected use case became a drafting assistant: a large language model connected, inside the firm's secured environment, to its historical base of investor communications. A new question arrives; the assistant retrieves how the firm has answered its closest precedents, and produces a sourced draft response for human review. Confidential data never leaves the firm's infrastructure. The design is deliberately model-agnostic, the firm can substitute or add models as the market evolves, because the value sits in the connection to its own validated history, not in any single model.

Results

The quantitative results are what a board would expect from a well-chosen use case: several thousand investor questions have now been answered through the assistant, with routine requests handled in a fraction of the previous time, and the team's senior hours redirected to the complex, judgment-heavy questions they were hired for.

The result we weight more heavily is behavioral: more than half of the target team uses the assistant daily, unprompted, months after launch. In enterprise AI, daily unforced usage is the scarcest metric there is, most deployed tools plateau at a fraction of that. Three factors produced it, none of them technical:

• The use case was surfaced from the team's own described friction, not imposed from a technology roadmap
• The future users shaped the tool before a line of it was built
• Leadership sponsored the program visibly and consistently, treating adoption, not deployment, as the definition of done

The firm has since extended the approach: the same secure foundation now supports broader research and due-diligence assistance, and the roadmap is set by the same principle that opened the engagement, use cases are earned by measured value and adopted use, not announced by ambition.

Observation

Private markets firms are unusually well positioned for AI, for an unglamorous reason: their most valuable asset is already written down. Decades of validated answers, analyses and decisions sit in their archives, a proprietary corpus no competitor can replicate. The firms that convert that archive into a working memory, with tools their teams actually use, compound an advantage every quarter. The firms that buy tools first and ask adoption questions later fund the statistics everyone cites.

The lesson of this engagement travels well beyond private markets: the technology was the easiest part, the use-case selection was the important part, and the adoption was the whole game.