Three questions investment banks should ask about their data

Investment banking teams are experimenting with AI in growing numbers. The use cases are well understood: faster research, better meeting preparation, more efficient deal documentation. But a pattern keeps emerging. Tools that perform well in a demo produce unreliable output in practice, and the reason is almost always the same. The data underneath them was not ready.

This is not a new problem. Private equity firms worked through exactly this challenge over the past several years, often after investing in AI tools that underdelivered. The ones that made meaningful progress did so by addressing the data foundation first: understanding what proprietary context they actually held, whether it was structured and current, and how it connected to the workflows where it needed to be useful. The tools followed. The data came first.

For investment banking teams asking the same questions now, three diagnostics are worth working through before drawing any conclusions about what AI can or cannot do for your firm.

1. Is your CRM a record of the past, or a picture of the present?

Most CRM data in investment banking reflects what happened, not what is happening now. Contact records were accurate when they were created. Relationship notes capture meetings that occurred. Deal history logs transactions that closed.

But the market does not stay still between mandates. The PE firm that passed on a sector two years ago may have since raised a new fund with a different thesis. The family-owned business on your coverage list may have appointed a new CFO who has quietly signaled a desire to exit. The strategic acquirer who was working through post-merger integration may now be ready to move again.

None of that shows up in a CRM unless someone has manually updated it. The diagnostic question is not whether your CRM is well-organized. It is whether the data in it is live. If the answer is no, then every buyer list, every outreach decision, and every pitch built on that data is starting from an incomplete picture.

2. How much of your firm's institutional knowledge is actually findable?

Every investment bank accumulates knowledge over time. Which buyers are active in which sectors. How different sponsors think about valuation. What worked in a previous mandate and what did not. The question is whether that knowledge is accessible, or whether it exists only in the memory of the people who were in the room.

In most firms, it is the latter. Past mandates contain information that would be genuinely useful for future ones: comparable transactions, buyer behavior, valuation dynamics. That information exists somewhere in your document repositories. The question is whether anyone can find it quickly enough to be useful, or whether every new mandate starts from scratch by necessity rather than choice.

When Cavendish tested an agentic approach to buyer list construction, the result was not just faster. It surfaced buyers the original process had missed entirely, drawing on deal history and market data that existed within the firm but had never been systematically connected. A list that had taken an associate three weeks to build was reconstructed in four hours, with more than double the number of credible buyers identified.

It made us realize we had actually missed out on a whole host of opportunities, or of identifying the right investor, or the right trade acquirer for our client. I would not go back to the old way of doing things.

John Farrugia, CEO, Cavendish

3. Are your AI tools working on reliable data, or producing confident-sounding noise?

Most investment banking teams are already experimenting with AI. The use cases are obvious and the tools are increasingly capable. The risk that tends to get underweighted is what happens when those tools are applied to data that was not built to support them.

The concern is not that a model generates obvious nonsense. It is the more consequential problem of AI producing plausible, authoritative-sounding output built on stale or fragmented inputs. A hallucinated buyer on a pitch list. A misread fund status. A contact shown as active at a firm they left eighteen months ago. These are not edge cases. They are what happens when the data infrastructure has not kept pace with the tools running on top of it.

The diagnostic question is not whether your team is using AI. It is whether the data those tools are working from is structured, current, and connected enough that the output is trustworthy. If it is not, the risk is not that the tools are unhelpful. It is that they are confidently wrong at exactly the moment it matters most.

None of these questions are primarily about technology. They are about data architecture: the foundational work of understanding what information your firm actually holds, whether it is current, and how it connects to the workflows where it needs to be useful. Getting that right is what determines whether AI becomes a genuine advantage or just another source of noise to manage.

Download: Agentic AI for investment banking