Why senior bankers get generic answers from AI (and how to fix it)

Most senior investment bankers are using AI. Not experimentally, not reluctantly. They are using it to draft, to reason through complex situations, to prepare for client conversations, and to pressure-test their own thinking. The tools are genuinely useful.

But there is a problem that rarely gets named directly, and it is costing firms more than they realise.

The AI they are using does not know what they know.

The public information problem

When a partner types a question into ChatGPT or Claude, the model draws on everything it was trained on: news, filings, research, general market commentary, sector data available on the open web. It produces a credible, well-structured answer.

What it cannot do is draw on the eighteen months of buyer conversations your coverage team has had with a particular acquirer. It does not know that the strategic rationale for a deal in your sector shifted after a key portfolio exit. It has no access to the proprietary thesis your team has been refining across three consecutive mandates. It cannot tell you what your firm has learned, only what the internet broadly knows.

The result is that a significant portion of the AI-assisted work happening in investment banks right now is producing generic outputs. They look polished. They are well-reasoned. But they are not grounded in the institutional knowledge that makes an advisory firm genuinely differentiated.

That knowledge exists. It is just not connected to the tools people are actually using.

Where the knowledge is

Investment banks and M&A advisory firms generate substantial proprietary intelligence over time. Prior deal outcomes, buyer and seller behaviour across mandates, relationship history, sector thesis development, management assessments, the reasons a particular acquirer passed and under what conditions they might re-engage.

All of this is valuable. Most of it is inaccessible in any structured sense.

It lives in inboxes. In old CRM notes written in formats nobody standardised. In the memory of partners who have moved on. In Excel files maintained by one analyst who has since changed team. Firms know this is a problem. What they are less clear on is that it is precisely this problem that limits how much value they can extract from AI.

General-purpose AI is a powerful reasoning engine. But reasoning on incomplete or fragmented information produces incomplete answers. The constraint is not the model. It is the data the model has access to.

What changes when AI is connected to your data

The firms that are beginning to get materially better outputs from AI are the ones that have connected their models to a structured, proprietary data layer.

In practice, this means a partner can ask a natural language question and receive an answer grounded in the firm's actual knowledge, not just public market data. Questions like: who are the most credible buyers for this asset given our previous conversations with strategic acquirers in this sector? Which targets in our coverage universe have shown the signals that typically precede a fundraise? What does our deal history suggest about valuation expectations in this category?

These are not questions a general-purpose AI can answer reliably. They require the kind of context that only exists inside the firm. But when that context is structured, maintained, and made accessible, they become questions that can be answered in seconds.

The workflow shift is significant. Preparation that previously took hours becomes a conversation. Analysis that previously required a junior team working across multiple databases becomes a query. The partner's time is spent on judgement and client relationship, which is where it should be.

The reliability gap

There is another dimension to this beyond capability. It is reliability.

Senior bankers are appropriately cautious about AI outputs. The concern about hallucination is real, and the stakes of acting on incorrect information in an M&A context are significant. This caution is one reason AI adoption among senior professionals has been more measured than headlines suggest.

When AI is grounded in a firm's own structured data, the reliability profile changes. The model is not speculating from public sources. It is reasoning from information the firm has verified, structured, and maintained. The scope for confabulation narrows. The outputs become something a partner can actually build a client conversation around.

This is the shift that matters for senior professionals: from AI as a drafting tool to AI as a reliable intelligence layer. The first is useful. The second is genuinely transformative.

What this requires

Connecting AI to proprietary knowledge is not a matter of prompting differently or choosing a better model. It requires building the underlying data infrastructure that makes the connection possible.

That means structuring the firm's knowledge in a way that is queryable, maintained, and connected to live market signals. It means establishing how new intelligence enters the system and how it compounds over time. It means thinking about data architecture as a strategic asset, not an IT project.

Firms that do this well will have a materially different capability from those that do not. The underlying AI models are largely the same across the industry. What is not the same is the proprietary data those models are connected to. That is the differentiator that is hard to replicate and that compounds with every mandate completed.

The generic answer problem is solvable. The solution is not a better AI tool. It is better data infrastructure underneath the tools you are already using.

Deal Engine helps investment banks build proprietary data infrastructure that connects internal knowledge with live market signals, giving your AI tools something proprietary to work with. To learn more, speak to our team.