Three layers of AI capability for investment banks: which one are you actually missing?

Investment banks are investing heavily in AI. Most of the attention, and most of the budget, is going to the same two layers.

The first is workflow automation: tools that handle the time-consuming, process-heavy work that junior teams have historically carried. CIM research, comparable analysis, initial screening, document preparation. Vendors like Rogo are building specifically for this layer. The value proposition is clear and the market is competitive.

The second is conversational AI: the ability for professionals at any level to ask questions and get intelligent responses. ChatGPT and Claude are the default tools here. Most senior bankers are already using them regularly. Adoption is high, the tools are accessible, and the use cases are intuitive.

Both of these layers are real and useful. They are also not where the durable competitive advantage is being built.

The layer most firms are missing is the first one. Not first in the list above, but first in order of importance: the intelligence warehouse.

Layer 1: The intelligence warehouse

This is the foundation that the other two layers depend on, and the one that receives the least attention in most AI investment conversations.

An intelligence warehouse is a structured, governed data layer that combines a firm's proprietary knowledge with live external market signals. It captures what the firm knows: deal history, buyer and seller behaviour across mandates, relationship context, sector thesis development, management assessments, ownership changes, company signals relevant to the firm's coverage universe.

Without this layer, the other two are operating on generic inputs. Workflow tools surface generic results. Conversational AI gives generic answers. The intelligence is public, not proprietary.

With this layer, everything changes. Workflow outputs are grounded in the firm's actual view of the market. Conversational AI answers questions based on what the firm genuinely knows, not what the internet broadly knows. The system compounds: every deal, every conversation, every signal enriches the knowledge base and improves future outputs.

This is the layer that does not depreciate. Models improve constantly and commoditise quickly. Proprietary data, structured and maintained over time, is an asset that only a particular firm can have.

Layer 2: The operator layer

Once the intelligence warehouse exists, the question becomes how people interact with it.

Senior bankers are not going to use a dashboard. They are not going to filter through a CRM interface to find what they need before a client call. They want to ask a question and get a reliable answer, in the same way they would ask a trusted colleague who happened to know everything the firm knows.

This is the operator layer: the interface that connects professionals to the intelligence warehouse in a way that is natural, fast, and reliable. The output should feel like a conversation, not a database query.

The important distinction here is that this is not the same as using ChatGPT. A general-purpose chat interface draws on public information. The operator layer draws on the firm's proprietary data. The interface looks similar. The underlying capability is fundamentally different.

For senior professionals and partners, this is the layer that changes how they prepare, how they advise, and how quickly they can develop conviction on a deal or a sector. The questions that previously required a junior team working across multiple sources for half a day become a two-minute conversation with the firm's own knowledge base.

Layer 3: Workflows and dashboards

The third layer is where most of the current market focus sits, and where the competition is most intense.

Workflow tools handle the process-heavy tasks: building buyer lists, screening targets, producing initial research packs, monitoring portfolio companies, tracking pipeline. These are real time savings and the value is tangible at the junior and mid-level of the team.

This layer is genuinely useful. It is also the most replicable and the most commoditised. As general-purpose AI models improve, an increasing proportion of workflow tasks become table stakes rather than differentiators. The tools that exist in this layer are competing on speed, accuracy, and ease of use. The gap between them will continue to narrow.

For firms making AI investment decisions, this is the layer where the return diminishes fastest as the market matures.

Why the order matters

The instinct to start with Layer 3 is understandable. The outputs are visible, the use cases are concrete, and it is easy to demonstrate value to a sceptical team. But firms that start here without building Layer 1 are optimising the symptom rather than the cause.

Workflow tools are only as good as the data they run on. If the data is fragmented, incomplete, or disconnected from the firm's actual knowledge, the outputs reflect that. Better process applied to poor information produces polished but limited results.

The firms building a genuine AI advantage in investment banking are starting with the data architecture and building out from there. Layer 1 makes Layer 2 reliable. Layer 2 makes Layer 3 meaningful. The sequence matters.

What this means in practice

For most investment banks and M&A advisory firms, the practical implication is a question about where your data actually is.

Not the data you subscribe to. The data you own. Your deal history, your relationship context, your sector thesis, your institutional memory. Is it structured? Is it maintained? Is it connected to the live market signals your coverage depends on? Can any member of your team query it reliably and get back an answer they can act on?

If the honest answer is no, then the AI tools you are investing in are running on a weaker foundation than they could be. The gap is not in the tools. It is in the layer underneath them.

Building that layer is not a quick project. But it is the one that compounds. And the firms that start now will have a materially stronger intelligence asset than the firms that address Layer 3 first and come back to Layer 1 later.

Deal Engine helps investment banks build the intelligence infrastructure that makes AI applications reliable and proprietary. To learn more, speak to our team for a personalized demo of the platform.