Every major bank now has an AI strategy, but most deployments stall before they scale. This framework, drawn from conversations with banking MDs, COOs, and AI implementation leads, organizes the critical gaps into four evaluation pillars, a set of platform design criteria, and a competitive landscape. If your firm has committed seven figures to AI infrastructure and isn’t seeing returns, these are the questions worth asking.

The four pillars:

• Data Capture: Does intelligence flow to the banker automatically, or does the system wait to be fed?

• Data Integrity: Is the system grounded in your firm’s verified data, or generating answers that merely sound grounded?

• Business Context: Does the platform create a structured intelligence layer the firm owns, or is data still scattered across inboxes?

• Revenue Agency: Does the tool help win mandates, or format the pitch book for mandates someone else originated?

CONTEXT

The state of AI in banking

The pattern I hear described most often: a team demos an AI tool that summarizes earnings transcripts or formats pitch books in seconds. Leadership is impressed. A budget is approved. Six months later, a handful of analysts use it sporadically, and nobody can point to a mandate it helped win. No data-driven ways to understand ROI.

And this isn’t anecdotal. A 2025 McKinsey-IACPM study of 44 financial institutions found that more than two in five have slowed AI development because of disappointing outcomes. Only 7% have fully scaled AI use cases. The institutions seeing actual returns are the ones that redesigned workflows end-to-end rather than layering new tools onto existing processes.

A second pattern worth noting: firms often focus on which model a vendor uses as though the model itself is the product. But model capability is the one variable that's rapidly commoditizing. Within a year or two, every firm will have access to the same frontier reasoning at roughly the same per-token cost. The lasting advantage won't come from which model you call. It will come from the data infrastructure underneath it. You can swap to a better model the week it ships, but you cannot retroactively build a proprietary data layer.

PILLAR 1

Data Capture

Does intelligence flow to the banker automatically, or does the system wait to be fed?

Revenue producers rarely log their own interactions consistently. CRM implementations that depend on voluntary input tend to struggle with adoption for precisely this reason. When input is inconsistent, the system loses credibility, adoption collapses, and institutional knowledge disappears with the next departure.

• Is data ingestion fully passive, or does it require the banker to take an action?

• Does it capture the content of interactions, not just metadata (who met whom, when)?

• Does coverage extend across the firm?

PILLAR 2

Data Integrity

Is the system grounded in your firm's actual data, or generating answers that sound grounded?

In a profession where you stake your reputation on what you say in a room, a plausible-sounding wrong answer is a liability. Fluency is not accuracy. The more coherent an AI output sounds, the less likely a user is to verify it.

Outputs without source attribution shift the verification burden back onto the banker, which is the exact work the tool was supposed to eliminate. Accuracy in banking is binary: a 90% correct output doesn't save time, it creates a verification step.

• Can every output be traced to specific source records?

• Is retrieval deterministic and permission-aware?

• Is there an auditable lineage trail for outputs and workflow actions?

• Can it satisfy internal model risk and audit requirements?

PILLAR 3

Business Context

Does the platform create a structured intelligence layer ready for humans and systems to act on?

Growing a banking business relies on compounding relationships (which also means compounding data). This becomes more consequential as AI agents enter front-office workflows: an agent reasoning against raw communications in real time is constrained by context window limits, and inference quality degrades as data volume grows.

A pre-structured semantic layer encoding a firm's business definitions, entity relationships, and deal-stage conventions removes both constraints, making years of relationship history available to reason over at substantially higher accuracy, and eliminating the hallucination and incomplete results that come with layering AI on unprocessed data.

• Does the system create a structured semantic layer with standardized metrics, entity relationships, and qualitative context (sentiments, transaction signals, investment criteria)?

• Can a new banker or AI agent reconstruct a full client relationship narrative from what the system holds?

• Can business users configure the data schema or a dashboard reliably without filing an engineering ticket?

• Is the intelligence layer accessible to external agents and other tools through open protocols, or does it require the vendor’s own interface?

PILLAR 4

Revenue Agency

Does the tool directly help win deals, or merely saves time?

Many of the tools being developed are designed for junior execution: easier company profile building, faster comps, better formatting. Those are real gains. But the value chain has two layers: productivity improvement applied to slide formatting saves analyst hours; applied to origination intelligence, it wins mandates.

Senior bankers put it plainly: the key to winning mandates isn’t in the 100-page deck. It’s in the five minutes of dialogue that convince a CEO to hire your firm. That requires knowing the client’s history, their peers’ recent moves, and why now is the right moment. That is the gap most current tools don’t address.

• Does the tool proactively surface revenue-generating signals without being prompted? Does it flag relationships going cold or clients whose sentiment has shifted?

• Can it generate meeting preparation grounded in the firm’s own deal and interaction history?

• Is it designed for the person who wins the mandate, or only the person who executes afterward?

PLATFORM DESIGN CRITERIA

Four implementation-level questions that separate tools that work in a demo from tools that work in production:

LANDSCAPE

The Competitive Landscape

No existing category solves all four pillars.

Each category has a distinct strength:

• General-purpose AI tools are strong on reasoning and useful for analysis and drafting when context is provided. They have broad domain knowledge and web search grounding for public data, but no persistent connection to firm-specific intelligence and no proactive origination capability.

• Finance AI tools are purpose-built for financial document search, with source citation against filings as the core differentiator. They largely operate against public data rather than a firm's own relationship context, and carry the same hallucination risks as general-purpose AI in generative Q&A mode. For senior bankers, the use case largely stops at initial brainstorming.

• Legacy CRMs are the incumbent systems of record. Email and calendar integrations give them partial data capture, and they hold years of relationship history. Their gap is the intelligence layer: stored data without semantic structure, and no proactive signal surfacing.

The gap none of them close is the same: a structured intelligence layer built automatically from a firm’s proprietary data / communications, governed, queryable, and exposed as a platform for both humans and agents to build on.

WHAT WE BUILT

A disclosure: I’m co-founder of Debreu.

From my time in banking, I experienced these gaps directly: CRM systems that couldn’t hold adoption, institutional knowledge that walked out with the banker, and Outlook search as the industry’s default intelligence tool.

Debreu is an AI intelligence and orchestration layer for banks and financial institutions. It captures communications, meeting data and files automatically, extracts finance-native structured intelligence with source attribution, and exposes that intelligence through an intuitive interface, modular APIs, and open protocols for third-party agent connectivity.

The framework above is an attempt to sharpen the evaluation criteria at this pivotal moment for the industry. The right question matters. Firms that get this right probably won't be distinguished by how many tools they’ve deployed, but by how clearly they understand the problem.

The recent SaaS index plunge reminded me of Brave New World. It depicts a society engineered for comfort, stability, and manufactured belief. John, our protagonist, grows up in a remote tribe untouched by the modern system—he’s raised on disease, religion, and Shakespeare. Upon being brought to London, the clash destroys him. He spirals into self-punishment, violence, and suicide.

The typical takeaway is: humanity gets crushed by the post-modern world. But given everything happening in tech right now, I’d like to offer a different take: you don’t want to be John.

Our read of SaaSpocalypse

The premise: Foundation models enable AI agents that can automate human workflows. The effects:

• Build internally: Companies can build their own custom agents or use ones provided by the models. Agents can use SaaS products (reducing seats) or directly perform the function (weaning off vendors)

• Vendor compression: Point solutions and general offerings from CRM, data enrichment, and workflow giants become vulnerable to data moat erosion and price undercutting from LLMs and modern/open-source disruptors

• Traditional SaaS strategy is deprecated: Seat-based pricing, systems designed to be used by human workers, and value props rooted in functionality are all incompatible with agents that directly drive business outcomes (side note: this reminds me of the advertising shift from impressions and clicks -> intent -> conversion)

• Not all SaaS is equally exposed: Vertical solutions with proprietary/full-stack data loops (Toast, Procore) or deep regulatory moats (Veeva) are defensible because agents can’t easily replicate domain-specific workflows. Core infrastructure systems that support agents (Databricks, Snowflake) may even benefit.

Why this matters to you

The market has spoken on AI agents. Nonetheless, folks are taking sides and both sides are loud. Amid the chaos, it may be difficult to evaluate what an agentic economy means for your firm, how you can benefit from it, and how you can get there. I want to focus on financial services and present some thoughts on:

1. What agentic finance looks like

2. What role do companies like Debreu play in the new world?

3. Why LLMs won’t eat our lunch

1. The Future of Financial Services Front-Office

Here’s the work stack today:

The system of record exposes data to human workers, who use it to perform business functions. LLMs are now used to assist workflows, but do not yet directly drive outcomes end-to-end. Three underpinning premises not explicitly laid out in the diagram are:

• The data layer today is mostly raw, incomplete, and often low-quality (read about our view of the ideal state of financial services enterprise data)

• Context across different interfaces (email, files, CRM, LLMs, market data providers) is not coordinated. People have to stitch together the truth and are unable to fully trust any single source

• Running manual workflows on top of fragmented data means that human bandwidth, error margins, and communication overhead are limiting factors, even with the best talent

Now here’s our rendering of the future stack (inspired by OpenAI Frontier):

Notable changes:

• Interface above workflows: workflows are now executed by agents instead of humans, humans input goals and monitor/intervene

• Orchestration layer: This layer tells agents what data they can and should fetch for each task and defines agent behavior. It establishes precise security controls and defines a standard protocol for connecting to a rich (and growing) ecosystem of modern solutions.

  • Put another way, this is the infrastructure that enables fine-tuned agent performance and lets teams easily 1) build internal agents in a low/no-code manner and 2) test out new third-party solutions without incurring integration costs or increasing risk surface.

• Intelligence layer: This level centralizes and standardizes business context – ingests raw data from various channels, validates and normalizes it, and assigns it meaning and traceability. Structured data is key for agents (and humans) to assess and complete tasks without hallucinating or running into context limits.

  • The data schema should be easily configurable to your business and systems.

• System of record: Note that in this world, traditional systems of record (e.g., CRMs) are relegated to the same status as any other source of raw data. They remain useful as record-keeping infrastructure, but they are not the intelligence layer (case in point: aggressive branding efforts like “Agentforce” and “CRMx”).

2. Where Debreu Fits In

At Debreu, we are building the Intelligence and Orchestration platform for financial institutions, enabling customers to seamlessly adopt agentic systems without rebuilding their infrastructure. We also provide domain-specific workflows across CIB/S&T/PE for use out-of-the-box. A bit more color on these points:

• The Platform

  • This is Debreu’s core offering

  • We enrich, clean, and persist your data, and provide the infra for safely adding new product integrations while ensuring agent behavior does not suffer from inaccurate or incomplete data or context

  • Combining the Intelligence/Orchestration layers is essential

    • Enables you to both access and control all data points: primary, derived, or enriched

    • Clean, MCP-backed APIs for agents mean any data reads are targeted and intentional, which is a win-win—you control what others see, and AI tools get the exact context they need

  • Want to use Debreu’s workflow agents, other products, or build your own? Do it easily and compliantly

• Data Schema

  • At the foundation of Debreu’s platform is a domain-specific, fully-configurable data schema

  • All data (internal and external) gets mapped to a personalized spec. Your intelligence layer should understand properties specific to your needs (e.g. healthcare vs. tech, debt vs. equity) at a granular level (firm, business line, team)

  • Intelligence is built into the schema. Beyond primary fields (e.g., company/people/projects), the schema captures derived context (e.g., sentiments, signals, relationships between entities, and how they evolve over time) that’s traceable to the source.

• Workflows

  • Tailored and proactive workflows

    • We believe you need both. Tailored workflows that are not proactive rely on repeated human prompting. Proactive automations that don’t understand your context and goals just generate noise

  • Examples we’ve built out: outreach initiation and tracking, action item identification, target list generation, meeting prep, deal management

That leads us to the next question: “even if Debreu can do all this, why wouldn’t I wait for GPT or Claude to build it?”

3. Existential Crises

Foundation model companies (Anthropic, OpenAI, Google) are going “product-mode”. The B2B landscape looks like an open feast. Can AI finance startups survive? How should financial institutions determine whether vendors are worth starting contracts with? (More on this to come in a later post.)

Here are three reasons why we believe we’ll survive and thrive:

Infrastructure, not feature

Debreu is an infrastructure platform for all your workstreams and data. We turn your institutional knowledge into a self-compounding asset. AI is simply one of many tools for us to serve our users, not the core product.

Until OpenAI becomes Skynet, your modern stack will comprise many layers, not just a single AI app. Processes and agents depend on a reliable, structured substrate.

Domain Specificity

Even if LLMs cater to “finance”, they need opinionated, finance-native ontologies, entity relationships, and signal taxonomies built into the schema to effectively serve users and agents; these opinions should be aligned with your specific industry, business-line, and team. If a CFO emails me “we’re evaluating our options”, a system with last-mile domain expertise knows exactly what additional context to search for, how to interpret it, and can differentiate between live deal signal and routine commentary.

In financial services, intelligence lives in nuance, and the cost of a wrong interpretation isn’t just wasted time; it’s a lost mandate.

Model-Agnosticism

LLMs are in a tight and ever-fluctuating race. You shouldn’t need to constantly pick winners or bet your intelligence infrastructure on a single provider’s roadmap.

We’ve seen massive gaps in effectiveness at both the provider and model level on different aspects of tasks. That’s why we use a variety of models, and constantly refine our composition based on new releases, fine-tuning with our internal test data, and user feedback. Debreu handles model volatility. You focus on getting mandates and doing deals.

Back to John

Whether you’re a financial services company or an AI SaaS startup, we believe now is a watershed moment to think about how your work stack can adapt to the future (i.e. the next couple years).

John spent his last days whipping himself in an abandoned lighthouse. Putting romance aside, John’s downfall wasn’t that the new world was inherently bad; it’s that he had no framework for operating in it. While this isn’t the greatest analogy, you certainly don’t want to wake up in a year and realize your wallet share has been eaten away by competitors using agents that have delivered faster, charged less, and expanded ruthlessly.

The future we’ve painted may seem distant (or is it?). Decisions today, however, matter because agents will bring about an exponential step in cost-efficiency and productivity (and humans are bad at exponential forecasting). Every contract refresh, every new vendor evaluation, every internal project today plays a role in whether your firm will benefit from compounding efficiency drivers or turn to stone.

On the flip side, this is also a great opportunity. There are so many AI solutions that companies can use to configure their “ideal” agentic work stack. The decision tree is massive and complex. My co-founder Lucy, who has firsthand experience evaluating AI vendors for a bank, is releasing a framework for financial institutions to set themselves up for success in the agentic new world—stay tuned.

I was recently chatting with a friend and fellow founder who re-interpreted Maslow’s hierarchy of needs for enterprise data (fun fact: Maslow didn’t come up with the pyramid). I wanted to share some thoughts on how this applies to financial services.

The key assumptions:

1. Each level is only achievable by completing the levels below

2. Levels progressively yield stronger outcomes and are worth pursuing

3. The top level is aligned with top-level personal/firm-wide goals

Why we’re talking about pyramids

We started Debreu with a clear goal: Help financial institutions turn their own data into a durable growth advantage.

To do this, we quickly realized institutional knowledge is key and we needed to deeply understand users’ data. Three axes we’ve used to segment data are: type, source, and quality:

• Type: core CRM data models (companies, people, interactions, files, notes, …), interpretation/intelligence layer (e.g. themes, sentiments, signals, analyses), orchestration/compliance layer (deal/pipeline management, reporting, auditing)

• Source: internal sources (emails, meetings, files), external databases, public data, and LLM-enrichments (for interpretations and analyses)

• Quality: correctness, completeness, usability.

Every datapoint you see, use, and store today falls onto different places along these axes, and their holistic effect is key to your and your firm’s bottom-line success (deals). Financial institutions use a combination of internal and external tools to capture data, but results are incomplete, unharmonious, and expensive. The adapted Maslow’s hierarchy provides a clean lens through which we can identify gaps in origination efficiency and evaluate data-driven solutions.

Why this matters for you

AI is here and it’s not stopping–we can all agree on that. The question is: how can financial institutions use technology to generate more returns? The raw answer: your data is your IP. Useful data and effective access to it puts you in a better position to initiate conversations, win mandates, and foster relationships, while also reducing human error and operational overhead.

If you’re reading this, you’re likely already evaluating solutions: gen-AI models, tools built on top of them, and in-house products. The challenge is: will these tools actually reduce friction, yield concrete results, and support long-term business goals? We believe that the most important factor is how any solution sources, augments, and leverages data.

Back to the pyramid

Our version of the data hierarchy is as follows:

Each layer compounds. Outcomes of workflow solutions are constrained by the quality of core data. Hooking up different products at each layer is costly and burdensome; they end up being glued together by human memory, effort, and time. As firms look toward the future and AI continues to redefine what is possible, we hope this is a useful framework for evaluating origination solutions.

A shameless plug to finish

We’ve spoken to dozens of advisory firms; they’ve told us that their setup for levels 1 and 2 is fragmented, incomplete, and cumbersome, resulting in low-adoption. For levels 3 and 4 (insights and action drivers), tech hasn’t provided any systematic approaches to origination at all! Most are focused on junior efficiency or generic chatbots (more to come on this point in a separate article).

Debreu is the full-stack solution to drive growth for financial institutions. Please reach out to info@debreu.ai if you want to try our product, learn more, or have any comments.