Is Claude better than ChatGPT for finance work?

Both are capable of similar work on similar tasks, and whichever is ahead this quarter usually won't be by next quarter. The more useful question isn't which model is better, it's whether either model can actually reach your deal data, your past decisions, and your firm's context. Without that connection, the choice of model barely changes the outcome.

Why isn't AI working at my firm?

Almost always because the data isn't structured or connected, not because the model is weak. Firms paste a document into a chatbot, get a decent summary, and then hit a wall, because that summary doesn't know about the last three deals in the same sector or how the firm adjusts for stock-based comp. The constraint is the infrastructure underneath, not the model on top.

What is a connected data layer or ontology in AI strategy?

A connected data layer means everything that has ever touched your firm, financials, customer summaries, budgets, forecasts, IC notes, is structured, linked, and accessible in one place. An ontology is the structured map of how all of that context relates to itself, so a question about one deal can draw on every relevant deal, adjustment, and past decision the firm has ever made.

What should a firm fix before investing more in AI models?

Fix the data layer first: get deal history, financials, and notes out of scattered folders, inboxes, and untrusted CRMs and into one structured, connected system. Once that exists, any capable model can use it well, and future model upgrades become a free improvement rather than something the firm has to chase.

Which AI Model Should My Firm Use? Why It Doesn't Matter

Q: Which AI model should my firm use?

It matters far less than most firms assume. Frontier models leapfrog each other every few months and settle back to rough parity within weeks, so having access to more than one and seeing which you prefer for a given task is reasonable. But that preference is not a strategy, and switching models will not fix a firm that isn't getting value from AI.

Q: What are the levels of AI maturity?

Level 1 is basic visibility, such as dropping a document into a chatbot and getting a summary back. Level 5 is autonomous decision-making across a connected data network, where agents plugged into a firm's operational infrastructure act on its behalf. Most firms sit at Level 1 today and mistake it for the frontier, when the real jump is to Level 2: using connected data to make operational decisions, which depends on infrastructure rather than model choice.

Every few months a new model release resets the conversation. Firms drop what they’re doing to test it, decide it’s finally the one that can do real finance work, and start planning a switch. None of that changes whether the firm gets value from AI, because the model was never the constraint.

Does the AI model you choose actually matter?

Less than most firms think. Frontier models converge in capability, leapfrogging each other every quarter and settling back to rough parity within weeks of the next release. Agonising over one model versus another is like debating which protein powder to buy: the ingredients are roughly the same, both get the job done, and the value was never in the powder. It’s in the workout. Most firms spend all their time comparing labels and none in the gym.

Having access to more than one model and noticing which one you prefer for a given task is sensible. That’s a preference. The strategy is what sits underneath it.

What are the levels of AI maturity?

AI maturity runs from basic visibility to autonomous action across a connected data network, and the gap between the bottom and the top has almost nothing to do with which model a firm subscribes to.

Level 1 is dropping a CIM into a chatbot and getting a summary back. Level 5 is agents plugged into your operational infrastructure, acting on your behalf across a fully connected data network. Most firms sit at Level 1, getting decent summaries and believing they’re at the frontier of what AI can do for them. They’re at the starting line.

The jump from Level 1 to Level 2, making operational decisions from your own data rather than just summarising a single document, has nothing to do with which model you use. It depends entirely on whether your data is structured, connected, and accessible enough for any model to work with it properly.

Why does a chatbot summary stop being useful after the first read?

Because the summary doesn’t connect to anything. It answers the question in front of it and then forgets everything the moment the conversation ends.

A firm has thousands of deals sitting in the CRM, hundreds of thousands of pages spread across Box or Dropbox, years of notes scattered through email and Word documents. Paste one document into a chatbot and you get a competent summary. But then what? That summary doesn’t know about the last three deals you did in the same sector. It doesn’t remember how your firm adjusts for stock-based comp. It has no idea what your IC tends to flag in businesses like this one. Every new chat starts from zero.

That’s the equivalent of hiring a brilliant analyst who forgets everything at the end of each conversation and has never met anyone else at the firm. Impressive in the room, useless as a colleague.

Where does the real edge in AI actually come from?

From connected context: every piece of information that has ever touched your firm, structured and linked so a question about one thing can draw on everything related to it.

Every set of financials linked to the company they belong to. Every customer summary, budget, and forecast connected to the deals they touched. Every data point connected to every other through an ontology, a structured map of how all your deal context relates. Get that in place and a single question can draw on all of it at once: compare this CIM to the last three deals you did in the same sector, check how this company’s customer concentration stacks up against the one you acquired last year, surface what your IC flagged in similar businesses before.

The model you’re using is a preference. What sits underneath it is a strategy.

None of that requires a specific model. It requires the structured data layer that any capable model can then work against.

Why do so many firms feel like AI “isn’t working” for them?

Because their data isn’t in a bad place, it’s in no place at all, and no amount of switching models fixes that.

One PE professional put it plainly: his firm had everything it needed, but scattered across Dropbox, email, a CRM nobody fully trusted, and Word documents nobody could search. Give that firm the best model on the market tomorrow and it wouldn’t change much, because the constraint was never the model. It was the infrastructure sitting underneath it. The model is the engine, but a lot of firms are debating engine specs while sitting in a car with no wheels.

What should a firm actually do instead of chasing the next model release?

Stop evaluating models and start structuring data. Get deal history, financials, customer detail, and IC notes out of scattered folders and inboxes and into one connected system where every piece of context links to every other piece.

That work is unglamorous compared with testing the newest release, and it’s also the only part of this that compounds. A model upgrade next quarter is free once the data underneath it is already structured and connected. Firms that skip this step and keep chasing model releases instead will find themselves exactly as under-served by the frontier model of two years from now as they are by today’s.

If you want a second opinion on where your own data actually stands, talk to us.

The AI Model Doesn't Matter. Your Data Layer Does.