Firm AI: AI that runs the business of the firm, not just the practice.

Thad Jampol

We showed a quote to a group of managing partners that said individual productivity has improved 10x with AI, but firm productivity has not increased 10x. The explicit point is that individual productivity is not equal to firm productivity. And before I had even hit the crescendo on that, the body language was so positive. They were all looking at each other and nodding vigorously.

Many of these leaders said, “Look, we have to give our associates and junior partners some help in an area where we haven’t had a lot of technology. So we’re going to go buy one of those tools. But it’s kind of just a tool. What we’re really excited about is AI helping us — the firm.”

Thad Jampol

We showed a slide that said there are going to be three major AI categories within a firm. Everyone is going to have general AI — you’ll either pick Copilot or you’ll go with Claude Cowork or OpenAI Enterprise. You’re probably running two or three right now, but you’ll consolidate eventually. Then there’s the practice of law and document-oriented AI — you’re probably piloting Harvey or Legora or maybe Spellbook.

But there’s a third category that we are calling Firm AI. We had Intapp in that third column, and the room really understood it. They asked a lot of questions. They appreciated the framing that says all AI is not the same. There are different subcategories within the industry, but you’re going to need Firm AI for leadership.

Thad Jampol

What Firm AI means is really building these industry agents and playbooks that are going to automate all of the different business-of-the-firm processes — and doing so with the institutional expertise that typically resides only in the most senior people. So, it’s not just faster. It’s better.

What firms racing to deploy AI tools are discovering is that faster individuals aren’t translating into better-run firms. The conflicts bottleneck is still there. Client intelligence is still siloed. Leadership still can’t see the full picture without assembling it by hand. Adding a model on top of that doesn’t change any of it.

Don Coleman

Right. The competitive edge in AI isn’t the model; it’s what the model runs on. And that foundation — the firm’s data, its governance policies, its institutional methods — is what our Intapp Celeste agentic platform is built to deliver.

Thad Jampol

Work backwards from a firm that says, “We need to make a major jump into AI.” Okay — the next challenge is: How do you infuse these models with the firm’s context? A lot of that context, institutional knowledge, and experience resides in documents. And in a lot of firms — particularly accounting — content is everywhere: shared drives, desktops, different SharePoint instances, Dropbox. It’s madness.

So they say, “We’re going to have to get control of this information before we’re comfortable sharing it with a model.” That means you need to be able to secure it, lock it down, protect MNPI, protect independence, protect confidential information. To do that, you need to put it into some kind of organized structure. That’s a lot of what Intapp Collaboration’s provisioning engine does: Based on the engagement type, it creates the right structure so you can put content into it. And by putting it into that structure, Intapp Walls can then apply the security policies correctly and protect the information.

And then the coup de grâce: Even if you have access to that information, there may still be sensitive, secretive content in there you really wouldn’t want to feed into a model. Because once it goes in, you lose control. Anyone can surface it, and sometimes it gets commingled with other information. So you strip that out. You redact it. You anonymize it. You feed the model sanitized versions that have all the good substance in it but not the secret information.

That’s a two- or three-step value proposition: Organize your information, protect it with confidentiality policies, and then build the foundation for your AI models in a safe, responsible way. And we provide that whole foundation for whatever AI solution you choose — whether that’s Copilot, Claude, or Celeste. 

Thad Jampol

AI agents inherit the same obligations as the people deploying them. Information barriers, independence rules, and confidentiality obligations don’t pause when an AI is running. Firms that skip compliance architecture aren’t moving faster. They’re accumulating liability with every agent they run.

Don Coleman

That’s what makes classic intake and conflict management so foundational. The whole point is to make sure you’re not in an embarrassing situation — or worse, a legally compromised one. It’s not about making the AI sound smart. It’s about ensuring the firm doesn’t operate blind to its own obligations.

Thad Jampol

The path forward isn’t more AI adoption. It’s AI built for the way professional firms actually operate; AI that serves as a business services workforce multiplier, capturing the methods, relationships, compliance posture, and comprehensive institutional knowledge of the firm. The firms that will pull ahead are building the data and governance foundation that makes AI compound over time — where every engagement entered, every relationship logged, and every conflict cleared makes the next decision faster and more precise.

Don Coleman

It’s critical to address the business side of running the firm — a necessary, requisite precursor to AI deployment. And the firms that understand that are the ones building something durable — not just deploying faster.