Services Firms Don't Need AI Yet. They Need a Data Spine.

Every services firm I talk to is getting the same pitch. An agent that drafts the deliverable. A copilot in the document editor. A chatbot that knows all your past work.

Almost none of them can run it.

Not because the tools are bad. Because the thing those tools require does not exist inside most services firms, and nobody selling them is going to raise it.

Read this knowing that I sell the work I am about to recommend. A post arguing you should fix your data before you buy AI is a post that sells what I do. So do not take my word for any of it. Every claim below is checkable from inside your own building this week.

The Demo Runs on a Company That Does Not Exist

Watch the demo again and look at the data instead of the output.

One schema. Clean client names. No duplicates. Every document tagged and in one place. No access restrictions. No engagement that ended badly and got half its files deleted.

Now your firm. Engagement records in the practice management system. Hours in a different one. Deliverables spread across a document store, three years of email, and a partner's laptop. The same client spelled four ways because four people set them up. A CRM last updated by somebody who left in March.

The demo is not dishonest. It is just running on a fictional company, and the gap between that company and yours is the entire project.

The Part Nobody Mentions

Here is the one that stops firms cold, usually after they have already bought something.

A services firm's most valuable data is its past engagements. That is also the data you have the least freedom to use.

Look at what your client agreements actually say. Many restrict what you can do with client data. Some prohibit using it to build or improve anything that serves other clients. Which means the obvious move, indexing all your past work so the model can learn from it, is not a technical project. It is a contract question, and the answer varies client by client.

I am not a lawyer and this is not legal advice. But you can find out fast. Pull your ten largest agreements and read the confidentiality and data-use clauses. Ask your counsel before you ask your engineer. If the answer comes back mixed, and it will, then permissions are not a feature you add later. They are the foundation.

The Billing Model Fights the Tool

Second thing nobody says out loud.

If you bill by the hour, a tool that makes a senior associate thirty percent faster reduces revenue on that engagement by thirty percent. The efficiency is real. It just lands on the wrong side of your income statement.

Firms hit this and conclude the AI did not work. The AI worked fine. The pricing model did not.

So there is a step ahead of the technology one, and almost everybody skips it. Decide what you are selling. Hours, deliverables, outcomes, or access. If the honest answer is still hours, then you are shopping for a tool that cuts your own revenue, and you should know that before you sign, not during the renewal conversation.

What a Data Spine Actually Is

Four things. None of them are exciting and all of them are load-bearing.

One canonical list of clients, engagements, and people. Not four systems with four spellings and a spreadsheet that reconciles them. One list that other systems point at.

Documents with provenance and permissions attached. Every deliverable carrying which client it belongs to, which engagement produced it, and what you are allowed to do with it. Without this the contract problem above has no solution, because you cannot enforce a rule you cannot express.

Hours joined to what they produced. Most firms can tell you hours booked by client. Far fewer can connect those hours to a deliverable and an outcome. That join is where the actual economics of your firm are hiding.

Retention and deletion with a name on it. What gets kept, for how long, and who decides. If nothing is ever deleted, your storage grows forever and so does your exposure.

That is the boring stuff. It is also the reason the firms that do it get value from AI in a quarter and the firms that skip it are still running pilots a year later.

What I'd Do This Week

Pick your largest client and count the systems their name lives in. Every place. That number is your real project scope, and it is usually higher than anyone guesses.

Read the data clauses in your ten biggest agreements. Confidentiality, data use, derivative works, aggregation. Ninety minutes with counsel will tell you what you are actually permitted to build.

Decide what you sell before you compress hours. Write it down. If it is still hours, say so plainly and price the tooling accordingly.

Name one owner for the client record. Not a committee. A person who decides how a client is spelled and where that lives.

Do not buy the agent yet. It will still be there in ninety days, it will be cheaper, and it will work when it arrives.

The Part That Matters

The AI market is selling services firms the last mile and skipping the road.

Your competitors buying agents this quarter are not getting ahead of you. Most of them are going to spend a year discovering the same four problems and calling it an AI failure.

Build the spine. The tools are the easy part, and they are getting easier every month.

The data has never once gotten easier on its own.