Changing your practice management system? Choose your AI gatekeeper with the same care

Every supplier is adding an AI connector

The software your firm uses is being connected to AI. Document management, practice management, transaction and data room platforms are all adding connectors built on the Model Context Protocol (MCP), an open standard that lets any compatible AI assistant read from, and act in, another system.

That is good news. AI that can see where a matter stands, and update the task list afterwards, is far more useful than AI that only sees what someone pastes into it. A recent supplier whitepaper put it well: today’s AI works next to the firm’s systems, not within them.

The same guidance, sensibly, tells firms to ask each supplier whether it has an MCP connector, whether the connector respects user permissions, and whether it keeps an audit log. Those are the right questions. But asked one supplier at a time, they lead somewhere firms should think hard about.

Six systems, six sets of rules, no one picture

Follow the one-supplier-at-a-time approach across a typical firm and you end up with:

• Several AI connectors, each with its own permissions model. Each may be well built. None of them knows what the others allow.
• Several audit logs, held by several suppliers. If a client or the SRA asks what AI did with that client’s information, the answer is spread across systems you do not control.
• No firm-wide rules. A client’s instruction that AI must not be used on its work has to be set up separately in every system, if each system allows it at all.
• A quiet contest to become the hub. Some platforms with built-in AI can also reach into other suppliers’ systems. Whichever one becomes the place your people use AI ends up drawing on all of your data, under its terms and with its choice of AI company.

None of this is anyone’s fault. It is what happens when each system is connected separately and nobody owns the connection as a whole.

The rabbit is out of the hat

Your staff are going to use AI, and so are your clients. Most of it will be well meant: a fee earner who wants help with a long bundle at 9pm, a client who wants to understand their own contract before the meeting. Banning it will not work. People will simply use their own tools, on their own phones, with your clients’ information.

So the question is not whether AI is used, but whether it is governed. A firm needs to protect its paying clients from staff who just want the tools to help them, and sometimes protect clients from themselves:

• Give staff an approved route that is easier than the unapproved one. If the firm’s AI is quick, capable and connected to the matter, nobody needs to paste client documents into a free chatbot. Pair it with a clear policy, short training and, where needed, technical blocks on consumer AI services.
• Make the rules follow the client. Client instructions on AI, confidentiality levels and ethical walls should apply to every AI request automatically, whoever makes it and whichever system it reaches.
• Give clients something better than a chatbot. Client-facing tools that cite their sources and run under the firm’s rules, with clear engagement terms about how AI-prepared material is reviewed and priced.
• Keep one record. When something goes wrong, and eventually it will, the firm needs to know exactly what was asked, by whom, and what each AI tool saw.

All four depend on the same thing: one place where the firm controls how AI reaches its data. That is the gatekeeper.

The gatekeeper outlasts the system

Firms have seen this pattern before. Many chose a specialist document management system alongside their practice management system, and kept it through more than one change of practice management supplier. The documents were too important to tie to one platform.

The layer that controls AI access to your firm’s data is the same kind of decision. Call it a gateway, a controller or a gatekeeper: it is the single route through which AI reaches every system, under the firm’s rules, with one record of what happened. Choose it well and you can change practice management system, add a document tool or switch AI provider without rebuilding your controls each time.

It also answers the regulator’s questions directly. The SRA’s August 2026 warning notice on AI expects firms to be satisfied that client data stays in a secure environment, is not accessed by unauthorised third parties and is not used to train AI models without explicit authority. That is far easier to show from one gatekeeper than from several suppliers’ settings.

What a good gatekeeper gives you

• One set of rules, including rules by client or matter, to meet outside counsel guidelines that restrict AI.
• One record of who asked what, what each AI tool saw and what it did, in every system.
• Your choice of AI company, on your contract: UK or EU hosting, no training on your data, nothing retained.
• Read first, act later. Suppliers’ guidance rightly recommends giving AI read access first and allowing it to change things only once its results have been checked. A gatekeeper is where that is enforced, across every system at once.
• One switch to turn AI off if something goes wrong.

It does not mean rebuilding permissions. A well-designed gatekeeper calls each system as the signed-in user, so each system still decides what that person can see. It adds the firm-wide layer on top.

What to ask in your practice management selection

If you are choosing a new system, add these to your requirements. Our free requirements builder now includes them.

1. Does your MCP connector give AI tools our data (records we can search, read and, where we allow it, update), or only your AI’s answers?
2. Does AI sign in through our identity provider and act as the individual user, so matter permissions and ethical walls apply?
3. Is AI and MCP access included in our licence, with published rate limits and notice before anything changes?
4. Can we route all AI access through our own gateway, apply our own client-level rules, and export a complete log of AI access and actions?
5. Can we allow AI to read without allowing it to change anything, and require approval for any action it takes?

Download the checklist for COLPs and partners (Word)

“AI and your client data: questions before you choose a system”. Written for COLPs, managing partners and business owners: 26 plain-English questions to have answered in writing, plus what the firm needs in place itself.

A supplier that answers these clearly is building for a firm that stays in control. Vague answers, extra charges or “only with our AI” are worth noting before you sign, not after.

Decide the gatekeeper on purpose

Choosing a practice management system is a once-in-a-decade decision. So, increasingly, is the way AI reaches it. Make both on purpose, in the same selection, and make sure the system you choose fits the gatekeeper you want, not the other way round. (See also AI is an enabler. Why handcuff it?)

We help firms choose systems independently, write requirements that protect their position on AI and data, and plan a gatekeeper the firm owns, independently of any vendor.

Supplier guidance referred to: a supplier whitepaper on the Model Context Protocol for law firms, May 2026. We describe patterns, not suppliers. If you think anything here is out of date, please tell us and we will correct it.