In my last blog, I talked about a partner using an AI summary to support a client meeting. The client queried some of the information, including a citation of a regulation that they knew had been withdrawn. But the AI assistant had produced a summary based on old drafts sitting in the same folder. It wasn't AI's fault - it was the information it was given.
My idea was about getting the fueling for the AI rocket right before its launched. And I now want to talk about what happens when you do that. Once your content is connected by business context using metadata- what it is, which client it relates to, who should have access then the big question changes. It moves from "can we trust these answers?" to "what work can be handed over?"
Before I move on, I want to be clear on "handing over" work. These agents that I want to talk about don't sign contracts, approve content or move a contract to next stage. A person still decides and is still accountable. It's about helping arrive at decisions faster.
"We've done this work before. Who did it? And where is it?"
Firms aren't typically short of expertise, but they may be short of ways to access it. A senior manager knows that the firm did similar work for a client in a similar industry 2 years ago. But a new proposal will only benefit from this knowledge if they ask her about it before the deadline.
When documents are connected to people, clients and engagements, instead of buried in folders, a knowledge agent can surface comparable work when a new proposal or project is created. It won't just be a list of file names but the actual relevant content with the chance to ask follow up questions from the document.
When your firm captures context on who did what and when, it becomes possible for an agent to support task allocation. It can suggest a shortlist of people based on the work they have done and their areas of expertise. A senior manager still picks the team, but their picks are backed by evidence.
Firms typically talk about "our people being our best asset" in marketing material and pitch decks - this is a chance to turn that idea into part of the client work delivery process.
"Everything starts from a blank page"
New engagements set ups might be a case of "rinse and repeat". Copy last quarter's plan, strip out what doesn't apply and then see who is free.
What about if it was possible to give an agent an engagement letter with a single instruction to build a plan with all the tasks in sequence? Over time, the agent develops into something that can create a plan including phases, approval gates with owners, due dates and an estimated finish date.
This idea works in another situation. On a production line, an intake agent walks a technician through logging a leak test failure conversationally, rather than having them find the right form to fill in line by line. If they type "leak test failed", and nothing more, nobody analyzing failure trends six months later gets anything out of it.
"Reviews are the bottleneck and reviewers don't agree"
This is another case for managing your content with business context.
A review agent could work through a contract clause by clause agents against your playbook, inside Word. Every finding points to the specific playbook clause where the draft falls short. It rates the impact, with an explanation, and offers a fix as a tracked change that can be accepted or rejected. In regulated work, it's a standard operating procedure (SOP) checked against the relevant standard before approval which creates evidence for future reviews or audits.
And that's a major point to consider; an agent can only check work against what it can read. If the current playbook is a Word document on a shared network folder and there are multiple versions of the SOP in circulation, then what should an agent check?
This is why document management matters for AI.
"Nobody looked at the contract since we signed it"
Obligations, service levels, renewal dates and closure conditions sits in signed contracts that nobody looks at until something goes wrong. "Nobody checked" isn't just a mistake, its an audit finding in regulated work.
An agent runs when a contract reaches a defined point in the negotiation process. It writes a risk level with a prioritized list of issues. Once the contract is signed, it pulls out the obligations and key dates that can be filtered. In quality management, an equivalent agent checks that corrective action addresses the root cause before anyone closes a non-conformance issue.
Results are recorded without a need to search for them in a chat message thread or email.
"The first draft eats too much time"
A drafting agent working from won proposals, including the source behind each section, is a very different proposition than creating plausible sounding text from thin air. We are all wary of AI filler so strong sources and citations matter. If I can't see which won proposal content came from, I must verify it manually which doesn't save me time. Highlighting where there are gaps to support the proposal drafting due to a lack of relevant source materials is just as important. This is all possible when your content is managed in business context.
The oversharing question
This comes up quite often. Firms and businesses are concerned that AI could read and use information that it shouldn't. Perhaps because of an old draft still visible or of what is in Teams messages, or because security policies haven't been properly applied.
The answer isn't to lock AI out of your content. Its permission aware intelligence based on what the document is and how it relates to other content. All the agents I have discussed would work inside those rules would only see and use the same information as their user. More on that here
Where to start
All of these agents exist today and are available through our industry and business product portfolio; M-Files for Tax Advisory, M-Files for Consulting, M-Files for Quality and M-Files for Contracts. If one of the issues I've discussed came up for you and your team this week, or this month do get in touch.
Though I would add one caveat to all of this…start where people already trust the documents. If you don't know where the foundation is at its weakest, then our AI Readiness Assessment is a great place to start.
