In the fourth of a series on safely adopting AI, Ed Molyneux sets out a four-rung ladder from copy-paste to fully governed, and argues most firms are standing a rung lower than they think.
Last week’s article described a destination: one controlled route to client data that authenticates the user, scopes access to their permissions, minimises what it releases, and records every request. No firm arrives there on Monday, and it is not the only place worth standing. Between personal accounts and that route lies a gradient, and each step along it buys a specific, nameable capability.
Firms tend to talk about AI as though there were only two states, allowed and banned. That framing is useless, because no firm is in either box. Knowing which rung you are actually on is most of the work.
The bottom rung is personal accounts and copy-paste. Staff use consumer tools on their own logins, and data goes in by hand, one paste at a time. There is no enforcement and no audit trail: ask whether client data has gone into an AI tool and the honest answer is that you do not know.
The trap is believing you are not on this rung. Most firms that think they have banned AI are sitting squarely on it, because the ban did not stop the activity, it moved it out of sight onto personal phones and home laptops. A prohibition you cannot enforce is not a rung below this one. It is this one, with worse visibility.
The first real step up is a managed home for AI. The firm moves onto an administrator-managed business plan, accounts are firm-owned and tied to your existing staff sign-on so joiners and leavers are handled once, and the settings that matter, data retention and whether the vendor may train on your material, are configured centrally rather than left to each individual. Enforcement is partial: you control who has access and under what settings, but not whether someone follows good practice on a given task. The trail is host-level: you can see that a conversation happened and administer retention, but not which client data was in it.
This is the biggest single jump available to most firms, and it is procurement and configuration rather than engineering. Firm-owned accounts, sensible settings, and a short written policy naming the approved tools. Most firms should be here before they attempt anything cleverer, and many are one purchasing decision away.
The next step is the controlled door itself. Instead of being fed by copy-paste, the AI reaches matter data through one interface that checks and records every transit. This step changes the shape of the problem, for a reason worth stating plainly: people paste files into chatbots because the AI has no other way to see the matter. Give it a governed route and the paste stops being necessary, and because every request passes through the door you enforce what a policy can only ask for. The trail is genuine too, at the level of what was disclosed rather than merely that something happened. The compliance record stops being a spreadsheet somebody is supposed to maintain and becomes a by-product of using the system.
The top rung has no open chatbox at all. The AI works inside the firm’s own case management or transaction platform, producing its analysis against data the system already holds, under the firm’s existing terms. Enforcement is total within that workflow, but it only covers the workflows somebody has built, and does nothing for the ad hoc, one-off work, which is exactly where the risky paste happens.
The last two rungs are complements rather than rivals. One industrialises routine, high-volume tasks; the other governs the improvised long tail. A firm that has done only the first has automated its known processes and left its unknown ones, the riskier ones, running on personal accounts.
What to do with this is fairly blunt. Find your real rung, without blame: ask people what they are actually using, because the answer is almost always more than you think, and a firm that punishes the honest answer will never get an accurate one again. Then get to a managed home, which for most firms is the highest-value move available this quarter. Then pick one high-volume flow and close the paste gap on it, starting where staff paste most, because that is where both the risk and the time saving concentrate. And accept that you cannot cover everything. A policy that admits the long tail exists and manages it is far more defensible than one that pretends it away.
One honest caveat runs through all four rungs: none of this makes the answers correct. Governing where data goes, and whether it is recorded, is a different question from whether the output is any good, and a well-governed system can still be confidently wrong. That deserves its own evidence, which a later article takes up. Before then, the most practical tool of the lot: the questions to ask a supplier before believing a word of their governance claims.
About the author
Ed Molyneux is co-founder and CTO of Moverly and the original author of the Property Data Trust Framework (PDTF), the open standard for machine-readable property data now being adopted across the industry. Ed writes about AI, property data infrastructure, and the future of conveyancing.
The views expressed in this article are those of the author and not necessarily those of Today’s Conveyancer. This article is general information, not legal advice.
