A recent article in these pages proposed a ‘rework ledger’ for measuring the true cost of conveyancing AI. Ed Molyneux agrees, and argues that the firms which treat that rework as a cost to tally will be beaten by the ones that treat it as something to learn from.
A recent article in these pages, by the behavioural scientist Gleb Tsipursky, responded to this column with something every firm piloting AI should adopt. He calls it a rework ledger.
Rather than measure only how fast an AI produces its first output, you also record the minutes spent checking it, the minutes spent correcting it, the escalations to more senior staff, and the downstream work caused by an error that slipped through. Then you calculate the net time saved. It is a genuinely useful discipline, because a task that appears to fall from thirty minutes to five can quietly cost more than it did before once verification and cleanup are counted. He is right, and I would encourage firms to keep that ledger.
But I want to add a column to it, because as described the ledger is a decision tool. It tells you whether to scale a workflow or stop using it, and it treats each instance of rework as a cost to be tallied. That is necessary, and it is also incomplete, because a mistake is not only a cost. It is information. The sharper question is not just whether a workflow saved net time this month, but whether the firm became permanently better because of what went wrong.
Picture two firms running the same pilot. Both log 40 minutes of rework on a workflow in a week. The first tallies it, decides the net saving is marginal, and moves on. The second asks, for every correction, a further question: why did the system get this wrong, and what do we change so it cannot happen again? Perhaps a data source the AI could not see gets wired in. Perhaps a check is added, or a house standard finally written down, or a guardrail set so a particular error is designed out. The first firm has recorded a cost. The second has written a curriculum. Over a year, the second firm’s rework on the tasks it has already met falls towards zero, while the first firm keeps paying the same tax every week.
This is the investment worth making, and most of it is not a technology investment. It is the discipline of turning each misstep into a durable change to the system: to the data the AI can reach, to the checks it runs, to the standards it is held to. A mistake that happens once and is engineered out is tuition. A mistake that recurs every week is simply waste. What separates the two is whether the firm has built a feedback loop at all, a route that runs from “this was wrong” to “this is now much harder to get wrong”. Without that loop, a firm pays for its errors twice, once in the rework and again in never learning from them.
There is a warning inside this that connects to the measurement point I made a fortnight ago. The errors most worth learning from are the quiet ones. An AI that over-flags a possible issue creates visible, annoying rework that lands squarely in the ledger. An AI that under-flags a real issue creates a much quieter and more expensive kind of risk, one that may not appear in the ledger at all until it surfaces on a completed matter. A firm that only learns from the rework it can see will steadily tidy up the irritating false alarms while doing nothing about the failures that actually carry liability. The feedback loop has to be pointed deliberately at the dangerous errors, not just the ones that shout.
The direction of the reform roadmap makes this more pressing, not less. Common data standards and clearer expectations for the use of AI in conveyancing are coming, and the firms that arrive with a working learning loop already turning will absorb those changes far faster than the firms still deciding whether AI earns its place. The capacity to improve is itself the capability, and it compounds.
So keep the rework ledger, and add one column to it: what did we change so this never recurs? The cost of rework is what you pay to find out where your system is weak. The value is what you build once you know. A firm that treats its AI’s mistakes as training data gets quietly and permanently stronger. A firm that treats them only as a cost stands still while the bill repeats. That, far more than the raw minutes saved, is the number worth watching.
Safe AI in Conveyancing, Ed’s series on adopting AI in a way a firm can genuinely stand behind, resumes next week.
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 those of Today’s Conveyancer.
