Ed Molyneux asked in Today’s Conveyancer how a firm would know when a conveyancing AI is wrong, which behavioural scientist Gleb Tsipursky says is an “unusually timely” measurement problem. HIs argument takes the next step: supplier accuracy tests matter, but firms also need to measure the hidden checking, correction, escalation, and downstream repair that appears after an AI workflow enters live practice. He gives conveyancers a practical 30-day rework ledger for deciding which tasks deserve more automation and which only move work out of sight.
The most important AI metric in a conveyancing firm may be the work the software creates after it appears to have finished a task.
Today’s Conveyancer recently asked how a firm would know when a conveyancing AI is wrong. That question is essential. Vendors should be able to show what their systems were tested against, who established the correct answers, and where errors carry the greatest risk.
But supplier accuracy is only half the measurement problem. Once a tool enters a live workflow, a firm needs to know whether the promised efficiency survives contact with actual files, actual clients, and actual professional judgment.
That requires a rework ledger.
A 30-day timeline
For 30 days, take one recurring AI-assisted workflow, such as summarising title material, extracting data, preparing an initial report, triaging documents, or generating a first draft. Record five things: gross minutes saved, minutes spent checking the output, minutes spent correcting it, escalations to more experienced staff, and downstream work caused by an error that escaped the first review.
Then calculate net time saved.
This sounds obvious, but many technology pilots measure the speed of the first output rather than the cost of making that output safe enough to use. A task that falls from 30 minutes to five looks transformative. If another 12 minutes go into verification, five into correction, and a colleague later spends 10 fixing a consequence, the business case changes completely.
Measuring workflows
The ledger should also distinguish between types of work. An AI tool that over-flags possible issues may create annoying but visible rework. A tool that under-flags a material issue can create quieter and much more expensive risk. Those two error patterns should never be blended into a single productivity number.
The government’s 2026 home-buying and selling reform roadmap points in the right direction. It explicitly identifies document classification, triage, and data extraction as tasks AI may automate while preserving human judgment for decisions. It also says clear standards for appropriate AI use in conveyancing will be developed and that the AI Growth Lab will support responsible testing in legal services.
Firms do not need to wait for those standards to start measuring their own workflows properly.
Process and accountability
A useful pilot should have a stop rule before it starts. Decide what level of net time saving, correction burden, and escalation rate would justify expansion. If the workflow misses the threshold after 30 days, redesign it, narrow its scope, add better data or guardrails, or stop using it for that task.
This approach also improves adoption. Employees become skeptical when leaders celebrate theoretical time savings while staff quietly absorb the checking and cleanup. Asking people to record rework treats their experience as operational evidence rather than resistance to technology.
That matters because conveyancing combines repeatable process with professional accountability. AI can handle more of the repeatable work, but only if firms can see the full cost of supervision and recovery.
The best automation does not merely produce an answer faster. It reduces the total work required to reach a result a conveyancer can responsibly stand behind.
About the author
Gleb Tsipursky PhD is a behavioural scientist, CEO of Disaster Avoidance Experts, and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026). His commentary has appeared regularly in The New York Times, The Guardian, the Toronto Star, and many others.
The views expressed in this article are those of the author and not Today’s Conveyancer.
One Response
One of the more sensible takes I’ve read on AI in conveyancing — measure net time saved, not headline time saved. The line that jumped out was that under-flagging is the “quieter and much more expensive risk”: exactly the same is true in security, where the problem you don’t spot always costs more than the false alarms you do. Build the verification habit into AI workflows from day one and you keep both the productivity and the professional accountability.