Less judgment than planning, more paperwork than anything, and the place a T&E practice loses its margin.
Start a conversation with the AI Adoption Concierge, already scoped to probate & administration. Pick a starting point, or describe your situation directly.
Administration is where the economics of a trusts and estates practice are usually decided. The legal judgment is concentrated in a few decisions; the volume is in assembling an asset picture from bank statements, brokerage records, tax filings and property records, keeping a deadline chain, and answering the same beneficiary questions repeatedly for months. That distribution — low judgment density, high document volume, repetitive correspondence — is close to the ideal profile for AI assistance, and it is also the work most likely to be written off or absorbed. The caution specific to this area is that some of these deadlines are jurisdictional and unforgiving, and a missed one is a malpractice event rather than an inconvenience.
Ranked by how much unbilled time each typically consumes.
Extracting holdings, account numbers, balances and titling from statements, tax filings and property records into a structured schedule.
Flagging accounts referenced once and never again, income with no corresponding asset, and gaps that suggest something has not surfaced.
Notice periods, creditor windows, filing dates and tax elections. Computed by rules-based systems, never by a generative model.
Status updates and plain-language explanations of what is happening and why, drafted from matter activity and approved before sending.
Assembling inventories and accountings into the form the court expects, verified against source records.
Why this takes so long, when distributions happen, what a fiduciary may and may not do — answered consistently rather than freshly each time.
How to run this safely.
Fiduciary duties run to people who did not choose you and frequently do not trust each other.
Missed deadlines are consistently among the largest single categories of legal malpractice claims across practice areas. Rules-based docketing with human verification is the control. A generative model may read a court notice and propose dates; it must never be the system of record for them, and a second human should confirm every extracted date.
It can build the first draft of one, and that is genuinely valuable because the alternative is a paralegal transcribing statements. Extraction from bank and brokerage statements, tax filings and property records into a structured schedule is well within current capability. What it cannot do is confirm the schedule is complete — the assets nobody mentioned are invisible to a tool reading only what it was given. The reconciliation step, tracing every figure to its source and asking what is conspicuously absent, stays human and is where the professional value sits.
Drafted, not sent. Proactive status updates measurably reduce the inbound "where are we" traffic that consumes administration time, and drafting them from matter activity is straightforward. But beneficiaries are not clients, they frequently disagree with each other, and administration correspondence lands in the middle of grief. An automated message that arrives at the wrong moment, or that says something slightly wrong about a distribution, causes damage disproportionate to the time it saved. Draft with AI, approve with a human, always.
In the work that was never being billed. Most administration matters carry a quantity of reading, chasing and re-explaining that gets absorbed because it cannot be justified line by line to a family. Compressing that is not a billing gain, it is a margin gain — and for firms doing administration on a fixed or percentage basis it is a direct one. Firms should measure it as time redeployed rather than hours saved, because the evidence from other professions is consistent that recovered time gets reinvested rather than banked unless someone deliberately redirects it.
It is a real constraint on how far to push automation, and it is under-discussed. Consumer research in adjacent legal contexts has consistently found significant resistance to AI appearing in client-facing communication, with resistance rising sharply among older clients — who are disproportionately the people in this practice. The workable pattern is AI behind the service and a person in front of it: use it to assemble, extract, draft and track, and keep the human in every conversation that touches a death, a family disagreement or money someone was expecting.
Describe where the unbilled hours go. The Institute will help you find what to change first.