3CK โ a narrowing into estate & bequest administration
AI agents for bequest administration.
When someone leaves a gift in their will, the paperwork lands on a desk. Most of it takes an afternoon. Some of it takes years โ and you can't tell which until someone has read the file.
Try the demo โ no login required โWhat bequest administration actually is
Someone dies. They left money to a university. Now somebody has to sort it out.
Read the will
A 20 to 50 page document arrives. Somewhere inside is one paragraph naming the university โ how much, to which department, and what it must be spent on.
Work out what it means today
The will might be thirty years old. The department it names may have merged, been renamed, or closed. Someone has to work out what it is called now, and prove it.
Make sure the money arrives, and is spent as promised
Chase the executor for two years. Check what arrives against what was owed. Then keep the gift spent the way the donor wrote it, for as long as the fund exists.
All three are done by hand today, by people with too many files. We build the AI agents that do the reading, the matching and the watching โ and hand a human the evidence to decide on.
Built for institutions holding restricted gifts
The demo
A resolved restriction, with evidence, in under a minute.
Live, not a mock-up. A language model reads every document below โ every field scored, with the sentence it came from. Scans go to a vision model, and the result names the model that ran. Entity matching is deliberately not a model: a guess is not evidence.
Every gift, tracked from will to distribution
We never touch it. Probate does.
No custody, no fiduciary duty. We make sure the funds are handled correctly at two stages.
At administration
We verify the money actually arrives, on time and in full.
Across the endowment's life
Institutions evolve. Faculties merge, programs close. The original gift terms do not. We monitor their published changes against every gift agreement they hold, and protect the money two ways.
Forward
Catch mismatches before funds are misapplied โ a loss prevented.
Backward
Find funds already stranded, sometimes for decades, and redirect them to the closest active program the institution still runs โ cash unstuck. Nobody re-checks these. We do, across the whole corpus.
The institution that never loses a gift.
Post-death is where the field thins. Most software stops when the will is written. We watch what happens after โ so a gift naming a department from 1994 still finds its way, and a fund nobody has checked in twenty years gets found.
How it works
We surface and evidence. A human decides.
Three agents run today, live, on every document. More join the corpus as we build.
Identifies what kind of document just arrived โ a will, a codicil, an executor's accounting โ and routes it to the right case.
Extracts donor, executor, gift type, amount or share, named designation, restrictions and dates โ with a confidence score and the exact source sentence for every field, so nothing is a black box.
Asks one question โ does this restriction still work? โ matched against our historical entity map, with the archive evidence chain attached. A curated map, never a model guess.
Coming next โ roadmap, not yet built
Probate Agent
Watches probate registries and death notices, and matches them against the institution's own history โ including names it hasn't used in decades.
Watcher Agent
Reads the institution's published changes โ mergers, closures, renames โ and flags every instrument each one touches.
Dormancy Agent
Screens the standing endowment corpus for funds whose purpose no longer maps to anything the institution does.
Asset Verification Agent
Reads an executor's interim accounting and checks it against what actually arrives.
Time saved
Hours become minutes.
One person does one file at a time, which is why files sit for months. Agents do not queue.
Money saved and made
And the numbers behind them.
The average bequest is $48,000. Today the hard ones go to a law firm on a 3% contingency โ $15,000 on a $500,000 estate, and again on the next one. We are flat: about $36,000 a year for a large university, however many estates come in. Two or three hard estates and we have paid for ourselves, before finding them anything.
Where the 120 files a year comes from
Roughly 1,840 working hours a year against 8โ15 hours per restricted file is a hard ceiling of about 120โ230 files a year โ and that assumes an officer does nothing else.
How we charge
- Annual subscription, tiered by instruments monitored โ restricted funds, endowments, and outstanding bequest intentions.
- ~$9,000 small college. ~$36,000 large university. $120,000+ largest endowments.
- Plus a fixed fee when we surface an estate the institution had no record of.
Team
Matthew
Legal researcher on wills and bequests, University of Sydney Advancement Services and Office of General Counsel. Founding engineer at Fluent Forever (0 โ $2.5M ARR).
Phuong
UNSW Centre for Social Impact.