AI has turned up in just about every corner of professional services.
Accountants are using it to summarise documents. Solicitors are leaning on it for research and drafting. Financial advisers are testing it on admin and client communication. Ag consultants are using it to capture and organise the information they’d otherwise lose in a notebook on the ute dashboard.
For most of that work, a general chatbot does a genuinely good job. It’s fast, it’s cheap and it’s right there in the browser.
Succession planning is where it comes unstuck — and not in the way you’d expect.
We tried it properly before we said any of this
Before building anything, we ran a full succession engagement through an off-the-shelf chatbot. Not a five-minute test. A real, messy, multi-meeting family scenario of the kind you’d actually pick up.
The first impression was excellent. It asked reasonable questions. It wrote summaries that read well. It picked up on tone. If you’d shown the output to a client, they’d have been impressed.
Then we went back and checked it against the source material.
Things had gone missing. Not dramatically — nothing crashed, nothing came back garbled. Details from earlier meetings had simply stopped appearing. A comment made in month two about who’d actually been paying the machinery loan never showed up again. An assumption someone made early on, later corrected by the accountant, was still being treated as fact in a summary six weeks later.
Here’s the part that should worry anyone doing this work: none of that announced itself.
You don’t get an error message when a chatbot loses track of something. You get a confident, well-written answer with a hole in it. And you’d need to already know what was missing to notice it was missing.
That’s not a knock on the technology. It’s doing exactly what it was designed to do — hold a conversation. It just wasn’t built to carry a three-to-nine-month engagement with a family’s assets and relationships riding on it.
Why succession breaks a general chatbot
Sit down at a kitchen table and you’re dealing with forty years of family history, a few million dollars of assets, conflicting memories, promises made in a shed in 1998, retirement plans nobody has said out loud, trust structures, debt, tax — and a lot of feelings.
That information doesn’t arrive in one sitting. It comes out in pieces over months. You’ll talk to family members separately. Financial details get confirmed weeks later. Assumptions get corrected as documents land and other perspectives come into the room.
A chatbot thread handles that badly for three reasons.
It has a limit, and it doesn’t tell you when you’ve hit it. Long engagements outgrow what the model can hold. Older material quietly falls out of view. The conversation keeps flowing as though nothing’s changed.
It can’t tell the difference between what was said and what was confirmed. Dad reckons the home block is worth $4 million. The valuation comes back at something quite different. To a chatbot, both statements are just text in a thread — and the older one is as likely to be repeated back to you as the newer one. Everyone assumes the company owns the machinery until the accountant says it doesn’t, and that correction has no more weight than the original guess.
There are no walls inside the thread. Everything you’ve typed sits in the same pool. The private concerns an adult child shared with you in confidence are sitting right alongside the summary you’re about to generate for the parents.
Individually, each of those is manageable if you’re watching for it. Across a nine-month engagement, with three or four families on the go, you won’t catch every one. Nobody would.
What that actually risks
The realistic failure here isn’t a wildly wrong answer that you’d spot immediately. It’s a good answer that’s quietly incomplete.
A structure recommendation that didn’t account for the off-farm sibling’s earlier contribution, because that detail dropped out of context in April. A fairness discussion that proceeds without the house deposit one child received in 2016. A summary that repeats back something an adult child told you they didn’t want repeated.
Any of those can be recovered from if you catch them. The problem is that catching them depends entirely on the facilitator’s own recall — which is exactly the thing the technology was supposed to be helping with.
You can absolutely use a general chatbot to speed up parts of your work. Draft a letter, tidy up meeting notes, summarise a document you’ve just read. That’s fine, and it’s useful.
What it can’t do is hold the file. Decisions in succession need a memory that’s structured, checkable and knows what’s been confirmed — not a long thread that’s doing its best to remember.
What we built instead
Succession Navigator was designed around the failures we found.
It stays out of the conversation. The quickest way to wreck one of these meetings is to put a screen between you and the family. Nobody wants to watch their adviser type, and the good stuff almost never comes from the answer in the field — it comes from the story wrapped around it. A farmer starts telling you about the year he bought the second block and mentions, almost in passing, that his eldest has effectively been running it for a decade. You don’t catch that if you’re working through eighty tick-boxes. So the system works like a quiet analyst in the corner: it follows the discussion, tracks what’s been covered, and shows you what still needs to be understood. When you want a hand, it’ll suggest the next question worth asking.
Confirmed information carries the weight. The AI can draw on the full history of the engagement, but approved facts — relationships, ownership, financials — become part of a structured case record. An offhand comment from a meeting back in March doesn’t quietly harden into fact by August. And the family isn’t answering the same questions twice because the software has forgotten.
Confidentiality is part of the workflow, not a matter of interpretation. Information is separated by where it came from and how it can be used, so individual family input doesn’t automatically flow back to parents or siblings. The point isn’t to bury difficult issues. It’s to make sure they’re raised deliberately, by the professional running the process, rather than accidentally by a piece of software.
It knows what it isn’t. Succession touches law, tax, finance, lending and estate planning, and those calls belong to qualified people. The system helps you gather information, spot gaps, surface inconsistencies and ask better questions. It doesn’t decide which child gets the farm, pick a structure, or hand out tax strategies — and it doesn’t replace the accountant, solicitor, planner, lender or ag adviser. If anything, one of its more valuable jobs is flagging exactly where those people need to be brought in.
The human stays responsible. The AI can suggest, draft and analyse. The facilitator decides what’s accepted into the case and how the engagement moves forward. For a firm carrying the liability, that line needs to be clear.
It works where farming actually happens. Not every meeting takes place in an office with five bars and fibre to the door. Plenty happen at a kitchen table or somewhere the signal comes and goes with the weather, so the system protects the conversation when connectivity drops rather than making everyone start again. Some of these things are hard enough to say once.
The point isn’t that AI is the problem
It isn’t. Used properly, AI is the best thing to happen to this kind of work in a long time, because it lets the adviser spend more of the meeting actually listening.
The point is that succession is a long, fragmented, high-stakes process, and a tool built for conversation isn’t built to carry it. The gap between how good the output looks and how complete it actually is doesn’t show up on day one. It shows up in month seven, in a room full of family, when something important has quietly gone missing.
That’s the gap Succession Navigator was built to close.
You don’t get an error message when a chatbot loses track. You get a confident answer with a hole in it.