NEOEDGE AIA document, underwriting-support, servicing or reconciliation workflow is holding up funding, service or reporting. You need to compare AI with a vendor product, simpler automation, or a process change.
← All AI decisions
Decision Sprint · For the AI decision on your desk this quarter
One AI decision. Real cases. A call you can defend.
In about two weeks, I test the options on your own cases and hand you a written call (build, buy, fix, continue, scale or stop) with the evidence and a named owner.
Sam Talasila Former Head of LLMs & Data Products, Wealthsimple
What you leave with
A decision record your team owns.
A short, plain document you can take to leadership, your board, or a vendor.
Decision record · Mortgage document reviewSynthetic example
- Call
- Fix before rollout. Routine files worked. Two gaps need fixing first.
- Tested
- 40 real files: 31 handled correctly, 6 needed a reviewer’s correction, 3 missing pages went unflagged.
- Next move
- Add a missing-page check, send non-standard income to a reviewer, then re-test on 20 new files.
- Owner
- Underwriting operations lead, with a named reviewer for exceptions.
Decision record · ContentsEvery sprint
- Test findings on routine, exception, and known-failure cases from your own work.
- Options compared: an AI approach, a vendor product, simpler automation, or a process change.
- Gaps and reviewer effort: what still needs a person, and how much of their time it takes.
- Cost and control assumptions: what it would take to run, and what has to be in place first.
- The recommended call, with the evidence for it and a named owner.
- If a pilot makes sense, what it should test and when to stop it.
When it fits
Three moments where a sprint earns its fee.
Each has a real workflow, a decision someone owns, and a cost to getting it wrong.
An AI tool gives unreliable answers, adds rework, or goes unused. You need to decide whether to fix it, stop it, or roll it out.
A proposed tool still has open questions about accuracy, cost, exceptions, privacy, or who owns it, and a contract is close.
How the two weeks run
Nothing is tested until the work is understood.
- Before we start
Agree the decision.
We name the workflow, the decision, who makes it, and by when. Your team arranges an authorized sample of cases and access to approved tools.
- Week one
Map the real work.
I sit with the people who do it: the handoffs, the checks, and the exceptions everyone already knows about. We set a baseline and what “better” has to mean.
- Week two
Test on your cases.
Routine cases, hard cases, and known failures. I record what worked, what needed correcting, and how much reviewer time each option really takes.
- The last day
Make the call.
We walk through the decision record with your team and whoever signs off, so the people who live with the decision understand it.
From a real engagement · Canadian mortgage lender · DevelopmentTrial first
Choosing an AI testing tool
“Which of these vendors is worth a contract?”
Software testing was slowing releases, and vendors were pitching AI testing tools. I compared more than 15 tools and recommended a short trial of two on the same tests, using made-up data so no client information left the company. A small trial instead of a contract.
Read the case study ↗︎Scope
Bounded on purpose.
A clear edge keeps the sprint quick, affordable, and honest about what it can show.
What you bring
A few people and real cases.
- A process owner who will make or sponsor the decision
- An authorized sample of routine, exception, and known-failure cases. We set the size once I’ve seen how complex they are.
- Time with the people who do and review the work
- Your technology owner, plus risk or compliance where client data or credit decisions are involved
Not part of the sprint
Kept for later, or elsewhere.
- Building or running a production system
- Open-ended implementation or extra staff
- A production support agreement
- Formal regulatory assurance or independent model validation
After the sprint
Three honest next moves.
Your team takes it forward.
The record is yours. Use it with your own team or a vendor you choose.
A clear reason to stop.
If the evidence says it isn’t worth it, you have that in writing and can move on.
I build it with you.
If a pilot makes sense, I scope it separately, build and test it with your team, then hand it over. About the pilot ↗︎
The offer
Everything to make the call, and act on it.
On our first call we size the decision. If it isn’t worth at least ten times the fee, I’ll tell you a Sprint isn’t worth it.
What’s included
- The decision recordThe call, the evidence from your cases, the options compared, and a named owner.
- Vendor Question PackThe questions to put to any vendor, drawn from what your own test showed.
- Board one-pagerOne page your leadership or board can read in two minutes.
- Pilot spec, if the call is buildWhat a pilot should test, how to judge it, and when to stop.
- 30-day follow-upA review a month later to check the decision is holding.
My guarantee
Clear Call Guarantee
You get a written call with a named owner: build, buy, fix, continue, scale or stop, backed by evidence from your cases. If the record leaves the decision unclear, I keep working at no cost until it doesn’t.
Stop is a clear call too. I guarantee the decision is clear and evidenced, not that AI is the answer.
Why now
I run two Sprints at a time. I do the work myself, so places are limited.
We work back from your decision date: a vendor deadline, a budget cycle or a board meeting. Starting earlier means more of your cases get tested.
At one lenderVendors were pitching AI testing tools. I compared more than 15 and recommended a short trial of two on the same tests, using made-up data. A small trial instead of a contract.
Questions
Before you book.
Is a Sprint worth it for our decision?
On our first call we size what’s at stake: the contract, the build, the hours, the risk. If the decision isn’t worth at least ten times the fee, I’ll tell you a Sprint isn’t worth it.
Does the answer have to be AI?
No. I compare an AI approach with a vendor product, simpler automation, and a process change. Sometimes the best move doesn’t involve AI at all.
What if the evidence says stop?
Then the record says stop, and why. That can save more than any rollout, and you won’t have the same debate again next quarter.
How will you handle our client data?
Before we start, we agree which data the test can use, in which approved tools, with what permissions and retention. If real data can’t be used yet, we start with realistic samples.
Do we need an AI pilot already?
No. Many sprints start from a manual process. Others start from a pilot that isn’t working. Both need the same thing: one workflow and real cases.
Is there a smaller first step?
If your team first needs hands-on practice with AI tools, the one-day Hands-on Workflow Workshop is an optional place to start.
Which decision is on your desk?
In 30 minutes, tell me about one workflow, pilot, or vendor decision. We’ll see whether a Decision Sprint, the Workshop, or neither is the right first step.