Services
AI-ready process improvement. Don't pay to automate a broken process.
AI-ready process improvement means making a process stable, measured and standardized before you automate it, then picking the specific steps where AI will pay off and sizing what each one is worth. AI doesn't fix variation. It scales it. If 4 people work an exception queue 4 different ways, the software scales the rework. We standardize the handoffs, cut the waste, then show you which steps AI should take over, with a dollar case your CFO can check.
By Bob Buckwalter, Lean Six Sigma Master Black Belt, PMP, MBA · Updated October 25, 2026
The order that works
Step 1
Standardize how the work is done
Step 2
Measure it for a few weeks
Step 3
Sort out the data
Step 4
Choose the platform
The Problem
Most AI pilots stall in the same place.
Most companies start AI at step 3 or 4: where's the data, which platform, who hosts it. Step 1 gets skipped. That's working out how the work gets done, and getting it done the same way twice.
If 4 people run a process 4 different ways, the model learns all 4. Exceptions with no owner become exceptions the model guesses at. Here's what that looks like on the floor: a routing rule sends every file with a blank reference field to a pend queue, nobody owns the queue, and it quietly grows until a customer calls. 6 months in, the pilot stalls and the technology takes the blame.
That isn't a data problem. It's a process problem, and process problems are what we've fixed for 20+ years in mortgage servicing, banking and insurance operations.
For context: in a senior operating role at a national commercial mortgage servicer, Bob Buckwalter's process work cut the error rate 90%, halved cycle time and produced $4M+ in annual savings. A process that stable is one AI can work with.
What AI-Ready Looks Like
One way of doing the work
The process runs the same way no matter who's on shift. It's written down and it has a named owner.
A baseline you trust
Cycle time, error rate, rework and exception volume, measured before anything changes. That's how you'll prove what the AI did.
Exceptions with a home
Every exception has a defined path and an owner, not “whoever notices.”
One source of truth
One agreed system of record, not competing spreadsheets that disagree by Thursday.
How We Work
Fix the flow first. Then put AI where it pays.
The method is Lean Six Sigma. The question it answers is new: which of these steps should a person do, which should a machine do, and which shouldn't exist at all?
-
1
Map the process as it runs today
With the people who run it. SIPOC and swimlane, exception paths included, because that's usually where the volume hides.
-
2
Measure it
Cycle time, cost per transaction, error and exception rates. A few weeks of real numbers beats a vendor's estimate.
-
3
Remove what shouldn't be automated
Redundant reviews and approval loops don't need AI. They need to go away. Automating them makes waste run faster.
-
4
Tag every remaining step
Fix it, standardize it, or AI candidate. In financial services the strong candidates tend to be document intake, classification, triage and first drafts a person reviews.
-
5
Size each candidate
Hours, error cost and cycle time per step, so the AI business case is built on your baseline and holds up in front of a CFO.
-
6
Hand over the controls
Owners, measures and a review loop, so your team sees when output quality drifts and can keep improving it without us.
Automating a process as-is vs. fixing it first
| Automate it as-is | Fix the process first | |
|---|---|---|
| Variation | Every way of doing the work gets automated, including the wrong ones | One documented way, so the model learns one pattern |
| Exceptions | Land in queues nobody owns and pile up | Have a defined path and a named owner before go-live |
| Business case | Built on the vendor's estimate | Built on your own measured baseline |
| After go-live | Rework moves downstream and the pilot takes the blame | Controls and an owner show when output quality drifts |
Where It Fits
The steps we look at first.
These are the kinds of steps that tend to come out of the tagging as AI candidates, once the process around them is stable.
Mortgage servicing
Document intake and indexing, exception queues, borrower correspondence drafts. See commercial mortgage servicing.
Insurance
First-notice-of-loss intake, claims triage and routing, correspondence. See insurance operations.
Banking
Wire and payment exceptions, KYC document review, reconciliation breaks. See financial services.
What We Don't Do
We don't sell software.
No platform to resell and no license quota to hit, so the recommendation is about your process. Sometimes the right answer is a simpler rule, a form, or deleting a step. We'll say so.
When it's time to build, we work alongside your IT team, your vendor, or one of our strategic partners.
AI Readiness Assessment
AI readiness assessment: 3 ways to find out if a process is ready.
Free · self-serve
Automation Readiness Score
Rate 12 statements about 1 process and get a verdict on whether it's ready to automate, with the gaps that matter most.
Score a process →$7,500 fixed · 2 weeks
First Win Diagnostic
1 process, mapped with your team. You get a SIPOC, 3 to 5 ranked improvements and a 90-day plan. Ask for the AI lens and every step gets tagged fix, standardize or AI candidate.
See the diagnostic →Free · 1 hour
Process Mapping Training
Teach your team the SIPOC in 1 hour, then let them map the process AI is supposed to fix. It's a fast way to find the 4 versions of it.
Book the session →Common Questions
Questions about AI and process improvement.
What is AI-ready process improvement?
It's getting a process stable, measured and standardized before you automate it, then choosing the steps where AI will pay. It's Lean Six Sigma pointed at one question: what should AI do here, and what should go away entirely?
What is an AI readiness assessment?
A review of whether a process is ready for AI. Is the work done the same way every time? Is there a baseline? Does every exception have an owner? Is there one source of truth? Our free Automation Readiness Score asks 12 of these questions about 1 process. The First Win Diagnostic answers them with your team over about 2 weeks.
Do we need to fix the process before using AI?
Usually, at least the first step. AI repeats what the process does. If the work varies by person, or exceptions have no owner, the model will be inconsistent too, only faster, with fewer people watching.
Which processes are good AI candidates?
High volume, rules you can write down, and a person who can check the output. Document intake, classification, triage and first drafts are common examples. Steps that exist only because of rework should be eliminated, not automated.
Do you build or sell AI software?
No. We don't sell a platform, so the recommendation is about your process, not a license. We get the process ready, pick the candidates and build the business case. When it's time to build, we work with your IT team, your vendor or our partners.
How do we get started?
Score 1 process with the free Automation Readiness Score, or book a free call at windyhillpartners.net/contact.html or 610-316-4056. We'll talk through the process you have in mind and whether it's ready.
Before you sign the AI contract, let's look at the process.
Bring one intake process or exception queue to a free call. We'll tell you honestly whether it's ready, and whether a 2-week First Win Diagnostic is the right next step.
Book a Free Call →