An alternative to hiring an AI strategy consultant
Instead of paying for an AI strategy, pick one customer journey, measure how long it takes today, then build a rough working version of an AI feature and watch real people try to use it. What you learn from that beats a strategy document.
This is for the person who has to show their boss something real from the customer journey work this month. Not another vendor deck. Something a customer used, with a number next to it.
You have probably already had the meeting where someone offers to assess your readiness and come back in six weeks with a roadmap. It will be a good roadmap, and it will sit in a folder.
The other route is smaller. Take one customer journey, get a before number, build a rough working version of an AI feature on it, and watch five real users try it. Two weeks of work instead of six months. You end up with something to show and something to compare against.
A strategy deck will not tell your boss whether anything changed
A readiness assessment usually runs like this. Interviews with the senior people. A maturity score. A ranked list of AI ideas, with impact and effort scores that are really guesses. Then a workshop where the same people agree with the ranking they were shown.
None of that touches the onboarding flow. Nobody has watched a customer stall on it. The document describes what the bank could do in theory, not what breaks when a branch officer is trying to get an application through.
So when your boss asks whether the AI work was worth it, all you have is the roadmap. That is not proof. The honest answer is that you don't know, and six months in, that is a hard thing to say out loud.
Instead, pick one customer journey you already know has a problem. Long turnaround time. High drop-off at one step. Then find out what is actually happening there before anyone writes a plan for it.
Get a before number, even a rough one
Most teams we work with have no baseline. Not because they are careless. Because nobody asked for one at the start, and by the time the question arrives the old version is gone.
Pick one number you can measure this week. Average turnaround time on one application type. Completion rate from start to submit. Number of calls to the branch about a single step. Write it down with the date, and write down how you got it, so the same method can be repeated after.
It does not have to be clean. Twenty applications pulled by hand is a baseline. A week of counting is a baseline. With no number at all, you have nothing to answer your boss with.
A rough number also stops you overclaiming. If the AI feature moves turnaround time from four days to three and a half, you can say so plainly. A small number holds up when someone asks how it was measured.
Build a rough working version before you write the business case
Most people arrive with the idea written up in a PowerPoint, and once they start building, the slides turn out to be wrong. The data is not where the diagram says. The step that looked simple needs a human to approve it. The model gives an answer that is fine four times out of five, which is not good enough for a credit decision and completely fine for drafting a reply to a customer.
You learn all of that by building. Not a production system. A rough version that runs, using a prompt or two and whatever tools your team already has, good enough for someone to try in front of you.
The hard part is knowing when rough is good enough to show someone. The test is simple. Can a person who has never seen it get through the task without you talking? If yes, it is ready to test. If you have to explain it, it is not ready yet.
Then watch five people use it. Five is usually enough to see the same problem come up more than once.
Fix the journey before you point AI at it
AI on a broken journey automates the failure faster. If the form asks for information the customer does not have to hand, a chatbot asking for the same information does not help. It just asks faster, and the customer still cannot answer.
So the first round of fixes usually has no AI in it. It is cutting steps, fixing the wording that confuses people, and removing the handoff where the file waits two days on someone's desk.
On a series of five-day design sprints with a large insurer, customer-facing web journeys that had previously taken six months or more were designed and tested with customers in two weeks and live in four. That is 26 weeks down to four. Completion rates on the new journeys rose 80 percent, and the steps where customers used to drop off were gone. The client measured all of that, not us.
Write-ups of design sprint weeks like that are at onoffgroup.com.
The test of AI training is the first real task after it
At some point someone will propose AI training for the whole department. A full room and a good feedback score tell you people showed up, but that's it.
The test is the first real piece of work after the room empties. Did anyone open the AI tool, and did it help? Pick the first real task you will use AI on, name it before the session, and check a week later whether it got done faster or better. Training that people are made to attend does not change how they work on Monday.
We have run workshops for Philippine banks and insurers since 2015. The workshops that change anything are the ones where people bring real work and leave with part of it done.
What to do this week
- Pick one customer journey with a known problem, and write down one number that describes it today, with the date and how you measured it.
- Choose one step in that journey where an AI feature might help, and describe it in a sentence a branch officer would understand.
- Build a rough working version of that feature, good enough for someone to use without you explaining it.
- Sit with five real users while they try it, and record what they do, not what they say about it afterwards.
- Re-measure the same number the same way, and report it honestly, including if it barely moved.
More on Diagnostics
- What the Operating Model Diagnostic looks at across 11 areas
- Custom workshop or strategy session: which one you need
- Instead of a platform project: ship one customer journey, measure it, then widen
- Everything on Diagnostics
Questions people ask
Why not start with a strategy deck?
A deck can sit safely for months without anyone finding out it is wrong. A rough working version gets used, or ignored, and you find out quickly which one it is.
What do we need before we build anything?
A baseline. If you have no before number for turnaround time or completion on the flow you are changing, you cannot answer your boss when they ask whether it worked.
What if the customer journey is already broken?
Fix the journey first. Putting AI on top of a broken flow just automates the failure faster.
Where can we see more of this work?
The workshops, the research and the things built with client teams are at onoffgroup.com.
On-Off Group trains teams, tests products with real customers, finds where a transformation has stalled and builds what gets it moving, for banks, insurers and enterprises in the Philippines, since 2015. How we help with diagnostics.
