Customer research for bank onboarding and account opening teams

On-Off Group

October 2026

Customer research for bank onboarding means watching real applicants try to open an account, recording where they stop, and taking a before number on drop-off and turnaround time so you can prove whether your fix worked.

This is for whoever owns account opening at a Philippine bank and keeps getting asked why so many applications never finish.

You have probably done journey mapping already. You know roughly what usability testing costs. What you cannot do is answer your boss when the question is whether the last round of fixes worked, because there is no before number to point at.

Watching real people open an account does two things. It shows you where applicants stop, and it gives you a number you can measure again after the fix.

Why nobody can say which fix worked

The dashboard says applications started, applications completed. Between those two numbers is a gap nobody can explain. Somebody in the meeting says it is the selfie step. Somebody else says it is the ID upload. Both have been saying that for a year.

Meanwhile branch staff have their own version. They know which field customers always fill in wrong, and they have a workaround for it that never made it into any document.

So the fix gets chosen by whoever argued hardest. It ships. Completion moves a bit, or does not, and nobody can tell whether it was the fix, the payday week, or a campaign that ran at the same time.

The thing nobody did was write down what the flow was doing before the change.

Take the before number first, even if it is ugly

Two numbers are enough to start. Completion rate by step, so you can see where people leave. And turnaround time from submitted application to opened account, measured end to end, including the days it sits with whoever checks documents.

Most bank teams can get the first number out of analytics in a week. The second one usually needs someone to pull thirty recent applications by hand and count the days. That is fine. Thirty applications counted honestly beats a system report nobody trusts.

Write both numbers on one page with the date and the period they cover. Send it to your boss before you change anything. It feels like admitting the flow is bad. It also means that when the pilot ends you can show what moved.

If the number turns out to be small, say so. A drop-off of four per cent at the ID step is not a project.

Watch five real people open an account

Not staff. Not people from the office. Five people who match the customers you are actually trying to win, on their own phones, opening a real account or getting as close as compliance allows.

A survey will not tell you any of this. Someone photographs their ID on a dark table and the upload keeps failing, so they give up and tell you later the app was fine. Someone reads the middle name field differently from how the system reads it. Someone gets to the funding step, sees no option they use, and closes the app.

Five is enough. You are looking for the places where several people stall, and those show up fast.

Record it. A clip of a customer stuck on one screen, unable to get past it, ends the argument about the selfie step. You can do one round of this, or keep a few sessions running every month. Both options are written up at onoffgroup.com.

Include branch staff and the people checking documents

Onboarding does not end when the applicant taps submit. Half the turnaround time usually sits after that, with whoever reviews documents and whoever handles the exceptions.

So watch two more things. A branch officer completing an application on a customer's behalf, on the actual branch machine. And the review queue, including what happens when a document comes back blurred or a name does not match.

Watching the review queue is where you find the loops. An application bounces back to the customer, the customer does not respond for days, and it ages in a report as if the customer went quiet. The report calls that customer drop-off. The cause is a rejection message that does not say what to fix.

When a bank team watches this together, the fix list comes out different. Less about app screens, more about wording and about what counts as an acceptable document. Getting those people in the same room for a day is the cheapest part of the whole exercise, and there is training for bank teams that does exactly that.

Fix the steps first, then look at AI

There is pressure to put AI somewhere in onboarding. Document checking, a chat assistant, automated review.

If the flow already sends people away at step four, an assistant just gets them to step four faster. If the rejection message does not tell customers what was wrong with their ID photo, automating the rejection produces more of them, sooner. We have written about that pattern in Fix the customer journey before you put AI on it.

Do it in this order. Find the two steps people fail on. Fix those. Measure again against your before number. Then look at where the remaining time goes, and if the answer is document review sitting in a queue overnight, that is a sensible place for AI, with a number attached to it.

A large insurer we worked with redesigned customer-facing web journeys this way, in short sprints with customers testing each version. Journeys that had taken six months or more were designed and tested in two weeks and live in four. Completion rates on the new journeys rose 80 per cent, and the old drop-off points went away. The client measured all of this, which they could only do because the numbers were taken before the work started.

What to do this week

Pull completion rate by step for the last full month and write it on one page with the date.
Count turnaround time by hand on the last thirty applications, from submit to account open.
Book five people who match your target customer and watch them try to open an account on their own phone.
Sit with one branch officer and one document reviewer, and write down every workaround they mention.
Pick the two steps with the biggest drop-off and agree what you will change, with the before number written next to it.

Questions people ask

How many people do we need to watch?
Five is enough to find the points where applications stall. You are looking for the places several people get stuck, not a statistically representative sample.

What should we measure before we change anything?
Completion rate by step, and turnaround time from submitted application to opened account. Both taken over a fixed period, written down and dated, before any fix goes live.

Should we add AI to the onboarding flow first?
No. If the form is unclear or documents keep bouncing back, AI makes the same failure happen faster. Fix the steps people fail on, then look at where AI helps.

Does this work for branch-assisted applications too?
Yes, and it usually needs to. Watch branch staff complete an application on a customer's behalf, and watch whoever reviews the documents afterwards.

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 customer research.