The second OTP looks like a scam. Explaining it first lifted completion by a fifth.
A journey most people abandon
India’s Account Aggregator framework lets someone share their financial data with a lender in a few taps instead of digging out six months of bank statements. It is a genuinely good piece of public infrastructure, and it has a problem everyone in the industry knows about: most journeys do not finish.
Across the ecosystem, roughly a quarter of consent requests raised are actually fulfilled. Our own completion sat nearer thirty percent, which was good for the market and still meant seven out of ten people walked away part-way through.
They were not walking away because the product was broken. They were walking away because of a moment in the middle of it.
The moment
The journey requires two one-time passwords. One links the bank account. A second, later, approves the sharing of the consent. Both are necessary and both are correct.
Now consider who is receiving them. This is a country where every bank SMS, every regulator campaign and every warning forwarded by a relative teaches one rule above all others: be suspicious when someone asks you for an OTP. And be far more suspicious the second time.
The product was behaving exactly as designed, and the design looked exactly like a scam.
That second OTP is the point where a customer on a phone call goes quiet, decides something is wrong, and hangs up. Nothing in the interface can fix it, because the interface is not what they distrust. They distrust the situation.
Where the idea came from
Nobody asked me to work on this. I noticed it from the side.
The product team kept getting the same questions from client operators, over and over, and was spending real time answering them individually. I had worked with that team on product decks, so I knew the journey well enough to see the pattern — and I knew from those conversations that when an operator walks a customer through the flow properly, it completes far more often.
So the question was not how to fix the product. It was whether the thing a good operator says on a call could be packaged so that every operator said it, without anyone from product having to be on the line.
I asked the product team whether we could build that as self-serve video. They said yes.
What I made
Five walkthrough videos in a month, aimed at operators and product staff at the banks and NBFCs consuming the data — the people who sit on the phone with a customer and talk them through it.
They are not product demos. Each one is built around the questions that actually reach the product team, and the two-OTP explanation is in there deliberately: you are going to receive a second code in a moment, here is why, it is not a mistake and it is not fraud. An operator who says that before the OTP arrives turns the scariest moment of the journey into proof that they know what they are doing.
I recorded them, wrote the scripts, produced the voiceover, cut the timing, added the highlights and zooms, kept the branding consistent, ran several rounds with product on which doubts came up most, and then distributed them to clients over email and WhatsApp groups.
Where the AI actually helped
Five videos in a month is not a normal production rate for one person doing every job. It was possible because I used an AI video tool, and it is worth being precise about what it did and did not do.
- Generated a first-pass script from my screen recording
- Laid down a rough timing sync
- Handled the mechanical assembly
- Made each new revision cheap
- Decided what the video had to say at all
- Rewrote the script it produced
- Fixed the timing, which was never quite right
- Added highlights, zooms and on-screen text
- Held the branding together
- Chose which doubts were worth answering
A video that would have taken me a week took a day. Each subsequent revision took under two hours instead of most of another day.
That is the honest shape of AI in marketing work right now. It did not know that two OTPs look like fraud in India, or that this was the sentence worth building a video around. It removed the week of assembly standing between having that idea and shipping it — which is not a small thing, because the reason good enablement content does not exist is almost never that nobody knew what to say.
Then the requests came in another language
Business teams came back asking whether the videos could exist in the languages their operators and customers actually speak. Which, once you think about what the videos are for, is obviously right. The whole job is convincing a nervous person that a second OTP is safe. Reassurance does not travel well in someone’s third language.
There was no budget for localisation, no agency and no vendor. So I made the Hindi and Marathi versions myself, and then asked around the office.
Five of these exist because colleagues sat down and recorded them out of goodwill. Nobody was paid, and nobody was assigned.
I am prouder of that row than of anything else in this project. It is the part that could not have been bought.
What happened
To be exact: twenty percent is a relative lift on the base, not twenty percentage points, and it is the top of the range rather than the average. The comparison is banks with trained, video-equipped operators against banks without them — not a before-and-after at a single client.
In plain terms, that moves completion from about thirty percent to about thirty-six. Six more journeys finishing in every hundred started, in a market where the average is twenty-five.
It is not a transformation. It is a fifth more of the thing the business was already trying to do, achieved by five videos and some borrowed afternoons, and it compounds every month it keeps running.
Why a video worked where a document had not
There was already documentation. Operators still asked the product team.
My read is that this is about who has to admit what. Picture a new operator a few weeks into a job at a bank they were proud to join. They do not fully understand why the customer gets a second code. Asking that out loud — of a manager, of a vendor’s product team — means admitting they do not know something that sounds basic.
A video their manager has told them to watch costs them nothing to admit. They learn the thing privately, and they arrive at the call already able to explain it.
Enablement content does not only transfer information. It gives people a way to acquire it without anyone watching.
What I would do differently
Instrument it. The videos went out over email and WhatsApp, which is exactly how the audience wanted to receive them and exactly how you learn nothing. I have no view counts, no completion data, no way of knowing which of the five did the work. That was a free thing to set up and I did not do it.
Start from the tickets, not from the journey. The first cut followed the product flow in order, because that is the intuitive way to explain something. The version that worked was organised around what people were confused by. Those turn out to be different documents, and only one of them is useful.
Go looking for the next one. This started because I happened to notice a product team answering the same question repeatedly. That signal exists in every company and nobody owns watching for it. A recurring support question is a marketing brief that has already been validated by the market.