So we moved the money to creators and to the events that actually cost something. 27% of the volume beat the other 73%, and ten months later the app had 1,686 signed accounts.
No measurement partner, Firebase events only. On the platforms that means you can see an app open and nothing after it. Most agencies treat that as a reason to start spending and fix the tracking later.
We went the other way and read the only honest signal available: what people did inside the product. Internal data, session by session, all the way down the funnel.
It showed the drop was not at acquisition. People were arriving and leaving during onboarding, before the app ever got to show them why it existed. Pouring media on that would have bought an audience the product was not ready to keep.
An app that has not solved onboarding does not have a media problem. It has a product launch problem.
So October and November went into the fix. Onboarding restructured around the moment users actually converted, then the measurement stack, the martech tooling and the off app instrumentation built underneath it. Roughly four times the investment into tools, tracking and structure before any meaningful ad money moved.
December was the first month we let it run: Google scaled, Meta launched, targeting deliberately unclever. 9,097 installs, arriving into a product that finally knew what to do with them.
January went into a creative testing framework rather than a budget increase. February added supply, because one platform absorbing a seven fold increase in eight weeks stops being a scale problem and becomes a frequency problem. Apple Search Ads, Telegram and affiliate partners came online together.
It worked. It also produced the most expensive lesson of the year, which arrives two chapters from here.
An install is the easiest number in marketing to buy and the hardest to defend.
The first measurement partner had been picked because it was affordable. That stopped being the deciding factor the moment the numbers had to survive an argument. We ripped it out and replaced it with one that gave us unfiltered, neutral event data, then moved Meta onto server to server events so the platforms optimised against what the app recorded rather than what an SDK guessed.
Six events went live. For the first time we could tell the difference between someone who downloaded the app and someone who used it.
Cheap data is not data. It is a story you paid a discount for.
One honest note. Every event level figure on this page starts on 14 February 2026, because that is the first day the app could describe itself properly. Everything before it is platform reported.
No, we are not nuts. This was on purpose.
The install objective buys the cheapest available download, and the cheapest available download belongs to the person least likely to open the app twice. So over half the budget moved off installs and onto signup, report viewed and report asked. Creator campaigns took more than 40% of Meta budget across both months.
Both accounts were then cleaned out for creative fatigue, with one rule: acquisition cost was not allowed to break while we did it. Snapchat and X ads joined Telegram as test channels.
Use the switch. Same audience, same creative pool, same country. The only thing changing is what the campaign was told to buy.
Meta campaigns, 14 Feb to 9 Aug 2026. Rates are events divided by installs on the same campaigns.
The client was extending from research into a full Demat account, and that work ran into the compliance engineering required to stay inside SEBI's guidelines. Until it cleared there was no honest reason to keep acquiring at full budget. Installs fell from 66,451 in February to 9,034 in June.
We have kept it in, because it is where the account earned its next six months. The pause produced two new events, e-sign and a refined signup, plus a client service team to handle leads that were finally worth handling.
If an agency has never shown you a flat quarter, you are looking at an edited one.
Budgets came back split. Google took the bottom of the funnel and the e-sign event, Meta went back to the top and fed it signups. Seventy to thirty.
The e-sign campaign on Google turned 5,328 installs into 1,343 signed accounts at a 4.19% click through rate, against an account average of 1.25%. Above it sat a YouTube campaign running 2,953,529 impressions at 0.18%. As a response line that is humiliating. It was never a response line. It was the retargeting pool the e-sign campaign closed.
Ten months to stop counting downloads and start counting signatures.
Weeks beginning 25 May to 3 Aug 2026. Grey is before the e-sign campaign launched.
Google and Meta wanted two completely different things, and pretending otherwise is how most app accounts waste a year.
On Google, roughly 78% of budget went into non creative assets. Not a slight on the work, just what a search led category rewards. The exception was a video only campaign that needed special approval from the Google team, which we used to buy views at the top and then retarget the same audience with e-sign at the bottom.
On Meta the opposite was true. The client's in house production was fast and their creator sourcing was better than ours, so our job was never the camera. It was the test matrix, the objective sitting underneath each asset, and knowing when to move money behind a result instead of a preference.
A clip the whole internet was watching that week, pointed at a stock report. Shelf life measured in days, which is exactly why the matrix mattered more than any single asset.
A trend nobody in finance would touch. It ran, it worked, and it dated fast. Both of those were expected.
The character everyone was quoting that month. Recognition does the first two seconds so the product only has to earn the next five.
The one window a year when the entire target audience is looking at the same thing at the same time. We treated it as inventory, not as a theme.
Built inside the formats Gen Z already scrolls, not adapted into them. The difference between the two is the difference between 8.9% and 24.4%.
Someone the viewer already followed, talking about money the way they normally talk. The highest report ask rate in the account came from here: 51.1%.
We were outperformed by people with phones. So we put them in the media plan.
Every row goes the same way, and not by a little. The best creator campaign converted 51.1% of its installs into a stock report question. The best brand made campaign managed 22.2%.
The second thing it solved was quieter. The client wanted people new to investing, only those, which is an audience no platform sells because it is harder to find and slower to convert. We built it from proxy signals, then ran it alongside the audiences we already trusted so the pairing held the brief and pulled the cost back down.
Creators got there from the other direction. Someone who has never bought a share does not respond to a brokerage. They respond to a person they already trust talking about money.
Cumulative lifetime value events across every attributed source. Not strictly sequential: some users request a portfolio without asking for a report first, and e-sign only existed from June.
The first number is the one every deck leads with. The last one is the only one the business can bank.
51.1% of its installs asked a stock question, the highest rate in the account, on 3,725 installs. It has never carried a top three budget. That is the easiest money on the table.
Weekly signatures went from six to 331 in a month, then settled near 250. Strong start, not a trend. We are optimising toward an event we have watched for six weeks.
Snapchat and X ads launched into the quarter that got interrupted. Both deserve a clean window against the new event structure, not a verdict based on when they happened to start.