Thinking · Newsletter
How do I use purchase data to decide which ad creative to make next?
Rank ads on purchases and purchase value per 1,000 impressions, add results up by idea, and turn the ranking into next week's brief. An eight-step method.
TL;DR. Rank your last few weeks of ads by purchases and purchase value per 1,000 impressions, not by click-through rate, then add those results up by idea (concept, hook, format) rather than by ad. Brief more of the top ideas, drop the bottom ones, and spend part of every batch on an idea you have not tried. Repeat it weekly, because one week of ranking is a guess and ten weeks is a playbook.
Why clicks are the wrong input
Click-through rate tells you an ad earned a tap. It does not tell you the tap was from someone who buys. An ad can lead the account on CTR and trail it on purchases, and if CTR picks the next brief, you will make more of that ad.
The platforms already measure the outcome. Meta's insights API reports purchase_roas, "The total return on ad spend (ROAS) from purchases", and action_values, "The total value of all conversions attributed to your ads." Google Ads defines conversion value per cost, which "estimates your return on investment." RocketShip HQ's AppLovin testing guide puts the rule plainly: "judge it on the campaign’s goal rather than click-through rate." We made the longer case in measure creative on purchases, not on the ad.
Why per impression, not per click or per ad
We read creative on purchases per 1,000 impressions and purchase value per 1,000 impressions. Purchases per click flatters an ad that earns few clicks; purchases per ad ignores how much delivery each ad got. Per impression puts every ad on the same denominator: the delivery it was given. It is also the scoreboard we report to clients: completed purchases per thousand impressions.
Then check value against cost. Purchase value per impression says whether the idea sells; ROAS (or Google's "Conv. value / cost") says whether it sells at a price you can afford.
How to use purchase data to decide your next ad creative
- Pull ad-level results for a fixed window. Impressions, spend, purchases and purchase value per ad. Leave out the last few days: Google reports its primary conversion columns "based on the time of the click, not the time of the conversion", so recent days fill in later, and Meta credits purchases inside an attribution window.
- Tag every ad by idea. At minimum: concept (the one-sentence promise), hook type, format and offer. Write the tags in the ad name or a sheet, from a fixed list, never free text.
- Add results up by tag, not by ad. Sum impressions, purchases, value and spend for every ad sharing a concept, then compute per 1,000 impressions. One idea runs as many ads, so ranking ads splits its evidence.
- Set a floor before you rank. Our rule of thumb: an idea needs at least 20 purchases across its ads before it is ranked. Below that it sits in a "not yet read" column, not at the bottom.
- Rank ideas on purchase value per 1,000 impressions, check ROAS. Read the top and the bottom of the list. The middle rarely changes a brief.
- Turn the ranking into the brief. Top ideas get new hooks, new formats and new performers, built on the same promise. Bottom ideas are retired. Repeat the same exercise for hook type and format inside the winning concept.
- Reserve part of every batch for an untested idea. Our rule of thumb is one brief in four. A ranking can only rank what you made, so without new ideas the list shrinks to one tired winner.
- Log the decision and run it again next week. Write down which idea was briefed and why. Next week's ranking tells you whether the bet paid.
Why you rank ideas, not ads
The networks will not do this grouping for you. Meta's documentation for dynamic creative says "we currently show only creative-asset level breakdowns such as metrics by image, title, body, video." Google's Performance Max asset report goes further and warns that "Conversions aren’t split across multiple components of an advertisement": an ad with a headline, description and image that drives one conversion credits all three assets with it. Asset reports tell you which file was present, not which idea sold.
Meta's own guidance points the same way. Its April 2025 post on creative diversification describes making a wide range of ad creatives "with different themes, messages, and visuals to cater to diverse audience segments." If the system is choosing between themes and messages, your report should be grouped by theme and message too.
Segwise, a creative analytics vendor, describes what the tags buy you: "You stop asking 'which ad performed best?' and start asking 'what do our best ads have in common?'"
What the ranking tells you to make
| What the ranking shows | What to brief next |
|---|---|
| One concept leads on value per 1,000 impressions and ROAS | New hooks, formats and performers on that concept |
| A concept sells (high value per impression) but ROAS is weak | Same concept, cheaper to deliver: shorter cuts, other placements, test the offer |
| High CTR, low purchases per impression | Fix the promise or the destination, not the hook. The ad attracts the wrong click |
| A format wins across several concepts | Make it the default format for new concepts |
| An idea has fewer than 20 purchases | Keep it running. Do not brief from it yet |
| Every top idea is the same concept | Brief a new concept now, before that one tires |
The third row is where most teams lose money: a hook that wins on clicks gets copied while purchases stay flat. If the ad and the page disagree, fix the page before you rebrief the hook.
Where to find untested ideas
Your own library is the fastest source. Group your top twenty ads by spend by the promise each one makes, and the concentration shows within an hour, as we wrote in the most useful creative research is your own ad library. The ideas missing from that list are where the "one in four" slot goes.
The weekly loop
The ranking only works if it runs on a schedule. Our loop is brief, make, launch, read, log, every week. It is the same five-stage system we lay out in creative operations, defined. Read on purchases, log the decision, and let next week's ranking grade it. When you want to know which part of a winning ad did the work, the method is in which part of my ad drives installs or purchases.
Flixr runs this loop for apps, games and commercial brands on AppLovin, Meta and TikTok, with read-only access to the client's results so we are judged on the same numbers as their team.
Questions people ask
Should I use ROAS or purchases per 1,000 impressions to pick creative? Use both, for different jobs. Purchase value per 1,000 impressions says whether an idea sells when shown; ROAS says whether it sells at a cost you can carry. An idea that wins the first and loses the second needs a cheaper version, not retirement.
How long should I wait before reading purchase data? Long enough for purchases to arrive. Google reports primary conversions against the click date, so the most recent days fill in later. Leave the latest few days out of every read. Our floor of 20 purchases per idea matters more than the calendar.
Can I just use Meta's or Google's asset reports? As an input, yes. As the ranking, no: Meta breaks dynamic creative down by image, title, body and video, and Google's Performance Max asset report credits every asset in a converting ad with the full conversion. Neither tells you which concept sold, so tag and add up by idea yourself.
What tags should I start with? Concept, hook type, format and offer. Add more only when a decision needs them, and keep each tag to a fixed list so results add up.
Does this work for app installs as well as purchases? The method is the same with a different outcome. Rank on the event you are paid on, whether that is installs, trials or purchases, per 1,000 impressions.
Sources: Meta for Developers, Ad Account Insights; Meta for Developers, Insights for dynamic creative (asset feed); Meta for Business, The Creative Advantage: Unlocking the Power of Diversification with Meta Andromeda; Google Ads Help, Conversion value per cost; Google Ads Help, Understand your conversion tracking data; Google Ads Help, Conversion reporting by asset in Performance Max campaigns; RocketShip HQ, How to test ad creatives on AppLovin; Segwise, Creative tagging.