Most retailers have Always On in name and seasonal campaigns in practice. Budget gets reallocated in season, the algorithm loses its model, and re-entry after the season always costs more than it should.
The pattern is predictable. In September the marketing team assigns a baseline budget to "always-on campaigns" and plans the year's activations: Buen Fin in November, Christmas in December, Three Kings in January. The plan looks good on paper.
November arrives. The Buen Fin budget is 5x the normal one. Someone decides it makes sense to take part of the Always On budget to amplify the season — "everyone converts in season anyway". The baseline campaigns are paused or cut to a minimum. Buen Fin is a success. The ROAS is celebrated.
December begins with the system rebuilding from scratch. The algorithm lost its model. The warm audience built up over previous months has cooled. Re-entry costs 40% more than keeping the base running would have. Nobody measures it, because nobody compares against the alternative scenario.
An ad set that has run 90 straight days has something a new one doesn't: a refined audience model. The algorithm learned which user profiles convert for that specific product, at that specific CPM, with that signal architecture. That learning isn't saved in any file — it lives in the ad set's history and in how Meta's delivery system distributes impressions.
When you pause an ad set for more than 7 days, the algorithm treats it as new when you switch it back on. It doesn't resume where you left off — it starts the learning cycle over. In pharmacy retail that cycle takes 2 to 3 weeks of spend to reach the previous efficiency. The cost is invisible in the seasonal report and very real in the December budget.
Always On is running, the algorithm has its model, frequency is under control, CAC is in range. The system is ready to be amplified.
The team makes a short-term call: redirect baseline budget into the activation. Seasonal ROAS rises. The base loses momentum. The accumulated warm audience starts to cool.
When the baseline campaigns are switched back on, the system enters the learning phase. CPM is higher (post-season inventory is expensive), the audience model is at zero, the creative may carry accumulated fatigue. The team asks why it "doesn't work like it did before the season".
The rule that breaks the cycle: The Always On budget is not reallocatable during peak season. It is protected. Seasonal activations get their own incremental budget — they do not compete with the baseline. If there is no incremental budget for the season, the season is smaller. The baseline is not touched.
The second common mistake in retail is treating Always On creative as static. The team produces a batch of assets for the "always running" campaigns, uploads them, and lets them run until CPL rises or someone complains.
In a retail operation with 20-30 live SKUs and diverse audiences, creative has a useful life of 4-6 weeks before fatigue starts eating into CTR. In high-volume seasons that cycle shortens to 2-3 weeks, because frequency climbs faster.
Real Always On has a creative calendar tied to signal, not to dates. When CTR drops 15% over 3 consecutive days or frequency passes 2.8 in the main segment, creative rotates. Not because "four weeks have gone by" — because the data says so.
If CAC rises every time a seasonal activation ends, the baseline system is being sacrificed. I diagnose the structure and propose how to protect it without softening the impact of the seasons.
References
Primary documentation for the platforms cited on this page:
Case figures refer to the periods and accounts described in each one; they are not promises of replicable results.
Por Cesar Baz — Performance marketing, data and AI consultant
Publicado el · Actualizado el