Campaign Case Study
When Does Usage Move After a Launch Campaign?
A new large language model launched on OpenRouter while Tutti ran a creator content campaign alongside it. We put the campaign's effective views and the model's public request counts on one timeline and watched how the two curves talk to each other. Client and model are anonymized under our confidentiality terms.
The most honest test of a campaign is putting the exposure we delivered next to the client's own business curve. In this case that curve is public: anyone can look up the model's daily requests on OpenRouter. Every exposure figure in this report uses the effective-views standard — posts from banned accounts removed, detected injected traffic stripped, views without real engagement discounted on a curve — the same numbers used to settle with creators.
- The model's usage acceleration came 1–2 days after the campaign's exposure peak (Spearman ρ ≈ 0.86) — the timeline supports "the campaign drove trial."
- The campaign delivered 1.05M effective views against a 1M target (105%); daily requests accelerated from ~1.7M to ~3M during the campaign window.
- The first wave after launch and the later ~3M/day plateau were not the campaign's doing — we don't count them as campaign effect.
- For brands: give evaluation at least 2–3 days after exposure; concentrated delivery makes the inflection visible; ask your vendor for effective views and explicit attribution boundaries.
Exposure First, Usage After
The campaign's exposure peaked on day +3 after launch; the model's requests peaked on day +7. The only acceleration in requests (+3 → +7, +1.34M/day) landed 1–2 days after the campaign concentrated ~0.91M effective views on days +2 to +5.
OpenRouter values are manual readings of its public model chart, ±5% tolerance; the partial final day is excluded.
| Days since launch | OpenRouter requests (M) | New posts | Daily effective views | Cumulative effective (M) |
|---|---|---|---|---|
| Launch | 0.10 | 0 | 0 | 0.00 |
| +1 | 1.63 | 2 | 4,869 | 0.00 |
| +2 | 1.68 | 10 | 139,743 | 0.15 |
| +3 · exposure peak | 1.88 | 58 | 349,144 | 0.49 |
| +4 | 2.42 | 12 | 260,149 | 0.75 |
| +5 | 3.00 | 11 | 159,890 | 0.91 |
| +6 | 2.97 | 1 | 54,154 | 0.97 |
| +7 · request peak | 3.22 | 1 | 34,289 | 1.00 |
| +8 | 2.93 | 0 | 15,473 | 1.02 |
Delivery counts creator originals only — boost replies and quotes are amplification, not double-counted delivery. The table's day-by-day observation sums to 1.02M; final settlement confirmed 1.05M (settlement ran on slightly fresher snapshots).
Shift Exposure Two Days Forward and the Ranks Line Up
Within the 9-day window, same-day exposure and requests are nearly uncorrelated; shift the campaign's exposure forward 1–2 days and rank correlation rises to 0.86. Cumulative exposure explains requests better than same-day exposure (r 0.90).
ρ is rank correlation (Spearman): it only asks whether the rankings match — were the biggest exposure days also the biggest request-growth days? 1 means the rankings match perfectly, 0 means no relationship; it isn't inflated by a single extreme day, which makes it the right directional read for a short window. r is ordinary linear correlation (Pearson).
The reading: consistent with a "see the content → try it a day or two later" path. Cumulative exposure explaining usage better means content builds awareness that compounds — it isn't same-day click-through.
What Was the Campaign, and What Wasn't
Writing off what we can't claim is what makes the rest stand. The defensible statement: during the campaign window, the model's daily requests accelerated from ~1.7M to ~3M, and the timing of that acceleration matches the timing of exposure delivery.
- People try 1–2 days after seeing the content — lagged correlation climbs from 0.32 to 0.52 / 0.86, a clear conversion path.
- The only request acceleration (+3 → +7) came right after the campaign concentrated ~0.91M effective views on days +2 to +5.
- Cumulative exposure explains usage better than daily exposure — content builds awareness that outlasts the posting window.
- The first wave the day after launch wasn't ours: the campaign had barely started. That was launch buzz plus the platform's limited-time free tier.
- Neither was the later plateau: requests held at ~3M/day when the campaign's new exposure had already faded — retention is the product's doing.
- Nine days of observation, global requests vs a Chinese-language social campaign — correlation isn't full causation, so we take the most conservative reading.
Three Takeaways for Brands
The three things worth carrying out of this case.
- 01Effects lag 1–2 days — don't judge a campaign on day-one data
Same-day exposure and usage are nearly uncorrelated; two days later the ranks line up. Kill a campaign on day two and you stop right before the conversion arrives. Give evaluation at least 2–3 days after exposure.
- 02Content builds awareness — concentrated delivery makes the inflection visible
Cumulative exposure explains usage better than daily exposure (r 0.90): what works is accumulated awareness, not same-day clicks. The same budget concentrated into a few days is far more likely to leave a recognizable inflection in the usage curve than a slow drip.
- 03Ask for effective views — and for the boundaries
Reported exposure should strip banned accounts and injected traffic and discount views without real engagement — the number you're shown should be the number that gets paid. Just as important: your vendor should volunteer which traffic was not the campaign's doing. Data that dares to report net figures and draw boundaries is data you can make decisions with.
Client and model are anonymized under confidentiality terms; the time axis is anchored to the model's launch day; the case ran in summer 2026. Effective views share the creator-settlement standard: posts from banned accounts removed, detected injected traffic stripped, views without real engagement discounted — the number we report to brands is the number we actually pay on.
Make your next launch visible in the usage curve
Tutti connects vetted 𝕏 creators with brands; delivery is counted in effective views and real engagement only.