Metrics
Fill rate is a feature, not a target
15 July 2026 · 5 min read · Monetzly team
Fill rate — the share of opportunities that get an ad — is a number ad networks are trained to push toward 100%. In display, that instinct is roughly correct. In conversation, it is how you destroy the channel.
The argument for keeping it low is not aesthetic. It is that fill rate and rate per placement are coupled, and the coupling runs the wrong way.
Why the two numbers fight each other
A sponsored mention inside an answer is worth something because users read the answer. Read rate is a function of whether previous answers were worth reading. Push a placement into every turn, including the ones where nothing genuinely fits, and users start skimming past the sponsored parts — then past the answers themselves.
Once that happens, click-through falls, advertisers see worse performance, and the rate they will pay for your inventory falls with it. Higher fill rate, lower value per placement, and the trade is not close.
What a healthy distribution looks like
In an assistant with real commercial intent, only a minority of turns contain a need a campaign can legitimately serve. The rest are clarifications, follow-ups, chit-chat, and questions with no product answer at all. A system matching most of those turns is not finding intent; it is manufacturing it.
- No match is the default outcome, and the response passes through unmodified.
- Placement frequency is capped per session, independent of how many turns would technically qualify.
- The threshold is a relevance judgement, not a keyword hit.
Read fill rate as a diagnostic instead
The metric is still useful — just not as a goal. Sudden movements are the signal.
| What you see | Likely cause | What to check |
|---|---|---|
| Fill rate near zero | Category blocks too broad, or the app's subject matter carries little commercial intent | Blocked categories; whether an ad model fits this app at all |
| Fill rate climbing without a traffic change | Threshold drift or a new broad-match campaign | Sample recent placements by hand and judge relevance |
| High fill, falling click-through | Matches are technically valid but contextually weak | Tighten the threshold; expect revenue per placement to recover |
The habit worth building is periodically reading real placements in real conversations. Aggregate metrics will not tell you that an ad was technically relevant and obviously wrong to a human.
Related
Monetize the conversations you already have
Wrap your existing response stream with the Monetzly SDK and earn on sessions that were never going to convert to a subscription.