Call conversions: count every call over 30 seconds or only the ones that book?

Say a plumber gets 40 calls a week from Search ads, but the owner reckons only 12 turn into jobs. Most setups count any call over 30 seconds as a conversion, which feels generous to me. I'd make 60 seconds the primary conversion and keep the shorter calls as a secondary signal, since short calls are often wrong numbers, price shoppers, or someone who dialed the wrong branch. The catch is volume. At 12 real jobs a week Smart Bidding has very little to learn from, so a strict threshold can starve the campaign. I'm also not sure where the line belongs for a trade with long quotes versus one doing quick emergency jobs. Has anyone compared duration cutoffs against booked-job data?

Smart Bidding needs volume more than a perfect definition, so I'd keep the 30-second call as the conversion it optimizes on and fix the value side instead. If the owner's booked jobs get uploaded back as offline conversions with real values, the bidder learns that some calls are worth far more than others, and the short-call noise stops mattering much. It also handles your trade question better than a single cutoff can. A quote-heavy job and a quick emergency call can carry different values, so the signal reflects the actual business instead of a guess based on duration.

Does the owner's booking system record which call led to each job, or is that matching done by hand?

I'd drop the 30-second call as the optimization target. Offline values only work if every call can be tied to a booked job, and a long quote can easily close two weeks later, so the uploads show up late and in small numbers. Until that loop is reliable, the bidder is mostly learning from wrong numbers.

Agreed on the matching problem, but I wouldn't throw out the 30-second call yet. A decent rule of thumb is to optimize on the cheapest event that still has enough volume to learn from, and only move to booked jobs once the owner can match most of them back to a call. Without that match rate the bidder is learning from noise either way. Two or three weeks of matching data will tell you more than arguing over where the cutoff belongs.