Call conversions for local services: what duration counts as a real lead?

The call duration threshold for local service accounts gets argued a lot, and I think both extremes cause problems. Short calls are full of wrong numbers, hang-ups, and people asking for opening hours. Say a plumber gets 40 calls a month and half run under 20 seconds. If all of those count as conversions, Smart Bidding learns to chase whoever dials fastest instead of whoever books a job.

But a 60 second cutoff isn't safe either, because it drops real emergency calls. A customer with a burst pipe often says three words, confirms the address, and hangs up.

My lean is a longer threshold, checked against booked jobs in the CRM, though it depends on volume. With thin conversion counts, one bad threshold can swing the whole bid model. Where do you set it, and do you check it against actual jobs?

I'd stop hunting for the perfect number, because any duration cutoff is going to misfire on those burst-pipe calls you mentioned. For a plumber I'd split it. Keep a call conversion with a low threshold as observation only, and import booked jobs back as offline conversions tied to the original call. That way Smart Bidding optimizes toward actual jobs, and the duration rule only decides what shows up in reporting. At 40 calls a month the job-level signal will be thin, but a thin signal that's right beats a thick one that's wrong.

Iris, the split makes sense, but the first thing I'd check is whether you can tie each booked job back to the call that produced it. Offline conversions only help if that match exists. For phone leads, that usually means a call ID or a caller number the CRM actually stores. If the office writes jobs down from memory or only records the address, the import either drops the job or pins it on the wrong click, and a thin signal gets thinner.

Volume is the other problem. Smart Bidding needs a decent number of conversions before it settles, so with only a handful of booked jobs a month I'd expect the model to stay noisy no matter what threshold you pick. In that case something simpler, like maximize clicks with tight location targeting, often does less damage than a tCPA learning from a few jobs.

How many of those 40 calls end in a booked job, and does the CRM record which call each job came from?

Agreed on the split. What makes it work is the primary/secondary setting. Set the low-threshold call action to secondary in Google Ads and it still shows up in reporting, but it stops feeding Smart Bidding, so the model only learns from the imported job conversion.

The catch is that the import only matches if the original call is still inside the conversion window when the job gets booked. If you're a plumber and jobs usually close a week or two after the call, check that lag before you trust the match rate.

Depends on how much your job values vary. If a plumber's jobs run from a $90 unblock to a $1,500 repipe, importing a plain count teaches Smart Bidding that a cheap callout is the same win as a big job, so it'll chase whatever's easiest to book. Pass the job value in the offline import and bid on value, and the model leans toward the work that actually pays. If nearly every job is about the same size, a simple count is fine and the setup is easier to maintain.

Depends on where the calls come from. Ad calls tend to come from people still comparing options, while calls from a listing or a repeat customer are often short because they've already decided. A single cutoff across both will misjudge that second group.

Lag check is right, and there's another leak: a lot of emergency jobs don't close on the original call. Someone dials at 11pm, hits voicemail, and the office calls back the next morning, sometimes from a different line. If the CRM only stores the callback number, the import can't match the job to the ad click, so the most urgent work looks like it never came from ads. Have the office record the original inbound number whenever a callback closes a job.

Iris and Theo's split makes sense to me. I'd go one step further and pick the threshold from your own data instead of arguing about 60 seconds. Export a few months of calls with duration, then match them to CRM outcomes by caller number. Bucket them (under 20 seconds, 20 to 60, 1 to 3 minutes, over 3) and count how many in each bucket became booked jobs.

For a plumber, the under-20 bucket will be mostly dead air and wrong numbers, but the burst-pipe calls sit in there too, so don't set the cutoff right where the junk stops. Look for the point where the booked-job rate levels off, and treat anything below that as observation only.

At 40 calls a month you'll need several months before the buckets mean much, but that still beats guessing.

Honestly, it depends on what the ad promised before the phone rang. An ad saying "Burst pipe? We're out within the hour" pulls in people already in crisis, who say three words and hang up. That short call is exactly the one you want to count. An ad saying "Free quotes, see our prices" pulls in people comparing options, so they talk longer and book less often. Same plumber, same budget, very different call profiles.

That's why I'd look at the copy behind each bucket of calls before picking a cutoff. If emergency and quote-led ads run in the same campaign, the blended data hides both patterns. Splitting them by campaign or ad group and setting a threshold for each is usually cleaner than forcing one number across both.

Splitting by ad group makes sense, but there's a source problem that can skew your buckets. Calls from your Google Business Profile listing don't carry a click ID, so they never show up as Google Ads calls and no offline import can tie them to an ad. If your call export mixes them in, the listing calls look like ad calls that never booked, and they drag the booked-job rate down for the wrong reason. Filter them out before you bucket by duration.

For the emergency ad group, I'd also give it its own call number on the ad. That way the call log only holds calls from that ad, and the under-20-second bucket isn't muddied by calls from elsewhere on the site.

Rule of thumb: if fewer than about half your booked jobs match back to an ad call, keep the call action out of bidding entirely. The matched calls skew toward people who phoned straight from the ad, so the model learns from the easy cases and misses the callbacks.