Screening Junk Calls Without Blocking a Real Customer

Every trades shop has the same two complaints about its phone, and they pull in opposite directions. The line rings all day with people selling merchant services and SEO. And the one call you cannot afford to miss came from a number nobody recognised. Any filter aggressive enough to fix the first complaint will eventually cause the second.

Quick answer: Do not screen on a carrier spam score, because a real customer calling from a cheap mobile plan can carry one. Screen on your own history instead. Let the first call from any number through, review it, and mark the ones that were not real callers. A blocklist built from those verdicts is accurate for your shop and reversible the moment you get one wrong.

Why the obvious approach does not work

The instinct is to buy a spam score. A carrier or a lookup service rates the incoming number, and anything above a threshold gets dropped.

Two problems. The first is availability: the Twilio Lookup add-ons that once carried a reputation score are gone, so for a lot of stacks that data simply is not there to buy any more.

The second problem is worse, and it would matter even if the data existed. Reputation scores are built from complaint patterns across the whole phone network. Your customer calling about a dead furnace from a prepaid handset, or from a number recycled six months ago by someone who did run a call centre, looks like a risk to that model. A homeowner in a bad situation is exactly the person least likely to try twice.

A false positive on a marketing call costs you nothing. A false positive on a no-heat call in February costs you a customer and probably their neighbour too.

Screen on your own judgement instead

The signal you actually trust already exists: your own opinion of the calls you have already had.

In Patchment (that's us) the blocklist is not a list you maintain. It is derived from the verdicts you leave when you review calls. When you mark a call as not a real caller, that number stops getting through. Three useful things follow from deriving it rather than storing it.

It maintains itself. You are already reviewing calls, so the filter improves as a side effect of work you were doing anyway. Nobody has to remember to add numbers to a screen they never open.

There is no second copy to drift. A stored blocklist is a separate piece of state that slowly stops matching what you believe. A derived one cannot disagree with your verdicts, because it is only ever a reading of them.

Getting it wrong is cheap to undo. If you mis-flag a real customer, you correct the verdict and the block lifts immediately. With a stored list you would need to find the entry and remove it, which means an extra screen and a support conversation at exactly the wrong moment.

The rule that protects real customers

The first call from any number always gets through. Screening only stops the second one onward.

This is the whole safety property, and it is worth being blunt about why. A new customer is, by definition, a number you have never seen. Any system that can silently drop a first-time caller is a system that can lose you a new customer with no evidence it ever happened. Persistent nuisance callers do not stop at one attempt, so letting the first call through costs you very little and removes the failure mode you cannot detect.

Keep it inside your own business

Sharing caller reputation between businesses looks like free accuracy. It is not a spam question, it is a data sharing question, and it deserves to be treated as one.

Patchment scopes the blocklist to a single organization on purpose. Another company's opinion of a number never affects your line. That means the filter is slower to learn than a shared one would be, and it also means a competitor's mistake, or a dispute you know nothing about, cannot cost you a job.

What to do with the calls you do screen

Screened calls should not vanish. You want to be able to look at what was filtered, partly to confirm it was junk and partly because the volume itself tells you something. A sudden run of calls from one area code is usually a list you ended up on, and knowing that is more useful than silently dropping them.

The same review pass that produces your verdicts is what makes the rest of the front office better over time. The approval record is also what tells you whether it is safe to let the system act on its own, which we set out in deciding what your AI front office may do without asking.

A first month that will not burn you

Review every call for the first couple of weeks, and mark only the ones you are certain about. Certainty is the bar, not annoyance. A vendor call you did not want is still a real human calling your business, and marking it as not a real caller will teach the filter something you did not mean.

After that, check the screened list weekly. If nothing surprising is in it, drop to monthly. If a customer ever tells you they could not get through, that is the signal to look at your verdicts rather than the filter.

Also worth separating in your head: junk calls and missed real calls are different problems with different fixes. Filtering does nothing for the customer who reached a busy line, which is the larger loss in most shops and the subject of how to stop missing calls in field service.