By now, you have probably heard that STIR/SHAKEN is intended to prevent scammers from spoofing caller ID, in order to aid robocall blocking efforts and help trace those who placed them.
But if STIR/SHAKEN is so important, how come there are effective robocall blocking solutions available today? Even though STIR/SHAKEN is not widely implemented?
Machine learning stops robocalls
It’s all through the magic of machine learning. Several years ago I gave a presentation about machine learning and identifying robocalls was one of the use cases.
Machine learning excels when it has access to large data sets and the opportunity to get feedback on its performance.
In the case of robocalls, most data analytics tools have access to large volumes of PSTN call data, along with some designated “honey pot” blocks of unallocated phone numbers. They feed this all into the machine learning algorithms which are able to identify patterns based on call source, call duration, caller ID, volume of calls, patterns of dialed numbers, etc.
In other words, the tools are able to find patterns in the way robocallers behave, which allows these calls to be identified with a reasonable degree of accuracy.
YouMail identifies bad actors by what they say
If you read many FCC filings about robocalls, you may be surprised to see that the independent voicemail provider YouMail is regularly mentioned in providing stats and data to the FCC.
This seems a little odd at first, but YouMail has access to a different kind of data than most network operators – voicemail messages. YouMail have over 10 million individuals using their voicemail service (out of ~250-300M mobile phone users in the US), so we can estimate that 3-4% of all US mobile voicemails end up on their platform.
That’s a lot of voicemails, and (as I said earlier) machine learning tools thrive on large amounts of data. So YouMail is able to analyze the actual sound of the robocalls and compare that audio to the audio from known scammers.

This pattern-matching of the call audio against samples of known robocalls gives a new data point that can be fed into the data analytic tools – which should hopefully allow robocall blocking solutions to more accurately identify illegal robocalls.
So is STIR/SHAKEN a waste of time?
Nice try, but no. These are all pieces of the puzzle:
- data analytics based on calling patterns
- feedback based on voicemail audio matching
- prevention of caller ID spoofing.
Each of these pieces helps to more accurately identify an illegal robocall before it’s delivered to a user, but if you can combine them together you get a much more accurate and powerful tool.
If we use just one of the tools available to us, then maybe we can reduce robocalls by 60-70%. That’s great. But if there are 58 billion robocalls each year, then a lot are still getting through.
Hopefully by combining these approaches (and more) we can get to 99% effectiveness and beyond. If scammers find themselves with an ineffective tool and aggressive law enforcement traceback efforts, they’ll hopefully give up and dedicate their lives to mining bitcoin or trading on Robinhood.