Monday, May 3rd, 02021 at 11:11 UTC

Before you implement your own robocalling system, it is important to understand the legality in your area. With our projects, we have consent, the receiver if expecting to be reminded and has the option to opted-out. If your situation is different, be sure you understand what you’re getting into.

We’ve found robocalling (Programmable Voice) a very easy and successful tool in our survey toolbox.

For several years, we have run surveys in compulsory schools in Iceland. That required us to collect lists of student names and randomly sample them across the school year. We conduct surveys on behalf of our customers, the schools, so they work closely with us. For our student surveys, a designated contact person in each school helps to find and sit-down each of the randomly sampled students each month.

For us, we never contacted the these student participants directly, the school staff does. As we grew our offerings, we took on the task of surveying the children’s guardians, then staff. In doing so, we now reach out to those participants directly.


Our reminder regime always starts with a letter to parents letting them know about the survey and the ability to opt-out. Since schools have a legal requirement for evaluations, the scheme is opt-out rather than opt-in. Once the schools give us their participant list, we randomize it and start the first contact via email. A week later, we follow-up with an SMS. Then a week more, another email, then the following week, a human call.

This method worked really well, since the SMS caught anyone who’s email client marked us as spam. Also, by having a mobile friendly survey website, people could click a link in the SMS and start answering immediately.

But even after two emails and an SMS we still need to manually call to get the response rates up to the 80% mark. This took time and cost money, which was passed onto the customers.

Then, we switched our SMS provider from a local telecom, to Twilio. They have a simple, programmable API to send SMSes. But more importantly, they also offer a programmable voice interface todo robocalls!


Now, we send a first email reminder, an SMS, a second email, then a robocall, and if the participant still hasn’t started, we can make a manual human call. The script for the robocall is handled dynamically and anyone on the team can update the text and regenerate the audio in multiple languages.

As part of the system, we can accept touch tone input and make dynamic requests to our server for additional functionality. Twilio offers a webhook to give us the status of the call when it finished. We use this to log ‘connected’, ‘no answer’, ‘busy’, ‘failed’ or other statuses. That way we can follow-up or ignore participants as needed. All this for roughly the price of sending an SMS.

The response to the service has been amazing. This spring, we contacted approximately 15,000 people with robocalls. It boosted the response rate extraordinarily. This could be due to the fact it is ‘new’ in Iceland to get a robocall and the affect might wear-off or it is genuinely a useful tool in our toolbox for our customers.

There were a lot of things we took into consideration in the design of the robocalls. We never tried to fool anyone. The mp3 that is played is in Icelandic (or press 2 for English) and is a Text-to-Speech (TTS) voice. It sounds good, but it is obviously not perfect. We also start the message with “This is an automated call on behalf of …” as to not surprise anyone. It is important to announce that you are a robotic AI tool.

Since we have the opportunity to interact with participants via the touch tone buttons, we allow people to switch to English from Icelandic and to also request their survey login code to be resent to them with a press of a button. Making things more self-service increases the response rate. Previously, people got a reminder call, but then then needed to go dig out an old email to find the access code and link. We try to make that process as smooth and simple as possible via the press of a button.

Efficiency Downfall

The other amazing thing that the robocalling affords us, is also its spammy downfall. With a human, they can dial, read a script and answer any questions from the participant in about a minute. That means, maximally, a human can call around 60 participants per hour. To call 15,000 people once, it would take a single person calling non-stop 24h a day for 10.5 days! That’s just not possible if you want to respect people’s time and not call them while they are at work or dinner or after a children would be sleeping. Therefore, we have 2-3h window each day to contact as many people as possible. The way to solve the problem is to scale horrizontally and hire more people to call at the same time.

Using Twilio, we can call around 2,000 people per hour. That approximately a 35x increase in efficiency for a 50th the price. You can easily see how this can be abused!

Now our workflow takes programmatic voice into consideration. All the translations and Text-to-Speech (TTS) mp3 creation is done via AWS (Amazon Polly) the same as all other parts of the survey.

Our experience with Programmable Voice over the last two years has been great. We are excited to continue developing and exploring the possibilities for programmable voice tools.