The Creation of a Beast

Honestly: none of us at RadioAnalyzer knew where we would end up that day in spring 2023 when I announced a notable shift in development focus on our staff meeting.

“We’re going to create a breakthrough for the whole industry by testing spoken word content the same way radios test music. It will be a gamechanger for radio Worldwide”, was my bombastic statement.

One of our developers, Tamas, replied with a laconic: “Well, no pressure, then”.

But he and the team accepted the challenge with the ingenuity, tenacity and humble attitude they use for everything. And sitting here today; on the brink of releasing Content Testing in the summer, I feel like Dr. Frankenstein.

Because we have created a beast. A beautiful, powerful and highly useful beast.

Allow me to illustrate Content Testing

This is what I learned in 10 minutes looking at the morning show at a client station used for testing the new tool:

Lesson 1

• One of the least spoken about topics is politics. But politics is the best rating subject in a country where the economy and the political life can best be described as “tumultuous”. I see this as the radio station underestimating the listeners and their desire to discuss difficult subjects of interest to the whole of society.
• Lesson: if I were the producer or PD, I would look into ways to talk about political issues in a way that fits the format and flow.

Lesson 2

• Two of the most spoken about topics – Entertainment and Music – are also the two lowest rating topics. Looking further at the breaks from those categories, it seems to me that the hosts are on “autopilot”. It’s not creative. It’s not unique. It’s what the listener already read on social media yesterday.
• Lesson: Though the topics may be among the favorite subjects of the target audience, remember you have to bring something new or different to the table than Facebook already did yesterday.

Lesson 3

• Listener interaction rates poorly when it’s just a listener on the phone talking about a subject. But when the listener is challenged to be part of a game or a setup where they need to be actively involved with other listeners or the hosts, it works better.
• Lesson: adjust the interaction accordingly. Test new games and see how they work next week.

Lesson 4

• Superbowl might be something the hosts look forward to, but the listeners do not care. At all. I saw a lot of breaks where the upcoming sporting event was spoken about, so I tested the keyword “Superbowl”. Only the bits about Taylor Swift were remotely interesting to the listeners. The rest rated very poorly.
• Lesson: this is a classic: Kill your Darlings.

Imagine the level of insight a producer or PD could get by using this tool weekly. I just spent 10 minutes and the findings I uncovered were equal or better – plus way more actionable – than the intel an expensive and slow panel test or focus group would have provided.

Without giving away our secrets, this is what we do:

1. Record the audio, isolate the talk breaks and put them through a “speech to text” mechanism.
2. Let an AI algorithm compress the transcribed voicebreak into a short and quickly read summary of the voicebreak.
3. Boil the summary down even further into keywords and topics.
4. Test every single group of voicebreaks just like we would test a song. Example: all voicebreaks containing “entertainment” are tested as a group.
5. Display the results graphically including an indication showing which percentage of voicebreaks contain the tested keyword. This can be compared to “number of plays” in a music test.
6. Make user modification possible. So if you want to test something else than the pre-defined topics – like the Superbowl example – it’s easy and fast. Just search the keyword you want, start the calculation and get an almost instant result.

Just like music testing it *is* possible to get too much of a good thing, so our system makes a test once pr. week. We also display the score development within the different topics so you can track how the adjustments you make work and how listener interest changes over time.

The reception our new pet tool got at RadioDays in Munich was overwhelming. It seems we were right that this is exactly what the radio industry craves to get valuable data within a new areas instead of just more data in the same area we have always been researching.

Currently we are putting the finishing touches to different language models and we are ready to let clients roam free alongside our new beast when summer is upon us.

If you want to know more about Content Testing – reach out to us!