My eight year old son figured out a hack to make the music service work better for him.

The kids have a Google smart speaker that is attached to a Spotify account so they can just ask for any of their favourite music. Anyone who has pre-teens in the house probably knows how much such a setup is used - all day every day.

Coming from someone who had to run to the double-cassette boom box to press the record button any time a new favourite song came on the radio just so I could listen to it whenever I wanted to, this really does seem like the Jetsons in reality.

Here's the thing - they don’t really even have to know the title of the song and who sings it exactly - that combination of Google and Spotify is actually pretty good at figuring out what song they want through the vaguest of kinda-sounds-like prompts and, I suppose, their listening history and a playlist my wife keeps up-to-date.

Yes, it’s a marvel of modern technology by my standards, but for them it’s just normal - they can’t imagine a time when you couldn’t just wish any song you want onto the radio right then and there.

There’s a problem though. Every once in a while the system refuses to recognise a song, even when the eight-year-old repeatedly said the same prompt that worked last time, or his older brother has intervened with the real title and artist’s name.

And it doesn’t just respond with a “I didn’t understand your request” - it just starts playing random stuff from Spotify, and he has to tell it to stop, and then try again.

Needless to say this is very frustrating for an eight year old who’s still struggling to make his needs understood by the world.

But my son has found a solution to this. A simple hack, as they say: He asks for the approximate song name that he wants, and adds “by John Williams”.

Now, if you don’t know, John Williams is a contemporary composer popular in film scores. My son never actually wants something by John Williams. But through what must be thousands of frustrated trial and error experiments he has determined that, as a last resort, declaring that the song was actually composed by John Williams will work.

It’s bonkers. He swears by it. And has proven his theory to me on several occasions.

He did it again yesterday. The system was refusing to play some terrible track from the Eurovision Song Contest that his brother has been repeating 1700 times a day for the past few months.

He said “I got this”, and he asked for the song title “by John Williams”. Lo and behold, Google responded “Ok, playing {whatever that song’s name is} by the Eurovision Song Contest 2025”.

Amazing. Is it a feature? Some weird bug or easter egg that he’s discovered?

Maybe.

My theory is that the combination of two company’s gamed interfaces - voice recognition, search and the products they want to push your towards - sometimes puts the system in a box where it just keeps offering answers from a small set of cached data. And that small set doesn’t include the correct answer.

So how do you break out of this box? You need to ask for something different enough from the original query. Using the original title with an artist like John Williams works because you’re still searching for the title you want, but John Williams is different enough to break it out of its box. Add to that the fact that there’s almost no chance that John Williams wrote a song with anything like that title means it'll find the next best answer from, say, your history.

An bang. The hack works.

Alright, a cute story. But there’s actually a relevant lesson in this!

When you have a complex system in place - let's say it's something that that's been configured, tested and validated for a particular purpose - sometimes your users may find ways of ‘hacking’ the system to make it work better for them, in surprising ways.

This kind of behaviour is especially prevalent in this day and age where have more prevelance of ‘fuzzy’ ways of inputting data. ‘Fuzzy’ like natural language queries, voice input, gestures or image recognition.

Is it ok for that data to be entered in by speech-to-text?

How about that person who's taking a photo of the paper form and using text recognition to cut and paste into the system?

The way that the users of a system interface with the it may change over time as the tools they have access to change, too. You may need to constrain how they are allowed to interface with that system - for example by configuration or by saying what’s allowed in the procedures.

Either way, should visit the users once in a while and see how they’re using the system - they might surprise you.

– Brendan

p.s. Enjoy this message? Read more at the Hyland Quality Systems website.