The Problem
Last year, 315 people were admitted to hospital with knife-related injuries around Mother’s Day. Sadly, too many didn’t make it home.
The Campaign
Highlighting the heartbreaking fears of a mother whose child may have been involved in a knife crime, the campaign aims to engage with an audience of young males aged 16-to-24 and make them feel differently about carrying a knife. With unique, bespoke technology, the hard-hitting ad creative is triggered by the sound of ambulance sirens nearby.
The campaign launched on Mother’s Day 2023, the campaign was due to go live before the Covid pandemic and was pushed back by three years.
Good Morning Britain picked up the Campaign. This was a fantastic opportunity to spread the important message even further.
The Technology
M&C Saatchi London and Clear Channel approached me to determine if/how we could make an OOH poster understand the sounds in its environment, specifically the sounds of a siren. At the time, machine learning wasn't as mainstream as it is today. After some research, I found Teachable Machine and TensorFlow (the technology behind Teachable Machine).
I trained the machine learning model to listen to the city and detect sirens. I then wrapped a custom Javascript layer around the machine learning system to control the poster visuals directly. The biggest challenge was removing the false positives, as the results from the TensorFlow data could be very erratic. I got around this issue by passing all the ML data into an array and then normalising the data by taking an average over a few seconds. This was the breakthrough moment that took weeks to figure out.
I trained the machine learning model to listen to the city and detect sirens. I then wrapped a custom Javascript layer around the machine learning system to control the poster visuals directly. The biggest challenge was removing the false positives, as the results from the TensorFlow data could be very erratic. I got around this issue by passing all the ML data into an array and then normalising the data by taking an average over a few seconds. This was the breakthrough moment that took weeks to figure out.
I recorded background sounds from central London as a base sound and then recorded an ambulance to get the trigger sounds. Luckily my cousin drives an ambulance so I could gather all the sounds an ambulance makes in a field in the depths of Oxfordshire one evening in January.
Next came testing, so much testing! I became so obsessed with the tech that I set my hotel room up as a test area with a live webcam recording the street's sounds and the system's logs. How the hotel I stayed at didn't think I was up to some nefarious activity and reported me to the police was a miracle!
After many weeks of testing, I'd tweaked the ML model and Javascript to a level where not only did it fire accurately, but it actually started to visualise the sounds of the city. I could now get a result that ranged from 0 - 100, depending on the background level of the sirens in that location. It was amazing to see the machine learn what the city sounded like and respond accurately. For me, this was a chilling moment linked directly with the anxiety a mother must go through. The thing about Anxiety is that it ebbs and flows; it plays with your mind. Was that an ambulance? Is it a faraway ambulance or just a bus breaking? Is that for my son? The machine started to emulate the fear and confusion a mum must go through and represent it visually. We used this data and linked it to the text messages and the speed they came through on the poster.
I then worked with the incredibly talented Jason Prout at Clear Channel to integrate the technology into a custom 6-sheet bus shelter. Jason worked his magic, rigging the right parts and cramming PCs into the shell to create the final poster.
The work was developed in partnership with creative agency M&C Saatchi and Out of Home media and infrastructure company Clear Channel for the Ben Kinsella Foundation.