Something I've noticed about how most companies approach their use of AI is that they treat it like a product launch. There's a go-live date, an announcement, a big moment, and then the team moves on to the next initiative. The system gets left to run in the background as if it will keep up with the technological times, which are changing faster than ever before.
That's not what we did with DSTV, and I think it explains why, years later, it remains one of the most robust things we've built.
We started the engagement the way we'd want to start every engagement, though it doesn't always work out this way. Before writing a single line of code, we sat with the data to understand what was actually driving call centre volume. Not what we assumed was driving it, but what the evidence said. The team noticed that a small set of query types accounted for the bulk of calls coming in, so we built for those first. It sounds obvious when you say it out loud, but it's surprisingly rare to see it done properly.
From there, the work didn't stop when the system went live. The model has been retrained continuously, integrations updated, and new use cases have been added as the client's needs have evolved over the many, many years we’ve been working with them. Utterance match rates are close to perfect and customer satisfaction has kept climbing rather than plateauing, which is what you'd expect from a system that's been actively looked after rather than handed over and forgotten.
