Surprisingly Useful AI Article Enhancements
Before we became a global crash management platform, DingGo started with a housemate who dinged cars on a regular basis. Getting the damage fixed meant driving from smash repair to smash repair, collecting quotes in person, wondering why the hell this process was so intimidating, so opaque and so damn hard. In short, wondering why no one had digitised it yet. That small annoyance, and the years we then spent obsessing over it, is the reason there is a digital solution to this problem and why our AI works today. Our journey to effective artificial intelligence integration is the reason I’m a believer in building backwards with AI.
Consider the current state of play. MIT’s 2025 study of enterprise AI, The GenAI Divide, found that 95 per cent of generative AI pilots deliver no measurable return. Only five per cent create real value, despite tens of billions of dollars in spending. The report is blunt about the cause: too often companies bolt AI onto a workflow it does not understand, then wait for magic.
Our experience points the other way. Before you automate anything, understand the problem so completely that you know precisely what you are trying to solve. Show what good looks like then automate that. AI should come last, not first.
That lesson began with a single observation. When my housemate dinged his car, we noticed repairers often needed only a quick look at the damage before they knew roughly what was involved. So we tried emailing them photographs rather than driving over. They quoted from the photos. The whole business was hiding in plain sight.
For years afterwards we obsessed over the moment around that quote: coaching customers to take the right photograph, structuring the damage, vetting repairers, producing comparable prices, booking the job. When AI arrived, everything we had painstakingly structured suddenly became far more valuable. We had already spent years defining the problem. The technology simply made it faster.
That experience leaves founders with two things worth doing immediately:
Don’t look for an AI-shaped problem. Just broaden your perspective on what problems are now possible to solve. If the right solution is not AI that is important to support rather than favouring an unnecessarily AI-powered one. To put it another way many businesses are waving the AI hammer looking for an AI-shaped nail. We just broadened our perspective and realised we now simply had a bigger toolbelt that would let us hit a lot of nails we couldn’t reach before. The ability to instantly quote damage from photos was never possible for us before. But all the workflows, data we captured and user inputs we had could now power a unique model to make that possible. That’s not using AI as another way to solve a back-end automation problem, that’s using AI to solve a novel problem we never could before.
Your data is not boring, treat it that way. Every quote, image and completed job we logged over the years became a learning asset we never planned for. Founders sitting on years of apparently mundane operational data should look again. As the base models turn into a commodity that anyone can buy, McKinsey’s 2025 research keeps landing on the same point: the durable advantage sits in proprietary data and the workflows around it, the one asset a competitor cannot purchase. A generic model recognises a bumper. An industry system understands why this bumper, on this vehicle, with this damage pattern, matters. It’s no surprise that Google just spent millions to acquire a defunct airlines data.
Some will say frontier AI models will eventually be able to simply replicate deep and granular digital plays through sheer power. Maybe, but our experience and the experience of many at the coal face of other industry plays, especially niche industries, is telling us this is not happening and won’t for a long while, if at all.
AI can collapse work that once took years, and integrated into the right process, populated by the best data, it can supercharge a tech offering. What it cannot do is hand you a deep, specific understanding of your customer’s problem and your hard-won, tirelessly built piece of the market. That still has to be earned and AI can’t earn it. The founders in MIT’s successful 5 per cent picked one pain point and executed it well. The ones chasing the newest model stalled.
So if our experience is anything to go by, resist the urge to start with the technology. Skip the opening question of which model to use or where to insert AI. Go to the coalface of your industry and learn one problem better than anyone alive. Get obsessive about every single problem your customers struggle with. Structure everything around them. Then, and only then, integrate AI.
AI is changing what a startup can build and the economics of building it. It has not touched the founder’s real job, which was always to find the gap and know it like no one else.
Shaun Janks is Co-Founder & CPO at DingGo
Last Updated on October 10, 2026 by Shaun Janks



