Electric Twin

Two years ago some of the News UK team called us the '𝗘𝘃𝗶𝗹 𝗧𝘄𝗶𝗻'. This week their Head of Customer Insights put his name to a quote calling us an agile, flexible and complementary tool for answering business questions. Not an easy feat closing a gap like that! It all was possible because Chris Courtney-Smith at The Times went first and was willing to experiment. He took the internal heat, ran the work, and shared the results with the business. Now, two years later The Times has cut their research time from over two weeks to 90 seconds. Decisions can now get made in the room, rather than weeks later. On a new parenting channel, they saw retention increase by 7%, new visitors increased 3x and reader engagement doubled since implementing insights derived from Electric Twin. How did they get there? Step by step. The team ran one query, checked it against what they already knew, and expanded from there, gaining trust at each step. Every enterprise deal we've closed has followed a similar process... Often the hardest part of moving a business forward is being the person that decides to take the plunge, standing in front of colleagues, defending something unproven and convincing colleagues who've built their careers on a method, asking them to try a new one. Chris did that and last week presented the results at ESOMAR with our VP of Partnerships Leanne Tomasevic.

Everything we promise at Electric Twin depends on one thing: accuracy. Speed and scale are worth nothing if the answers are wrong. A wrong answer delivered in minutes is more dangerous than no answer at all, because teams act on it with confidence. So accuracy is the standard we build everything else on, and we test it in a way that leaves us nowhere to hide. Prof. Michael Muthukrishna, our scientific advisor, explains the method in the clip below. We take a real dataset and remove 10 to 20 per cent of the questions before we build the synthetic population. The model never sees them. Then we ask it those hidden questions and score its answers against what real people actually said. It predicts them or it fails, and we see exactly where. Just as important is what we do with the results. We share the method, the scores and the gaps with our clients, and we run the same test on their audience with their data, rather than pointing at a number from someone else's study. If a client wants to design the test themselves, better still. That transparency is really important to us. Our clients have to defend this technology to their own boards, and the strongest position we can put them in is being able to show exactly how it was tested and what it scored. Claims are cheap but evidence isn't.

Martin Sorrell built the largest advertising empire in history. When he speaks about what agencies need to do to survive AI, it's worth paying attention. His argument is straightforward: agencies that hold onto production as their core value proposition are already losing. AI commoditises the output. What it can't commoditise is understanding the audience. That's the gap with which most agencies haven't yet fully reckoned. The agencies we work with that are moving fastest aren't the ones debating whether AI will change their model. They've accepted it will, and they're asking a better question: what do we know about our clients' audiences that nobody else does? Proprietary audience intelligence (data and the means of understanding it) is becoming the moat, rather than headcount and process process (and that applies not only to agencies but all consumer-centric businesses). 👉 Read the full post & link to the article in the comments.

“Two weeks down to 90 seconds.” That's the shift The Times described when they moved product testing over to Electric Twin. Their previous workflow ran on an external panel with a standard two-week turnaround. They had a solid process that was widely used, but it set a hard ceiling on how fast their team could move. Now they test in the morning, iterate by lunch, and run another round before the end of the day. Speed matters. But the bigger change is what speed makes possible: More assumptions tested before resource is committed. More questions asked before a direction is decided. Fewer calls made on instinct because the data wasn't there in time. When customer feedback takes weeks, teams stop asking questions they don't have to. The research queue shapes what gets asked. Which means it shapes what gets decided. When you can get feedback in minutes, that constraint disappears.

A prediction you can't check is just an opinion with a decimal point. So the question we get asked most is a fair one. How do you know Electric Twin is accurate? We test it two ways. The first is holdout testing. Which works by taking a survey with 40 questions and holding out five of them. You build a synthetic audience from the other 35, then put those same five questions to that synthetic audience. Now compare the synthetic answers with the real ones and see how close they are. It mirrors what an insights team does every day, where the question you really care about is usually just one more question on a survey you've already run. We also tested Electric Twin against seven identical surveys across a range of methods, pre-registering our predictions before we'd seen any results. Run with Professor Michael Muthukrishna of LSE and NYU, the experiment compared Electric Twin against traditional market research methods. It turns out we were just as accurate. Have a watch of the video if you want to know more about accuracy at Electric Twin.

A study just appeared in Science that used Reddit's "Am I The Asshole" posts to test whether AI tells you the truth. Across 11 major models, chatbots affirmed users 49% more often than humans did, even in cases involving deception, manipulation, or illegal behaviour. One conversation was enough to make people more certain they were right and less willing to apologise. This doesn't surprise me. These AI tools are designed for the work place, coaching and encouraging us, to deliver better work. However, use them for the wrong thing and they go from assistive to detrimental. This isn't a bug as much as it is a design decision. At Electric Twin, our platform isn't designed to flatter you, it is designed to give you accurate reflections of the real world - even if everyone hates your idea. Great for understanding customers, perhaps not always great for your ego.