Why conversion rate misleads: Bhavik Patel on reading the numbers

Bhavik did not plan a career in data. He joined Photobox in the acquisition team, building landing pages, running A/B tests and working across search, display and performance marketing. Nobody owned the web data, so he found himself inside it. When a head of acquisition role came up, his manager told him to wait: marketing was not where he would thrive. When a head of analytics role opened, colleagues pushed him to apply because he was the most data-driven person in the company. He puts his career down to luck, and to someone else seeing it first.

Aziz asks why he can translate numbers into plain English when many analysts cannot. Bhavik's answer is that he started on the receiving end. Working in acquisition, next to the product and trading teams, he learned the problems commercial teams face before he ever ran a data team. Context, he says, is a big part of what makes a data team successful. A data team removed from what is happening in the business cannot do the job.

He uses AI the way a boxer uses a sparring partner: to test work he has already done. "AI is great if you know what you're doing," he says, and it creates false confidence if you don't. He does the analysis himself, then uses AI to help structure the write-up or check a script, and he keeps the final edit. "I'm still writing the book," he says; AI is the editor.

His case for AI in a business is productivity. Used to cut waste, it should let revenue grow faster than headcount, so revenue per head rises. Aziz adds the warning from his own field: marketers who use AI to do a job they don't understand cannot judge what it gives them, so they cannot add value with it.

Bhavik was studying econometrics when they spoke. His reason is practical. Privacy rules and cookie consent mean tracking loses part of the signal between an ad and a sale; he estimates the loss at about 30% in the UK. And marketing drives things a click can never show, such as people walking into a shop.

His example is launching a feature in Egypt that cannot be A/B tested because everyone must get it at once. You build a "synthetic Egypt" from other markets that behave like it, and compare. That synthetic market is your counterfactual: what would have happened without the launch. The method comes from economics; Abadie, Diamond and Hainmueller set it out in a 2007 NBER paper that built a synthetic comparison region to estimate the effect of California's tobacco programme.

Asked what people in their twenties should do, he gives three things. First, meet people, but choose events full of practitioners at your level, not sponsors and fancy job titles. He started CRAP Talks (conversion rate, analytics and product) after a big conference put people on a data panel who knew nothing about data. Second, learn how your company makes money. "Go and speak to finance and understand how do we as a company make money." Third, write. Writing gives you clarity, clarity gets you buy-in, and once you have buy-in, he says, "you'll be unstoppable."

Aziz sees the Pareto principle in every monthly client report: three things the team did drove 90% of the value. In data, the list of things you could measure has no end, so how does someone new find the metrics that matter? Bhavik's answer is a tree. Revenue is orders times average order value. Orders are customers times order frequency. Customers are traffic times conversion rate. Each team sits on a different branch. "Go and find out what your North Star is for your area," he says, whether that is acquisition rate or add-to-cart rate.

He also borrows an idea from Jeff Bezos: build around what will not change. Customers will still want good prices, fast delivery and good service in five years. A data team's customers will still want insight, and they will still want it yesterday.

Then the question in the episode's title: which KPI is most misunderstood? Bhavik picks conversion rate. At Gousto, a subscription business, it was the primary metric. But a subscription depends on two things: breadth, how many people buy the first time, and depth, how often they come back. Chase conversion with cheap offers and you fill the base with low-value customers. Breadth rises and depth suffers. He links this to the grow-at-all-costs thinking behind the tech layoffs.

"Conversion rate is a metric that can be artificially inflated by just doing less of something. It doesn't mean that you've driven your commercials forward."

It works the other way too. A team may switch off a cheap channel that was bringing in extra sales, because it drags the rate down. And a high cost per acquisition can mean you are buying high-value customers. Aziz's rule: "A disaggregated conversion rate is the only useful conversion rate." A single top-line rate can move for any number of reasons. Split by channel, it tells you something. This is why our reporting starts from the business, and why we wrote up the marketing KPIs worth reporting.

Bhavik closes on what he has changed his mind about. A year earlier he took up boxing to get his health back, and found a community he had never expected. Face-to-face events matter for the same reason: you can learn the academic side alone at a screen, but the worldliness that comes from other people's lived experience cannot be replicated by AI.