The Price of Getting It Wrong

Deliveroo UAE was barely a year old. When you're moving fast and building from scratch, flat fee pricing makes sense. It's one less variable, one less thing to get wrong while you're figuring everything else out. Simple scales.

And then the business grew. The customer base diversified. Order volumes climbed. What had been good enough at year one was now actively working against us. The market had matured past the point where one number could do all the work.

Every high-demand moment was leaving money on the table. Every price-sensitive customer in a low-demand window was a churn risk we didn't need to take.

I had no data science team. No pricing function. No inherited framework to build on. What I had was SQL, Excel, and a conviction that the data already knew the answer.

I pulled order histories, mapped demand curves by time of day, day of week, distance band, and restaurant category. I ran price elasticity models in Excel, testing how sensitive different customer segments were to fee changes at different price points. The work that at a larger company would have gone to a team of analysts, I did alone, iterating the models until the patterns held.

What the data showed: customers weren't uniformly price sensitive. High-frequency users barely flinched at higher fees during peak hours. Occasional users dropped off sharply when fees crossed certain thresholds. And there were entire distance bands where the flat fee was actively suppressing order volume. We were charging the same for a 2km delivery as a 12km one, which meant we were undercharging for the hard deliveries and overcharging for the easy ones.

The model pointed clearly at one move: reduce delivery fees in certain distance bands and time slots to stimulate volume. The logic was sound. The data supported it.

But if it backfired, the revenue hit would be immediate and visible. No lag, no ambiguity. Every order in those bands would be worth less, and the volume uplift had to materialise fast enough to cover it.

I made the call. Tiered the fees. Watched the data daily for the first two weeks.

6% revenue growth. Built entirely on repricing, not on new customers or new markets. The same customers, the same restaurants, the same demand. Just priced properly for the first time.

The distance bands where we reduced fees saw the volume jump the model predicted. The peak-hour tiers held. The occasional users who had been dropping off at the fee threshold came back.

Good analysis only gets you to the right answer. You still have to be willing to act on it when the downside is real and immediate.

The model doesn't carry the risk. You do.