When you build a product, you spend a lot of time imagining how people will use it. Then real teams start using it, and you learn more in a week than in months of planning.
One of the things we’ve been focusing on while building Nelly is something quite simple: the same data doesn’t necessarily lead to the same recommendation.
Two merchants might have similar sales performance, advertising costs, and product portfolios. But one wants to maximize revenue, while the other is focused on increasing visibility or reducing refunds.
Those are different problems, and they require different approaches.
That’s why Nelly starts by understanding what the merchant is trying to achieve.
From an engineering perspective, this matters because recommendations aren’t just about finding patterns in data. They’re about evaluating possible actions against a specific business objective, while taking the available context and constraints into account.
And there’s another important part: Nelly doesn’t make the final business decision.
The merchant does.
Our job is to make that decision easier by bringing together relevant information, explaining the reasoning, and reducing the amount of manual analysis required.
As more teams start using Nelly, I’m particularly interested in learning which recommendations they find useful, which they choose to act on, and where additional business context changes the outcome.
Generating recommendations is one thing. Helping someone make a better decision is another.
That’s what we’re building toward.
Decode the why. Drive the buy.