Every design thinking workshop I've sat through opens with the same five-box diagram and a straight arrow running left to right: Empathize, Define, Ideate, Prototype, Test. It reads like a project plan with a start and an end. The first time I ran a real initiative this way (early in my career, moving a business from one-off purchases to subscriptions), I treated the arrow literally. We interviewed users, defined the problem, built the thing, tested it, shipped it. Linear, tidy, and wrong in exactly the way that mattered.
The test that sends you back to the start
Testing told us something the interviews hadn't: people weren't rejecting the subscription price, they were rejecting the idea of losing access if they ever lapsed. That's not a Prototype problem. It's an Empathize problem we'd actually gotten wrong. We'd framed the customer's fear as "too expensive" when it was "too risky." No amount of re-prototyping the paywall screen was going to fix a wrong definition of the problem sitting two steps upstream.
The funnel diagram has no honest way to draw "go back to step one, you misunderstood the person." So most teams don't. They stay in Prototype and Test, iterating on a fix for the wrong problem, because the diagram told them the hard part, understanding the customer, was already behind them.
What the loop actually looks like
In practice, every project I've run worth remembering looped at least twice through Empathize and Define before Prototype ever stabilized:
- A first pass that gets the obvious need right and the real constraint wrong.
- A test that surfaces the real constraint, usually as a surprise.
- A second Empathize pass, narrower and sharper, aimed specifically at that surprise.
- Only then does Prototype stop changing shape every week.
At Ovela, the loop ran through pricing tiers. We assumed three tiers (economy, standard, premium) would need three different UIs. Testing the first prototype showed patients didn't want to choose a tier at all; they wanted to describe their budget and let the concierge match them. That sent us back to Define, not forward to a nicer dropdown.
If your "Test" step only ever produces a backlog of polish items, you're not testing the problem anymore. You're testing the paint.
How to actually loop without losing momentum
The risk of taking the loop seriously is that it becomes an excuse to never ship. Three things keep it honest:
- Time-box each full pass. A loop is healthy at two weeks and an excuse at eight.
- Write down what you thought you knew before each Empathize pass, so you can see exactly what changed, not just that something did.
- Re-enter at the narrowest point possible. Going back to Empathize doesn't mean re-interviewing everyone; it means re-checking the one assumption the test just broke.
The diagram I'd actually draw
Five boxes, still in order, but with a short arrow curling back from Test to Empathize, thinner than the main loop, used often, and never treated as a sign that the process failed. It didn't fail. It worked exactly as intended: it told you something you didn't know, earlier and cheaper than shipping would have.