A research function is worth what it changes
Discovery infrastructure and a design system are expensive, and both are easy to lose against feature work. The argument that wins is the one made in the currency the business already uses, which means the work has to ship and the change has to be measured.
Two companies, the same test
Conrad Electronic. Ten initiatives shipped in 2025 and measured across the EU over their first twelve months, carrying €30.3M of revenue impact. Nine of the ten are A/B test results. Behind them, five decisions the function supported with evidence that will not produce a clean figure for years.
FTI Group. The homepage and the comparison page, which carried most of the traffic and most of the debt, rebuilt on Next.js and the Sunscreen design system. Launched in Austria at a 9 percent higher click-through rate and half the development time for later features, then stopped by an insolvency three months later.
Before both, four years at CHECK24 leading product and user experience for insurance comparison, where every change is judged by conversion and A/B testing was already the native language of the company. The work there was making qualitative customer evidence compete on the same terms as the numbers, in a culture that had no trouble measuring and no habit of asking why.
What is the same every time
A hypothesis written down before the work starts. The one-pager at FTI, the repository-backed hypothesis at Conrad. Nothing enters a backlog on a hunch.
A control. A variant against a control is the only claim that survives someone checking it. Where that is impossible, as with the consent banner, say which figure is modelled and who modelled it.
The number the business already uses. Revenue per visit, click-through rate, orders. Not design-quality scores, which persuade designers.
Separate surfaces, measured separately. Three changes to one page, each tested on its own, beats one redesign of everything with a single before and after.
Credit where it belongs. The function supplies the evidence and the design. The teams ship and other people own the decisions.
What I would do differently
Price the current state before proposing the replacement.
Infrastructure arguments are easy to lose against feature work, and the number that wins them is the measured cost of doing nothing. At FTI a return on investment model carried the funding decision for exactly that reason. At Conrad I started building before I had the equivalent figure, and spent a quarter arguing from principle when I could have argued from arithmetic.