Case study
A quiet problem, hiding in plain sight
Auditing the allowances given to employees
Some BT colleagues receive on-call and shift allowances, extra pay for being reachable at odd hours or working shift patterns. Nobody was reviewing whether those allowances were still warranted. Guidance existed, but wasn't being followed. Documentation wasn't handed cleanly from one manager to the next. And when a new manager inherited a team, the last thing they wanted to touch was someone's pay: it's someone's livelihood, and without full context, the safer move was to leave it alone.
So allowances just kept flowing, unreviewed, by default rather than by decision.
From "it's just a form" to "this is someone's income"
The initial framing was simple: build a form. See your team, see their allowance, decide if it's still right. Practically a Google Form with extra steps.
That framing didn't survive contact with the actual stakes. This wasn't a routine data-entry task: every dropdown selection could end or continue someone's income. The team was also working within the real constraints of Microsoft PowerApps, which meant design ambition had to be negotiated against what the platform could actually do.
Fast iteration, with real judgment calls
Alongside another researcher, I worked directly with the designer and developer in tight, rapid cycles: test, discuss, change, retest. The recurring hard question after every session wasn't "what did we learn," it was:
Is this enough evidence to justify changing it?
A judgment call with no formula, made collaboratively each time. What started as a bare decision form became something much more deliberate:
- A clear three-option decision: continue the allowance, discontinue it as incorrect (the colleague may be owed a different allowance instead), or discontinue it as the colleague being ineligible
- Instructional content, FAQ links, and guidance articles rewritten for managers with zero context on allowance policy
- Careful language on what submitting an audit actually meant, spelled out before a manager could commit to it
- Even the notification email got design attention: this wasn't "researcher hands off findings," it was research, content, and design decisions made by the same small team in real time
That shift, from running studies to actively shaping the copy, the flow, and the guidance, is where the role moved from researcher to strategist.
What shipped
The final tool guided managers through a four-step review: review each colleague, make a decision, discuss with your colleague, submit your review, with FAQ links surfaced at the top rather than buried in a help center. Each manager saw their team in a single table (framework, allowance type and value, a decision dropdown, and a comments field) tracked against a visible progress bar. Submission required confirming an explicit summary of consequences before it went through, closing with a clear success state so managers knew the audit had actually registered.
The numbers
The tool launched in early July. Over the following four weeks, 90% of 700+ managers completed and submitted their audit, covering 94% of colleagues receiving shift or on-call allowances. Across all of them, only two queries about how the tool worked, and zero system issues. The programme lead was able to self-serve through the entire audit period without needing extra support from the team.
Early analysis points to roughly a 5% error rate in allowances: potentially surfacing over £500,000 in incorrectly paid allowances.
The programme lead's feedback was unprompted, and specifically called out how little support the tool needed given the number of people who used it. It was escalated by their team to a Director, who shared it back with the whole delivery team.
What this taught me
I went in assuming this was a small, low-effort project: just a form, how much could it really need? The deep dive proved the opposite: at this scale, every word of instruction, every label on a dropdown, every line of an email is a decision with real downstream weight. Tactical, unglamorous work, done with real research discipline, can carry outsized impact. It's a lesson I carry into every project since, glamorous or not: the size of the ask has very little to do with the size of the impact.