Case studies
Real projects. Measurable outcomes.
Real delivered work, kept anonymous — client names are kept out of it, but nothing here is illustrative or hypothetical.
01
A cluttered screen, rebuilt around how people actually work
A logistics company running Salesforce for the better part of a decade.
One of the most-used screens in the system had grown to nearly 100 fields on it — most irrelevant to any one person's day-to-day, with no way to simplify it without redesigning the whole page for everyone else too.
Rebuilt as a personal view: a handful of fields everyone needs, plus a simple search box to add anything else that matters to how that person works, arranged however makes sense to them. Finished fields tidy themselves away automatically, so the screen always shows exactly what still needs attention — nothing else. Also closed a costly, invisible bug along the way: a setting that was silently defaulting to the wrong option and letting paperwork go out completely blank.
Built and proven as a clickable, working prototype in a single session, shaped by real feedback before any of it was built for real.
02
An AI assistant built into the everyday system, not bolted on the side
A professional services firm creating new client records and quotes dozens of times a week.
Staff were creating new records — new jobs, new clients, new pricing — through long, fiddly forms, and it took knowing exactly which screen and which fields to use for each situation.
Built a conversational assistant living right inside the business's own system: staff describe what they want in plain English, and it handles the paperwork underneath — showing a proper on-screen form to check and confirm whenever real data needs entering, never a black box that just writes things silently.
The trickiest part: one step was quietly defaulting to the wrong setting and letting quotes go out completely blank, with no error or warning anywhere. The problem was invisible to standard automated testing, because that only ever checks clean, simple scenarios — it only showed up by testing real conversations and watching exactly what the assistant actually did, step by step.
03
A secure dashboard for people outside the usual system
A multi-company group needing visibility into performance across its operating businesses, without handing out full system access to each one.
Needed visibility into a handful of key numbers, without needing full access to the underlying system or the risk that comes with a raw data export.
Built a live dashboard where every figure is calculated fresh each time, never pulled from raw records, with access switched off for absolutely everyone by default — even senior IT staff — until explicitly turned on for a named person. If anything goes wrong, it shows a plain message and hides the numbers entirely, rather than risking a wrong or misleading figure.
Agreed the look and the numbers with the finance lead using a working mockup before any of it was built for real, then refined it again after a second round of feedback — both rounds took days, not weeks. Every figure was checked line-by-line against the business's own existing reports first — one easy-to-miss detail would have overstated a key number by six times if it had gone unnoticed.
04
Tracking down three invisible causes behind one broken screen
A business running a lightweight external portal for partners and contractors.
A simple external-facing area — a shared calendar and a filtered list — needed to work for a non-technical outside user. Both looked completely broken, with no error message anywhere, not even in the technical logs.
Found three completely different causes behind the exact same blank screen: a missing permission, a hidden restriction on specific pieces of information, and a visibility setting that doesn't even show up when checking the system directly — that last one could only be confirmed by testing the real screen itself. Also hit, and solved, a dead end where the system reported a permission change as successful while silently not applying it at all — the real fix was a different kind of access rule entirely, once the actual cause was properly tracked down rather than assumed.
The result: a working external area, and a reliable way to diagnose “it just doesn't work, no error” problems — the kind that look identical on screen no matter which of several very different causes is actually behind them.