StarcountData intelligence · Enterprise · Desktop web

Hit £1M first-year target, in 91 days.

Starcount built a clever dataset of its own, and had rights to data nobody had monetized before. But all that data wouldn’t become a product, until I made it lie.

£1Mfirst-year target

91 daysto reach it

1st everTwitter data deal

RoleHead of Product Design, first designer

TimelineDec 2015 to May 2019

UsersEnterprise marketing teams

reports

Starcount had unique data, but sold it as consultancy reports.

Its inference layer tied what people follow to what they care about: relations advertisers pay for.
All insight flowed through one product, The Observatory, used only by Starcount’s own consultancy to write reports for big brands.
The company wanted a product clients could drive themselves.

If A follows B,
and B follows C,
A might have an interest in C too.

product strategy

Three datasets became one product: a live map of UK demand.

The overhauled Observatory caught on and led to two data deals: Twitter’s first ever, and Royal Mail.
Core data inferred interests; ~20M UK tweets a day refreshed them daily.
Royal Mail deliveries, crossed with location metadata, mapped demand on every UK postcode.

the lie

The map was accurate but useless, so I made it lie, correctly.

Nudging a filter by 8% blanked 90% of the country: London always won.
Clients needed to see where else demand existed, not where people live.
With the Data team I rebalanced the projection: technically wrong, like a subway map, and chosen for the decision it serves.

Accurate

100%top 75%top 50%top 10%

Useful

Most of our clients still ran software from the 90s. When sales walked in with a rough proof of concept, prospects wanted it on sight, price unset, features half-built. So every pass ran function and polish in equal parts, the prototype doubling as the sales deck, sharpening the tool and growing the prospect list before there was anything to buy.

prototyping

3D extrusion and animation made demand readable at a glance.

Extrusion tied height to value, freeing color for secondary cues.
Animation showed interest moving as waves: in and out of London, or spreading across neighboring areas.
Clients committed on the strength of the prototype alone.

feature design

Every feature ran through the map, in both directions.

Clients overlaid their own CRM and catchment files, and dropped pins to define bespoke geofences.
Those uploads ran through a six-state toggle: file upload, processing, cancel, ON, OFF, delete, in one component sized for the small laptops sales reps carried.
Filters fine-tuned an audience, with the summary pane updating live.
Lookalikes ran the logic in reverse: given a market that works, it suggested areas with matching interests and behaviors.

impact

Audiences passed its £1M first-year target on day 91.

The function grew from one solo designer reporting into engineering to a five-person department driving product.
Accurate is a property of the data; helpful is a design decision. Clients paid for the decision.
The rebalanced heatmap, questioned internally as a risk, was among the product’s best received features.