BBC — Turning scattered data into clearer commissioning decisions

Context

Commissioners had data, but no map

“I’m in a sea of data and I don’t have a map.”

Audience, content and production information existed across different tools and teams. Commissioners could see fragments, but not the complete picture.

This made practical questions difficult to answer: which topics interested younger audiences, where the BBC was underserved by age or region, whether content was being duplicated, and whether an audience need required a new commission or could be met through existing content.

One content group as it arrived from Data Science, repeated a few hundred times down the page. Everything a commissioner needed was in here somewhere, which is exactly the problem.

The value was not more data. It was greater confidence in the next editorial decision.

A wall-sized workflow map of the popular factual commissioning process, divided into columns for pre-commissioning, commissioning, production, scheduling and reporting, each filled with colour-coded sticky notes for tasks, people, pain points and data. — We mapped the commissioning process end to end before designing anything. The columns are stages; the colours separate what people did, what they needed to know and where the process broke down.

A promising prototype lacked a clear workflow

Before I joined, the team had built a prototype for the Controller of BBC Three. It used performance data from existing programmes to highlight opportunities for new content.

Testing produced mixed feedback. People understood the idea, but were unclear who the product was for or how it would help them day to day.

I was brought in to lead a UX team of three and help turn the concept into a production MVP that could serve more than one commissioning team.

The prototype home screen, headed Welcome to the BBC Three commissioning support tool, asking the user to select an audience need to explore from five cards: feel good about myself, step off from the world and wind down, build a positive future for myself, show me a world beyond my own, and revel in life itself no matter how extreme. — The prototype opened by asking which of five BBC Three audience needs you wanted to explore. A fair question for one channel, and the wrong first question for everybody else.

Approach

From showing data to supporting decisions

Pre-planning widened the product’s purpose

My first move was to shift the brief from “show me data” to “help me decide what to commission, reuse or prioritise.”

Research revealed a missing stage before commissioning: pre-planning. Teams often needed to understand whether an audience need could be met by curating, resurfacing or scheduling existing content before creating something new.

This made the product relevant beyond BBC Three. It could also support iPlayer teams building curations and schedulers looking for archive content to rebroadcast.

The tool was no longer only about finding opportunities for new commissions. It supported the earlier decision about whether a new commission was needed at all.

Pre-planning sat upstream of the existing content pipeline, feeding the Plan and Create stages with audience and production data.

Research revealed what to simplify

Earlier discovery had identified broad pain points, but not what information people needed at each stage of their workflow. I made the case for further research before the team committed to the MVP structure.

To move quickly, I led paired card-sorting sessions with 12 content creators from different parts of the BBC. Pairing participants helped us cover more workflows and exposed where needs overlapped across commissioning, scheduling and editorial planning.

The research changed four parts of the product:

Journey simplified so users could reach decision-relevant information faster.
Filters prioritised age and region because they appeared consistently across sessions.
Groupings broadened to match editorial planning language.
Naming improved so users could understand what each content cluster represented.

The design work was deciding what to simplify, what to expose and what the system needed to explain.

Card sort board one, commissioning for a target audience. Information cards placed along a high-value to low-value scale, with age group, appetite for content and estimated opportunity at the high-value end. — Four commissioning scenarios, sorted by 12 content creators against the same high-value to low-value scale. Age and region held their position across every board, which is why they became the primary filters.

Card sort board two, placed along the same high-value to low-value scale by a different pair of participants. — Four commissioning scenarios, sorted by 12 content creators against the same high-value to low-value scale. Age and region held their position across every board, which is why they became the primary filters.

The top of the scoring matrix. Information types run down the left with a weighted score beside them: age group 170, region 160, estimated opportunity 150, appetite for content 150, then social class, ethnicity, gender and Acorn. Each scenario is a column scored high, medium, low or not applicable. — The boards turned into one matrix: every piece of information anyone asked for, scored against every scenario and weighted by how consistently it came up. Age group and region finished first and second with a gap behind them, which is what ended the filter argument. The second slide is the bottom of the same sheet, and it decided just as much — three quarters of the list was wanted once, by one pair, and stayed out of the MVP.

The bottom of the same matrix. Rows including risk taking, retention, reputation, production team and talent, and global trend score five or ten and read not applicable across almost every scenario. — The boards turned into one matrix: every piece of information anyone asked for, scored against every scenario and weighted by how consistently it came up. Age group and region finished first and second with a gap behind them, which is what ended the filter argument. The second slide is the bottom of the same sheet, and it decided just as much — three quarters of the list was wanted once, by one pair, and stayed out of the MVP.

Better data made the information architecture harder

The prototype relied on BARB data from linear broadcasts, manually mapped to content groups. For the MVP, we added daily iPlayer data from signed-in users to better represent younger audience behaviour.

This made the product more useful, but also made its information architecture more complex. The system generated more than 300 content groups that changed each day.

I worked with Information Architecture and Data Science to introduce sub-genres and demographic information into the naming logic. This produced unique, meaningful names for 95% of the clusters.

The interface and data model could not be designed separately. If the groups were unclear, the product would remain difficult to use regardless of how the screens were organised.

The BARB programming cluster model, drawn as a branching hierarchy. Each cluster is a card listing its top programmes, top genres, audience share and total reach, splitting into further levels below. — The BARB cluster model the prototype inherited: a fixed hierarchy, updated on a slow cycle. The iPlayer data we added behaved differently, so the naming logic had to work for both.

A spreadsheet comparing eleven naming strategies for content clusters. Columns show a sample name, average and maximum characters per name, the most duplicated name, the number of uniquely and non-uniquely named clusters, and the percentage of clusters uniquely named, colour-coded from red to green. — Eleven ways of building a cluster name, each scored on length, duplication and how many of the groups it could name uniquely. Working this out as data rather than as copy is what made the naming defensible to Data Science.

Two wireframes side by side, labelled BARB and iPlayer. Both block out the same content group detail page, with modules colour-coded: identification in green, opportunity in pink and demographics in blue. The BARB version carries a BBC share module and a Mosaic index where the iPlayer version carries activation and Acorn. — The detail page blocked out twice, once for each data source, with modules colour-coded by the job they did. The second slide is the same pair after review, with the decisions written onto it: ethnicity dropped, one region index instead of several, and an open question about whether top genres existed only to generate the name.

The same two wireframes after review, in grey, with sticky notes attached. Notes read: top three genres, are we using them just for the generation of titles; all figures; got rid of ethnicity; only one; and different age brackets, why. — The detail page blocked out twice, once for each data source, with modules colour-coded by the job they did. The second slide is the same pair after review, with the decisions written onto it: ethnicity dropped, one region index instead of several, and an open question about whether top genres existed only to generate the name.

We organised the tool around commissioning questions

I used a “Now, Next, Later” structure to move the experience away from raw data exploration:

Now — what do current audience and content patterns show?
Next — where is there a gap or opportunity?
Later — what might require commissioning, curation or scheduling attention?

The MVP allowed users to explore patterns by age and region, understand content groups in editorial language, and compare audience opportunities with existing programmes.

I worked with Product to broaden the use case, with Data Science and Engineering to understand what could update dynamically, and with Information Architecture to make the results meaningful to non-technical users.

My role was to keep those decisions anchored to the editorial workflow rather than the structure of the underlying data.

An early content group list where each group is labelled with a generated name such as Moderate Factual 29 or Large Factual 36, followed by its programme titles run together as a paragraph. — The content group list, reworked. The generated names came first, then the programme titles that let people recognise a group at a glance, then a pass against the GEL components and the column grid.

A later version of the list where programme titles are broken out as a bulleted list under each group name, with audience reach, existing content and estimated opportunity aligned to the right. — The content group list, reworked. The generated names came first, then the programme titles that let people recognise a group at a glance, then a pass against the GEL components and the column grid.

The names had to survive the layout

Naming the clusters well made them longer. A single group could run to three lines, and it still had to sit in a table row next to three metrics without pushing them out of reach.

I tested the group heading at three sizes from the BBC’s type scale against the longest names the system could generate, rather than against sample text that happened to fit.

A grid of the content group detail page repeated at three BBC GEL type sizes, labelled Trafalgar, Paragon and Double Pica, each shown with four different lengths of group name and different amounts of supporting detail. — The same heading at Trafalgar, Paragon and Double Pica, against four lengths of generated name. Double Pica held the longest names in two lines without crowding the metrics beside them.

The prototype asked people to pick a pre-combined audience, from a list of eleven compound options.

The filters became a sentence people could read, with age and region as separate, plainly named controls.

Results

From a mixed prototype to a production MVP

The Commissioning Support Tool moved into production with a broader BBC-wide purpose than the original BBC Three prototype.

300+

Content groupsRebuilt daily from BARB and signed-in iPlayer data

95%

Uniquely named clustersUp from generated labels people could not tell apart

Content creatorsPaired card sorting across commissioning, scheduling and editorial planning

Teams servedCommissioning, curation and scheduling, from a BBC Three-only prototype

The shipped content group list filtered to 16 to 24 year olds. Groups carry editorial names such as Documentaries: Human interest and Documentaries: Factual for 16-24s in London, each tagged BARB or iPlayer, listing its top programmes with audience reach, existing content and estimated opportunity. — The list people ended up using. Every group is named in language an editorial team already has, and tagged with the data source behind it.

The work did not have a public adoption metric, so the strongest evidenced outcome is the product and organisational change: a mixed-feedback prototype became a production MVP with clearer foundations for wider BBC use.

Reflection

Looking back, the hardest part was not the data, but agreeing what the product was for. Widening the brief from one BBC Three prototype to a tool for commissioning, curation and scheduling meant the interface and the data model had to be redesigned together, not handed from one team to the next. That reframing is what let a mixed-feedback prototype become something the rest of the BBC could actually use.