Analyst briefings cut from six hours to twenty minutes

The Challenge

A team of analysts was responsible for knowing what was being said about a dozen policy areas across national and regional media, broadcast, and social platforms — in four languages. In practice that meant a morning of manual scanning before anyone could write a single line of analysis, and by the time a daily digest was circulated it was already describing yesterday. Anything that broke in the evening was found the next morning, if at all.

Off-the-shelf media monitoring tools covered the collection problem but not the judgement problem. They returned volume, not relevance, and none of them understood the directorate’s own taxonomy of topics, entities and thresholds.

The Solution

We built a continuous pipeline that collects from every source the team cared about, scores each item against their taxonomy, and only then applies analysis. Relevance filtering happens before anything expensive runs, which is what makes monitoring at this volume affordable.

Analysts get two surfaces: a ranked digest generated on their schedule, and threshold alerts when a tracked topic moves faster than its own baseline. Every claim in a briefing links back to the source item, so nothing has to be taken on trust — a requirement that shaped the architecture from the first week rather than being retrofitted.

The system runs inside the directorate’s own environment. Analysts adjust the taxonomy themselves, without a release cycle, which is the difference between a tool that stays accurate and one that quietly drifts out of date.