Year
2023
Client
David Piagure
Category
Business
Product Duration
3 - 4 Weeks
We spent a week observing the morning reporting process and interviewed the staff doing it and the senior leaders receiving it. The recipients wanted three things: on-time delivery stats, flagged exceptions, and a short plain-language summary of anything that needed attention. The staff producing it spent most of their time on data extraction and formatting, with the actual writing taking only 10 to 15 minutes once the numbers were in front of them.
We designed the automated report to mirror the format the senior team was already used to, so there would be no adjustment period. The structure included a headline metrics block, an exceptions section that surfaced anything outside tolerance thresholds, and a two-paragraph narrative summary written in plain language. We designed the narrative prompt to match the tone and structure of the best-written manual reports we had seen during research.
We built a scheduled pipeline that ran at 6:45am each morning, querying APIs from all five source systems, normalising the data, running threshold checks to flag exceptions, and passing the structured output to a language model to generate the narrative. The finished report was formatted in HTML and distributed by email at 7am. We built a monitoring layer that alerted the ops team if any data source failed to return within the expected window, with a manual fallback process documented for rare failures.
12 hours per week reclaimed across the ops team
