AI Coffee Chat: Quick Wins for Improving Processes-Presentation

Presentation - Durham University

Pdf Summary

Durham University used Copilot to help redefine and update campaign data as part of a move from one broad institutional campaign record to three more specific campaign pillars.<br /><br />First, the team drafted new campaign definitions from internal documents and used Copilot as a sounding board to check whether the definitions covered all bases and avoided gaps. Colleagues then reviewed the final version.<br /><br />The bigger challenge was adapting existing gift data. Campaign records were a mix of the old institutional campaign and the three new pillars, so all gifts within the campaign period needed to be reassigned. A key source was the gift attribute “specifically for,” which contained 20,056 rows of free text, with about 10% unique values. Because the data was highly varied and messy, the team asked Copilot to help categorise it.<br /><br />At first, the results were poor: copied data trailed off, blanks were converted into “nan,” and the outputs were not reliable. After refining the prompts—adding campaign names, brief definitions, keyword-based instructions, and clear rules such as leaving blank cells untouched and flagging uncategorisable entries as “Review”—Copilot produced much better, strictly categorised results.<br /><br />The main outcomes were that Copilot helped with defining the new campaigns and updating gift data to match them. The key lesson was that Copilot works best as a mirror: the quality of its output depends heavily on the clarity of the prompt. Prompting was compared to learning a new language using words you already know.

Keywords

Durham University

Copilot

campaign data

campaign definitions

gift data

data categorisation

prompt engineering

free text

campaign pillars

data migration