ramona-careem.pages.dev / brief
Careem Care serves 70M+ customers and 2.5M+ Captains across 10 countries. The automation program (self-service flows, AI triage, proactive notifications, escalation automation) lives or dies on one weekly question: which contact drivers do we automate next, and which do we leave with people? This tool answers it from the data, in the shape the Care Analytics, Automation and Project Management team could put in front of leadership.
Careem AI challenge, option 1: Decision Brief Generator. Built by Ramona Furter as part of her application for Automation & AI Manager, Care. The dataset is self-created and illustrative; the generator, its ranking logic and its prompt are real and work on your own CSV too.
the data
Illustrative dummy dataset, self-created for this challenge. Eight weeks of Care contact volumes across four verticals (Rides, Food, Quik, Pay) and three markets (UAE, KSA, Jordan), aggregated per contact driver. No Careem data was used or seen.
| week | vertical | market | contact driver | nature | contacts | self-served % | CSAT | FCR % |
|---|
columns: week, vertical, market, driver, nature (scripted|judgment), contacts, self_served_pct, csat, fcr_pct
how it thinks
The tool aggregates the raw rows into a metrics summary, then composes the brief from this prompt's rules. Copy the prompt together with the live metrics of the current scope and paste it into any LLM (ChatGPT, Claude, Gemini): you get the same brief, because the judgment lives in the prompt, not in a black box.
You are the analytics lead of a customer-care organisation. From the metrics JSON below, write a one-page decision brief for the head of Care. Rules: 1. Exactly three takeaways. Each must carry its number from the data. 2. Each takeaway ends in one action, naming the automation lever (self-service flow, AI triage, proactive notification, escalation automation) and the metric that will prove it worked (contact rate, deflection rate, CSAT, FCR, automation accuracy). 3. Rank automation candidates by volume x (1 - current self-service share), and only candidates whose nature is "scripted". A high-volume driver that needs human judgment is not a candidate. 4. Name one thing NOT to automate yet, with the reason. If the lowest-CSAT driver needs judgment, that is your hold: fix the root cause upstream, automate only the status notification, keep people in the conversation. 5. Plain language. No jargon. Under 250 words. It must read as a decision, not a dashboard.
How this was built: designed and shipped by Ramona with her own AI toolchain (Claude Code), the same way she builds automation at Swiss Post and runs Pedal Peak. Around an hour from idea to this page. Everything runs in your browser; nothing is sent anywhere.