Teaching a client to use Claude in Excel for data reporting
A client came to us with a familiar problem. Their reporting workflow was manual end to end: pull data, write formulas, format columns, draft a summary, hope nothing changed before the meeting. Every step was where mistakes entered, and every step required someone who knew Excel well enough to get it right. They wanted to change that without changing their tooling — they were already in Excel, and they wanted to stay there.
Claude in Excel (via the Microsoft 365 Copilot integration or the Claude.ai web interface with pasted data) is a practical answer to this. The challenge is not the tool — it is the habit shift. Business users who have spent years thinking in functions and cell references do not automatically switch to asking questions in plain English. The training has to undo that muscle memory before it can install something better.
The two mindset shifts
Before any hands-on work, we spent time on the underlying shift the session was trying to create.
Ask questions, not functions. The first instinct when reaching for a formula is to think about syntax — VLOOKUP, INDEX/MATCH, SUMIFS. With Claude, you describe the outcome you want: "pull the customer name from this sheet using the order ID." You manage the question; Claude picks the right tool. The moment users internalise this, the scope of what they can ask for expands immediately.
Use Claude as a second pair of eyes, not just a builder. Once a report is built, Claude is as useful as a checker as it is a constructor. Ask it to sanity-check the totals, flag anomalies, or explain why a trend happened rather than just charting it. That is the difference between a report that lists numbers and one that tells a story.
The five-stage training structure
The session was structured as a progression of five stages, each building on the last. All work was done on a real (or representative) dataset, not a demo file — the goal was a repeatable workflow the participant could take back on Monday morning.
Stage 1 — Get comfortable asking, not building. Open a real spreadsheet. Ask Claude to summarise what is in the sheet: column meanings, row count, obvious gaps. Ask it to clean something small — fix inconsistent date formats, trim whitespace, standardise text casing. The aim is to make the act of asking feel natural before anything is at stake.
Stage 2 — Formulas without formula-writing. Ask for a calculated column in plain English. Ask Claude to explain the formula it produced, so the user can read it back. Ask for a lookup across two sheets without naming a function. The goal is to decouple capability from syntax knowledge.
Stage 3 — Reporting and summarising. Build a pivot-style summary: totals by region, by month, by product category. Ask Claude to identify the top and bottom performers. Ask for a written executive summary paragraph based on the data, ready to paste into an email. This is where the time saving becomes visible.
Stage 4 — Analysis and judgment. Ask Claude to flag anomalies and give possible explanations. Ask for a month-on-month comparison with commentary on what changed. Ask Claude to suggest and build one chart that best tells the story. This stage moves from reporting to reasoning.
Stage 5 — Make it repeatable. Save the successful prompts as a personal prompt library. Re-run the same ask on a refreshed dataset to test consistency. Identify one weekly or monthly report to fully hand off to this workflow going forward. The session ends with a concrete commitment, not a general intention.
Training materials
We produced a complete set of materials for the session: a ten-slide deck and a single-page quick-reference handout. Both are available to download below.
The deck walks through the two mindset shifts and all five stages, with speaker notes on every slide. The handout distils the five stages into a reference card the participant keeps at their desk — a grid for stages one through four, a full-width stage five banner, and a checklist for the end of the session: the one report to hand off this week, two or three prompts worth saving, and one open question to bring back.