Chain Prompts to Turn KPI Exports into a Board-Ready Risk Memo

What This Builds

A raw KPI export, OTIF, freight cost per unit, inventory turns by category, tells you what changed but not what to do about it, and a single long prompt asking ChatGPT to "summarize and write a memo" tends to produce something that reads confidently while skipping the reasoning a board actually wants to see. This build breaks that one request into four smaller ones, each feeding the next, so you end up with a one-page risk memo you can trace back to the numbers, not a plausible-sounding guess dressed up as analysis.

Prerequisites

• Comfortable with structured, multi-step ChatGPT prompting (Level 3). See the Level 3 guide "Set Up a ChatGPT Project for Negotiation Prep and BATNA Playbooks"
• A Plus subscription ($20/month) for a long enough context window to hold a full KPI export across four prompts in one conversation
• This month's KPI export from the ERP or Power BI, with supplier names replaced by short codes before anything gets pasted in
• Confirmation of which AI tool, if any, your company has approved for board-level content, since these memos often touch margin and customer concentration data

Total ongoing cost: $20/month for the Plus plan. No other subscription is required for this chain.

The Concept

A single long prompt asks ChatGPT to do four different jobs at once: read the data, decide what matters, rank it, and write persuasively. It's easy for the model to skip a step silently and go straight to confident-sounding prose. A prompt chain forces each job into its own turn, and you can see the reasoning at every handoff before it becomes a sentence in a document a board member reads.

Before any of this, replace every supplier name in the export with a code: S1, S2, S3, and so on. Keep the key mapping codes to real names on your own machine, never inside the chat. Board-adjacent material with real supplier names has no business sitting in a chat log if it doesn't need to.

Build It Step by Step

Part 1: Get a clean read on what moved

Open a new ChatGPT conversation. This is a separate, one-off chain, not the negotiation Project from Level 3. Select GPT-5.6 Sol from the model picker, since this chain leans on multi-step reasoning more than fast turnaround.

Prompt 1:

Below is this month's supply chain KPI export. Supplier names have been
replaced with codes (S1, S2, S3, etc). Summarize the three or four biggest
movements versus last month in plain language. For each movement, state the
metric, the direction and size of the change, and which supplier code or
category it relates to if the data shows one. Do not speculate about causes
that are not supported by the numbers below.

[paste the KPI export here]

What you should see: three or four short, factual statements, each tied to a specific number in the export. If it invents a cause the data doesn't support, tell it directly to remove that sentence before moving on.

Part 2: Rank the risk, not just the movement

Prompt 2:

From the movements you just summarized, rank the three biggest risks to the
business, worst first. For each one, explain in two sentences why it ranks
where it does, using only the numbers already given. Note anywhere your
ranking depends on an assumption instead of a stated figure.

What you should see: a ranked list with reasoning attached to each item, not just a reordering of Prompt 1's list. Push back if two risks feel reversed. Ask why, and let it defend or revise the order before you continue.

Part 3: Turn the ranking into a memo

Switch to GPT-5.6 Sol for this step, since it is a writing task rather than a ranking task.

Prompt 3:

Turn the ranked risks above into a one-page memo for the board, written for
readers who have not seen the raw export. Structure: a two-sentence opening
naming the overall trend, then one short section per risk with the risk
stated plainly, the supporting numbers, and a recommended action. Close with
one sentence on what to watch next month. Keep supplier codes, not names, in
the memo text.

What you should see: a memo that reads like something a person wrote under a deadline: direct sentences, no filler, one recommended action per risk. If it hedges every sentence with qualifiers, ask for a tighter, more direct version.

Part 4: Verify before it leaves the chat

Switch back to GPT-5.6 Sol for this last check.

Prompt 4:

Review the memo above against the original export and your own summary and
ranking earlier in this conversation. Flag any number or claim in the memo
that is not directly traceable to the data pasted in the first message.
Also confirm the memo uses only supplier codes, never a real supplier name,
anywhere in the text.

What you should see: either a confirmation that everything traces back cleanly, or a short list of unsupported claims to fix by hand before the memo goes anywhere. Treat this step as mandatory, not optional. It is the difference between a memo you can defend in the room and one that just sounds good.

Real Example: Q3 KPI Review

Setup: A supply chain manager at a mid-market distributor pulls the monthly KPI export, replaces six supplier names with codes S1 through S6, and pastes it into a fresh ChatGPT conversation.

Input: OTIF down 6 points for S3, freight cost per unit up 9 percent overall, inventory turns flat for the top category but down sharply for a secondary one.

Output: Prompt 1 identifies the three movements. Prompt 2 ranks the OTIF drop for S3 as the top risk because it touches the largest customer segment. Prompt 3 produces a one-page memo naming supplier concentration risk on S3 as the headline item with a recommended dual-sourcing review. Prompt 4 flags one sentence in the draft that implied a cause the export didn't support, which gets cut before the memo goes to the CFO for review.

Time saved: This chain replaces roughly an hour of manual synthesis and drafting with 15 to 20 minutes of prompting and review, most of which is spent reading and verifying rather than writing from a blank page.

What to Do When It Breaks

• The memo states a number that isn't in the export → This is exactly what Prompt 4 exists to catch. If it slips through anyway, go back to Prompt 1 and check whether the original summary already contained the invented figure. Fix the earliest step where the error appears, not just the memo text.
• The ranking in Prompt 2 doesn't match your own read of the risk → Ask directly why it ranked things that way, referencing the specific numbers you'd weigh differently. Treat disagreement as a prompt to dig in, not a reason to overrule silently and move on.
• The chain runs long and ChatGPT starts losing track of earlier numbers → Start a fresh conversation with a shorter, cleaner export rather than continuing to paste corrections into an increasingly long thread.
• A real supplier name slips into the export by accident → Stop, delete the message if the interface allows it, and restart the chain with a properly coded version. Don't rely on Prompt 3's instruction to use codes as your only safeguard.

Variations

• Simpler version: Collapse Prompts 1 and 2 into one step for a lower-stakes internal read, keeping the memo draft and verification steps separate.
• Extended version: Add a fifth prompt asking for a one-line executive summary suitable for a meeting invite or Slack message, pulled straight from the finished memo.

What to Do Next

• This week: Run the chain on last month's export as a dry run before using it on live board material.
• This month: Save the four prompts in a personal notes file so you're not retyping them each cycle. A dedicated ChatGPT Project comes later if this becomes a recurring monthly habit.
• Advanced: Compare this month's flagged risks against the prior month's memo to see whether the same risk keeps resurfacing unaddressed.

Advanced guide for Supply Chain Manager professionals. These techniques use more sophisticated AI features that may require paid subscriptions.