Prompt of the day

New database sanity-check: before you trust a single number

A new data delivery, a restatement, or a changed market definition just landed. Run the QC gauntlet before anyone builds a story on it.

You are a syndicated data steward. I will describe my database setup and you will produce a NUMBERED QC CHECKLIST customized to it — each item with a pass/fail criterion and what a failure implies. Cover at least:

1. Totals vs known benchmarks (shipments, prior deliveries, retailer portals)
2. Hierarchy double-count traps (brand vs manufacturer vs total category rollups)
3. Distribution consistency (%ACV vs TDP moving together or diverging suspiciously)
4. Restatement differences vs the prior delivery and where they concentrate
5. Coverage gaps (retailers or channels missing from the market definition)
6. Item coding: new UPCs mapped, discontinued items still reporting, private label placement

My setup: [DESCRIBE: data source, channels covered, market definitions, time periods, measures you use most]

Order the checklist by how badly a failure would corrupt downstream analysis, worst first.

Rules: Do not invent, estimate, or extrapolate any figure — if a number is not in the data I give you, write "not provided" and flag it. Mark every claim I should verify against my syndicated data or internal reporting before using it externally. Never include retailer-confidential terms or personally identifiable shopper data.

Why it works — Every experienced analyst has shipped a conclusion built on a restated market or a double-counted hierarchy. A setup-specific QC checklist, ordered by blast radius, catches the wreck before the deck exists — which is worth more than any analysis prompt in this brief.

Build a reusable 'about me' block you paste into every chat

The single biggest quality jump for a busy professional isn't a fancy prompt — it's giving the model who you are and who you're writing for, every time, for free.

Write one paragraph: your role, your company, your audience, your tone, and what a great result looks like. Save it. Paste it at the top of any AI task so you never re-explain yourself.

1. Draft 4-6 lines: role, company/industry, audience, tone, 'good = ...'.
2. Keep it in a note you can paste in 2 seconds.
3. Lead every AI task with it.