It’s 9:00 a.m. A customer wants 24 desk lamps for Friday. The dashboard says 32 are available and shows a refresh at 8:58. Your AI draft is ready to send. “Yes, we have 24 in stock.” Would you click Send or open the warehouse record first?
Open the warehouse record. The dashboard refreshed two minutes ago, but the stock count came from a snapshot taken three hours earlier. That difference changes what you can promise.
Follow the number back to the warehouse
In this fictional example, all times are UTC on the same day. The count is for the same lamp SKU at the East warehouse, and it describes sellable units at the time of the snapshot.
The 08:56 CSV export and the 08:58 dashboard both show 32 because they use the same 06:00 snapshot. The later timestamps tell you when that number was packaged and displayed. They give you one stock observation, even though the number appears in two places.
Decide how recent the count needs to be
For this example, the team requires a source count no more than 30 minutes old before it will promise stock. At 09:00 UTC, a qualifying count would be from 08:30 UTC or later. The 06:00 count is three hours old, so the draft goes on hold.
Set the cutoff with whoever owns stock commitments. An older count can still be useful for a weekly trend report. A customer promise can require a fresh count, a check for new holds or reservations, and a reservation for the order if your process calls for one. Verify the delivery date separately.
Before. “Yes, we have 24 lamps in stock for Friday.”
After. “I’ll check whether we can allocate 24 lamps and confirm the earliest delivery date.”
For the internal handoff, keep the reason visible. The East warehouse snapshot showed 32 sellable units at 06:00 UTC. The dashboard still displayed that snapshot at 08:58 UTC. At 09:00 UTC, the count was beyond the example team’s 30-minute limit.
Ask AI to show its source before it drafts
Paste only the fields you’re allowed to share. The AI can organize the supplied evidence and draft a reply. A person with warehouse access still needs to verify live availability and make any reservation.
Open the copy-ready stock check prompt
Help me review a stock request before I send a customer reply. Use only the records I provide. The decision time is [date, time and time zone]. The maximum age for a stock observation is [minutes]. The requested SKU, warehouse and quantity are [details].
For each record, return its ID, SKU, warehouse, sellable quantity, timestamp and what the timestamp describes. Separate the source stock observation from file generation and dashboard refresh. Flag records that share the same source snapshot.
If the source timestamp is unverified, missing, has an unclear time zone, is in the future or is older than the cutoff, mark the decision HOLD. Also mark HOLD if the scope does not match, availability is insufficient or the records conflict without a verified explanation. Give the next check and draft a reply that leaves stock and delivery unconfirmed.
If the count meets the cutoff and covers the request, mark it READY FOR HUMAN REVIEW. Include the source ID and as-of time. Leave allocation and delivery for the person who can verify them.
Records to review: [paste approved excerpts here].
Run one quick check before you reuse it
Try changing the dashboard refresh to 08:59 UTC while leaving the source snapshot at 06:00 UTC. The decision should still be HOLD. If the draft flips to “in stock,” check which timestamp it used.
For another example of checking what a timestamp actually records, read our guide to reviewing AI-written incident recaps.
Sources
Background reading checked November 10, 2026. The stock figures, cutoff and replies above are fictional examples written for this guide. Artwork by WorkTool Note.