Recipe

For Restaurant Operators

How to Check Trends and Seasonality Using AI

Fifteen minutes, two reports, one prompt — and “are we up or down?” stops being a gut feeling and starts being a real answer that accounts for the season.

Every operator asks the same question on Monday — are we up or down? The trap is comparing this month to last month and panicking, when the real story is the season: a January dip isn't a problem if every January dips. The fix is a basic trend health check. Hand AI three numbers — this period, last period, and the same period a year ago — and ask what actually changed once you account for the season. You get the genuine movers separated from the normal rhythm, in plain English, in about fifteen minutes. Works with whatever your POS or accountant exports.

Total time

15 minutes

Difficulty

Easy

Makes

A plain-English read of what really changed vs. last period and last year

Tool

Claude or ChatGPT

Ingredients

• This period's sales or P&L summary (POS export, spreadsheet, or PDF)
• The prior period — last week vs last week, or last month vs last month
• The same period one year ago (this is the one most people skip, and it's the one that catches seasonality)
• 15 minutes

The prompt

You're a sharp restaurant CFO doing a trend health check. I'm attaching three numbers for the same business: the current period, the prior period, and the same period one year ago. In plain English, no jargon: 1) What actually changed versus last period, and for each mover, tell me whether last year shows the same pattern — i.e. is this a real shift or just the normal season? 2) The one number that's moving in a way last year does NOT explain — that's the one worth my attention. 3) One specific question I should answer before reacting. If a swing looks like ordinary seasonality, say so plainly so I don't chase it. If anything looks miscategorized or like a data error, flag it.

Copy

The method

“Are we up or down?” is the most-asked and worst-answered question in a restaurant. Worst-answered because the honest answer is almost always “compared to what?” Sales down 8% from last month means nothing until you know last month was a holiday and this month always softens. The job you're hiring AI for here isn't math — your POS already does the math. It's the compared to what: separating the real movers from the rhythm of the calendar.

1. Pull three exports, not two. This period, the prior period, and — the one that matters — the same period last year. Most POS systems and every accountant can produce a year-ago number; it's the difference between a snapshot and a trend.
2. Keep it summary-level. You don't need every line item; period totals for sales, labor, food cost, and covers are plenty. A clean three-column comparison beats a 40-page dump.
3. Open Claude or ChatGPT, drag the files in, paste the prompt above.
4. Read the answer for the one phrase that matters: “last year shows the same pattern” (relax — it's the season) versus “last year does not explain this” (look here). That single distinction is the whole point of the exercise.
5. Push back before you act. Ask “if I strip out the seasonal piece, what's the real year-over-year change?” or “is the labor jump tracking with the sales jump, or running ahead of it?” It answers from your actual numbers and never tires of the follow-up.

Do it the same morning your weekly or monthly numbers land and it becomes a fifteen-minute ritual: same chat, drag in the new period, one line — “new period attached, same three-way read.”

On the plate

The good draft

Sales are down 9% from last month — but last year shows the exact same March-to-April dip, so that's the season, not a problem. Labor held at 29% both periods and a year ago: steady, ignore. The one to look at: food cost ran 32% this period vs 28% last period AND 28% a year ago — last year does NOT explain that, so it's a real shift. Question before reacting: did a major supplier price change land this month, or is it a single category? Pull the food-cost detail by category before touching the menu.

The bad draft

Sales decreased period-over-period, indicating a potential area for attention. Labor costs remained relatively stable. Food costs experienced an upward trend. Continued monitoring of these key metrics is recommended to ensure optimal financial performance going forward.

Notes from the kitchen

Isn't this just forecasting?

No — and that's deliberate. Forecasting predicts next month, which needs weather, events, and a real model; doing it from three exports would be guessing dressed up as math. This is the opposite and far more reliable: explaining what already happened and whether the season explains it. AI is genuinely good at reading patterns in data you hand it; it's shaky at predicting data it hasn't seen. This recipe stays on the side it's good at.

What if I don't have last year's number?

Two columns still work — the prompt runs on this-period-vs-last-period alone. You just lose the seasonality check, which is the most valuable part. If you can get even a rough year-ago figure from your POS or accountant, it's worth the five minutes.

My periods aren't clean — different number of days, a holiday fell differently.

Tell it. Add a line: “this period had 31 days vs 28 last period, and Easter fell in this period last year.” It will normalize the comparison and call out the calendar quirks instead of being fooled by them — which is exactly what a good analyst does.

Does this work across multiple locations?

Drag in each location's three-way set and add: “do this per location, then tell me which store's trend is the real outlier.” That works until the exports stop fitting in a chat window — when you're consolidating across enough stores that this becomes a weekly chore, that's the problem Expo was built for: every store's trend, every period, called out automatically.