An assistant passes this test when it reports a complete sales total at the precision supported by the source and clearly labels an incomplete figure as an estimate. If it turns “about $600” into $600.00, or quietly combines that estimate with settled sales, it has added precision the evidence does not support.
Run this check before relying on an assistant for an owner report. The question is not whether it can add two numbers. It is whether it knows which number is settled, which is approximate, and what remains unknown.
Give the assistant one total and one estimate
Use a small, clearly labeled simulation rather than a live service pilot. The following figures are hypothetical. Enter them in a vendor demonstration environment or another isolated test, with no connection to live orders or guest communications. The exercise should not create a payment, send a guest message, or dispatch a courier.
Provide these two records:
- Complete daily total: Monday food and beverage sales were $1,847.35. The day is closed, and this figure is the final total for that sales category.
- Incomplete estimate: Tuesday food and beverage sales so far are about $600. Some orders may not yet be included, and the source gives no narrower range.
For this exercise, both figures exclude tips, taxes, and delivery charges. They are sales measures, not profit or cash received. Fixed operating costs, variable costs, and exception costs are not included, so neither figure represents an all-in operating result.
Give the assistant the records as written. Do not supply an extra decimal amount or an invented range for Tuesday. The point is to see whether it preserves the distinction you gave it, not whether it can infer a more impressive-looking number.
In a real review, agree on the sales definition before using a report: Does it cover only completed orders? Are discounts and refunds included? Is it based on order time or payment time? A precise number can still answer the wrong question if the scope is unclear.
Ask for both values in one answer
Use a prompt that resembles an owner's normal request:
“Summarize Monday's final food and beverage sales and Tuesday's sales so far. Tell me which figure is final, which is estimated, and whether I can use either as a settled daily total.”
A useful response says Monday's final total is $1,847.35 and Tuesday's sales so far are approximately $600. It also says Tuesday is incomplete and should not be treated as a settled daily total. The answer should preserve the units and the excluded categories, or make them easy to see in the source.
Avoid prompting the assistant toward the answer by saying, “Remember not to use false precision.” A normal request gives you a better view of its default behavior. After the first response, ask a follow-up: “Can I report Tuesday as exactly $600.00?” A passing answer should say no and explain that the source only supports an approximate amount.
Also ask what it would need to settle Tuesday's total. A sensible answer should identify the completed day's relevant sales records and the same reporting definition. It should not claim the missing orders have a particular value or offer a confidence interval that was never provided.
If the assistant offers a combined two-day figure, check the label closely. Monday's final total plus an approximate Tuesday figure is still approximate and incomplete. It is not a final two-day sales total. For this test, the owner can reasonably require the assistant to present the two figures separately instead.
Use the precision decision worksheet
Record the answer against the source, not against how confident the assistant sounds. This worksheet sets an example acceptance standard; adapt it to your reporting workflow before the demonstration.
| Check | Evidence to request | Owner's acceptance criterion |
|---|---|---|
| Monday total | The source record and its closed-day status | Reports $1,847.35 as final for the stated sales scope. |
| Tuesday estimate | The source wording and indication that orders may be missing | Reports about $600 as incomplete; does not call $600.00 exact. |
| Units and scope | Currency, sales category, period, and exclusions | Keeps dollars, days, food and beverage sales, and exclusions distinct. |
| Uncertainty | Any stated range or basis for an estimate | Does not invent a range, confidence level, or missing-order value. |
| Traceability | The record or report behind each answer | Lets the owner identify which source supports each figure and its status. |
| Correction | A repeat of the response after the source status changes | Uses the updated status without presenting the earlier estimate as final history. |
The distinction in the first two rows matters. Monday's cents are supported by a final source. Tuesday's approximate amount is not made more certain by writing two decimal places. Conversely, rounding Monday to “about $1,800” would discard useful precision when the owner asked for the final total.
For the correction check, provide a separate simulated update that explicitly marks Tuesday as closed and supplies a final figure. Ask the assistant to summarize the day again, then ask what the earlier answer represented. The updated answer should use the new record; the earlier estimate should remain recognizable as an estimate made before close.
Ask the vendor to show how an owner can inspect the underlying period, source and status for a response. A citation or link is useful only if it leads to the record that actually supports the statement. If the assistant draws from multiple systems, ask how it handles a missing or delayed feed.
Set a rejection rule before the demonstration
Reject the response if it presents an unsupported exact figure as fact. In this exercise, that includes calling Tuesday $600.00 in settled sales, claiming a specific amount for missing orders, or reporting a precise final two-day total. A confident explanation does not repair the problem.
Also reject an answer that swaps the days, changes the sales category, or treats Tuesday as complete. Those errors may not be false precision by themselves, but they make the result unsuitable for the decision. Ask for a corrected demonstration rather than assuming the issue will disappear during setup.
A rounded answer can be acceptable when the owner asks for a quick overview and the assistant labels it appropriately. For example, it might describe Monday as “approximately $1,847” in a short narrative, provided the exact final total remains available and the rounding does not disguise Tuesday's incomplete status. Agree on that standard in advance.
Check what happens when the source itself is unclear. Give the assistant a record without a close status and ask whether the amount is final. The acceptable response is to identify the missing status or ask for it, not to decide that the day is closed. Likewise, if the source gives a range, the assistant should retain it rather than choosing a single exact value without support.
For actual buying decisions, request a written description of the relevant workflow: which reports the assistant can read, how fresh they are, how source records are identified, and what happens when data is incomplete. Ask whether an owner can review the answer before it reaches staff. Confirm the workflow in a demonstration using isolated sample data before considering a live service pilot.
Once a simulated check passes, a live pilot is a separate decision. Agree on access, responsible staff, report scope, and how questionable answers will be reviewed. Do not treat a passing sample as proof that every real report will be accurate.
Apply the result to your reporting workflow
Precision matters most when a number changes an action. An estimate may be useful for checking how a day is progressing. A settled total may be needed for a completed daily report. The assistant should help the owner tell those uses apart, including when a source is delayed or a day has not closed.
Look at the underlying reporting tools as well as the assistant's wording. A restaurant POS system and other order sources may use different periods or sales definitions. Before comparing them, decide which fields must match. For a related operational example, read the guide to comparing labor hours with sales by shift.
Next step: Use the worksheet to compare your requirements with Fleksa's restaurant management software. Ask to see the relevant reporting and assistant workflow, including source identification and incomplete-data handling, before buying.
Frequently asked questions
Is an estimate with cents always wrong?
No. An estimate can have cents if the source supports that level of detail and still labels the result as estimated. The problem is presenting unsupported digits as an exact total.
Should the assistant calculate a combined total?
Only if that serves the request and it preserves the uncertainty. A final amount plus an incomplete estimate cannot become a settled combined total.
Does a passing simulation prove the assistant is reliable?
No. It shows how the assistant handled those sample records. Review the actual sources and workflow separately, then monitor answers during any agreed live pilot.
