The manager of a three-location restaurant group approves the weekend prep plan. A new AI forecast predicts a rush, so the kitchens prepare 180 extra portions. By Sunday night, 46 remain unsold.
The obvious conclusion is that the model failed. The more important question is whether the business ever defined what success meant, which data the forecast could trust, when a manager should override it, or how waste and stockouts would be measured together.
The restaurant is illustrative. The questions are the ones every team must answer before an impressive AI demo becomes part of a real operating day.
So, is your business ready for AI? The answer depends less on how impressive the demonstration looks and more on whether your business can answer seven practical questions.
Quick answer: Your business is ready to test an AI idea when you can explain the problem, how the work happens today, what information AI will use, what it must never do, how you will judge it, where it fits into daily work, and who will keep checking it. Any missing answer is the next thing to fix.
The weekend forecast reveals what “ready for AI” really means
Consider the restaurant as an illustrative example. The team wants AI to reduce food waste without increasing stockouts. It combines sales history, reservations, promotions, weather, and local events. The demo looks convincing because the model predicts last month's demand with impressive accuracy.
Real operations expose the gaps. Promotions were recorded differently by each location. A local festival was missing from the data. Managers received a number without a confidence range. Nobody knew whether the system was advising purchasing, prep quantities, or staffing. The team had an AI model, but it did not yet have an operating decision.
That is the starting point for an AI business check. The check does not ask whether AI is exciting. It asks whether the restaurant can turn a forecast into a safer, measurable, and repeatable action.
What should this AI business check help you decide?
This is not a pass-or-fail certificate for the whole company. A business can be ready to use AI to sort invoices and unready to let it draft regulated advice. Being ready depends on one business problem, the people affected, and the daily work around it.
This matters because the same AI tool can be helpful in one situation and harmful in another. The NIST AI Risk Management Framework starts with purpose, users, context, possible effects, and limits. In business language: understand where and why AI will be used before comparing tools.
NIST notes that AI RMF 1.0 is under revision. Use it as a practical reference point, then confirm the current version and any rules that apply to your industry before making a formal decision.
The seven questions should lead to one of three decisions:
- Try the idea in a small, controlled setting.
- Run a smaller test to answer the questions that remain unclear.
- Fix the work process or information first, then ask the seven questions again.
Seven questions to answer before you invest in AI
Use these questions for one business problem. Give each answer 0 if you do not know, 1 if it is partly clear, or 2 if it is clear and supported by evidence. This is a Xyric planning guide, not an industry certification.
Here is the complete business check:
- The problem: What business decision or action should improve, and who is responsible for it?
- Today: What happens now, and what does it cost in time, mistakes, money, or lost service?
- The information: Do you have useful, reliable information that you are allowed to use?
- The limits: What must AI never do, and who can stop it?
- The proof: Which real examples will show whether it works well enough?
- The daily work: Where will the AI suggestion appear, and what happens when it is unavailable?
- The responsibility: Who will keep checking the results and respond when something goes wrong?
The stop signal: A high total score does not matter if you do not know what AI must never do, whether you may use the information, or who remains responsible. Treat any critical zero as the next problem to solve.
An AI business check becomes useful when every answer points to something real. “The operations team will use it” is a hope. “Two claims analysts will review each suggestion inside the existing queue during a four-week test” is a working plan.
Question 1: What business decision should improve?
Write one sentence: “When this happens, this person uses this information to take this action.” If you cannot complete the sentence, stop. A chatbot, model, or AI tool has no business value unless it helps someone make a better decision or take a better action.
For the restaurant, the decision can be stated clearly: at 4 p.m. each day, the kitchen manager uses the demand range to approve tomorrow's prep quantity for each perishable item. That is testable. A vague goal to use AI for planning is not.
Google's Rules of Machine Learning recommends deciding what success means before building and starting with a simple comparison. The same discipline applies to generative AI. Write down how the work happens today. Then compare the proposed AI with something real.
Question 2: How does the work happen today?
Record what happens before AI enters the process. That starting picture may include handling time, completion rate, correction rate, backlog age, money recovered, or another measure tied to the decision.
Do not count “number of AI responses” as business value. It is activity. Pair speed or volume with a real result and a safety check. A customer support assistant, for example, might be judged on resolution time, correct handoffs to people, and whether reviewers accept its answers. One number alone can reward the wrong behavior.
In the restaurant story, the starting picture includes waste by item, stockouts, emergency purchases, and manager planning time. Reducing waste while doubling stockouts would not be a successful test. The business check keeps both the desired result and the safety limit visible.
Question 3: Does AI have the right information?
Ask where the information comes from, who owns it, how current it must be, whether it reflects real situations, and whether the business is allowed to use it. The business check should reveal missing information and unclear definitions before anyone builds around them.
More information is not automatically better. A smaller set of reviewed examples can be more useful than a large archive with unclear labels, changing definitions, or missing cases. This question is answered when the team can explain where the information came from and what it does not show.
The restaurant therefore checks whether reservations, delivery orders, promotions, closures, and local events are captured consistently across locations. More sales rows cannot compensate for a missing explanation of why demand changed.
Question 4: What must AI never do?
Safety is not a legal paragraph added at the end. Identify the people affected, sensitive information, ways the tool could be misused, answers that are never acceptable, and the person who can pause it.
NIST treats AI safety as an ongoing responsibility. Its AI RMF Core calls for clear roles, testing before use, regular checks, and planned responses. Turn that into rules a small trial can use: who may access it, when a person must approve the answer, which answers are never allowed, how problems are recorded, and how people work without it.
For the restaurant, the forecast should advise rather than silently place orders. A manager can override it, see which information is missing, and return to the existing prep method when the recommendation falls outside an agreed range. That turns a vague promise of human review into a clear business rule.
Question 5: How will you know it works?
Collect a small set of realistic examples and agree on how people will judge the answers. Record what counts as correct, useful, incomplete, unsafe, or uncertain. If reviewers disagree, make the rules clearer or narrow the business problem.
The score that matters is not the one shown in a software demonstration. It is how well AI performs on your work, your information, and your safety limits. Keep the examples where it fails and test them again whenever the tool or instructions change.
For the demand forecast, the examples should include ordinary weekdays, holidays, promotions, bad weather, and unusual local events. Reviewers should judge waste, stockouts, and the usefulness of the forecast range, not only the average accuracy score.
Question 6: Where will AI fit into daily work?
A working demonstration is not a working business process. Show how a request reaches the AI, where its answer appears, who reviews it, and what people should do when the tool or one of its information sources fails.
The restaurant could place the recommendation inside the existing prep sheet rather than asking managers to open a separate AI dashboard. The same screen should show the suggested range, the reason for unusual demand, and the manual fallback when fresh data is unavailable.
Google's MLOps guidance notes that the AI model is only a small part of a live system. Checking information, testing changes, delivering answers, watching performance, and managing the process all sit around it. Include those operating parts before calling the idea ready for everyday use.
Question 7: Who will keep checking it?
Decide who reviews the results, how often the review happens, and what warning forces a change. The final question asks whether the team can notice outdated information, falling quality, frequent manager overrides, broken rules, slow responses, and unexpected use.
Watching results without a responsible person is just another dashboard. Name the warning, the person who receives it, the first action, and how the team returns to the previous way of working if necessary.
In the restaurant example, a weekly review compares recommendations, overrides, waste, stockouts, and missing information by location. If forecasts become less accurate or managers repeatedly reject one category, an owner investigates before the system quietly shapes another weekend.
How to score the seven-question AI check
Add the seven scores. Treat the ranges as planning guidance:
- 0 to 5: The idea is not ready. Clarify the problem, how work happens today, and the information available.
- 6 to 10: Run a small test. Use it to answer the weakest questions.
- 11 to 14: Consider a small real-world trial, with specialist, legal, security, or industry review where needed.
The total can hide a critical zero. A high score does not cancel an unknown safety limit or information you are not allowed to use. Record each answer, keep the supporting evidence, and repeat the check when the business problem or working environment changes.
Turn the answers into the next action
The result of the AI business check should fit on one page: the problem, how work happens today, seven answers, supporting evidence, important unknowns, the responsible person, and the next small test. The check is complete when that page tells the team what to do next and what must be true before it spends more.
Return to the restaurant. If the information question scores zero because promotions and events are unreliable, the next move is not a larger model. It is a short information cleanup and a small trial on one category. The story moves from an impressive demo to a business test that can actually teach the team something.
Frequently asked questions about getting a business ready for AI
What should a business check before using AI?
A business should check seven things: the problem AI will solve, how work happens today, the information available, what AI must never do, how people will judge it, where it fits into daily work, and who remains responsible. The answers should lead to a clear decision to test, wait, or fix the foundations first.
How long does the AI business check take?
A focused check can often be completed in one or two working sessions when the right business, information, technical, and safety owners are present. Complex or regulated ideas may need deeper review. The goal is not to rush toward a score. It is to find uncertainty before it becomes expensive software.
Does a business need perfect data before trying AI?
No. A business needs information that is good enough for the specific problem, not perfect information across the whole company. A small set of reliable, permitted, and representative examples may be enough for a narrow test. A large archive does not help when its meaning, quality, or permitted use is unclear.
What score is good enough to start a small AI trial?
The score is planning guidance, not permission. A score of 11 to 14 can support a small real-world trial when no critical answer is missing. A zero in safety, permission, proof, or responsibility should stop progress regardless of the total. Fix the critical gap, record the evidence, and answer the questions again.
Xyric's related guide to data governance for SMEs can help when ownership and definitions are the weakest scores. When you are ready to build, the Launch track helps turn a clearly defined AI idea into a working product.
Book a discovery call to take one AI idea through these seven business questions and identify what to fix, test, or fund next.
Sources
NIST AI RMF Core, NIST AI Resource Center, accessed 2026-09-08.
Rules of Machine Learning, Google for Developers, accessed 2026-09-08.
