Choose an AI tool by identifying the operating decision it should improve, the evidence it needs and the manager who will review its output. Test it against your current process before expanding its role, with clear limits on data access and automated actions.
Key takeaways
- Start with a recurring operating decision, then assess whether a tool improves it.
- Compare results with the current process using data you are authorized to share.
- Keep a named manager responsible for exceptions and consequential decisions.
- Set data-use limits, success criteria and an exit path before wider adoption.
Which problem are you trying to solve?
“Use AI” is not an operating objective. “Help the manager prepare a better demand forecast before writing the schedule” is a question that can be examined. So is “identify recurring breakfast complaints without reading every review from scratch.”
Begin with a task that already happens. Identify who does it, which information they use and where the result falls short. If the team cannot explain the present process, software selection may be premature.
At Life House Hotels, I built the F&B department and set its technology platform alongside portfolio-wide standards and programs. That experience is relevant to choosing systems: the value depends on how the information reaches a decision, not simply on what appears on a dashboard.
What should a useful selection brief contain?
Write the brief before the vendor demonstration. Otherwise, the product’s strongest features can become your requirements by default.
Specify the decision, its owner and the information available. Describe an ordinary situation and an exception the manager finds difficult. Include the systems the business already uses and any limits on access or data sharing.
| Decision area | What a tool may support | Where a manager still makes the call |
|---|---|---|
| Forecasting and scheduling | Demand estimates and proposed coverage | Service standards, skill mix, events and unusual conditions |
| Guest feedback | Grouping recurring topics and locating relevant comments | Interpreting the complaint and choosing a response |
| Purchasing and inventory | Suggested orders and identification of unusual usage | Quality, substitutions, events and supplier conditions |
| Reservations and guest relationships | Organizing demand and permitted guest preferences | Table decisions, guest promises and individual service judgment |
The AI & Technology section of our capabilities follows these decision areas. The point is to make ownership and oversight explicit before a tool becomes part of the routine.
How should you evaluate forecasting and scheduling tools?
Ask what the forecast includes and what it cannot see. Restaurant demand may be affected by events, closures or changes in service that are missing from historical records. A hotel forecast also needs to distinguish occupied rooms from the guests likely to use a particular outlet.
Compare the tool’s output with the forecast your manager would otherwise use. Look at where it helps and where the manager needs to intervene. A correct total can still hide the wrong timing or distribution across outlets.
Scheduling adds another layer. Coverage depends on skills and the service promise, not only expected activity. The responsible manager should review availability, role requirements and applicable employment rules with the appropriate support. A recommendation needs an explanation and an override, especially when the context has changed.
What can AI usefully do with guest feedback?
A tool may help group comments by topic and bring repeated complaints to a manager’s attention. Require a way to open the original comments. A summary without traceable evidence makes it difficult to distinguish a pattern from an inaccurate interpretation.
Test ambiguous examples. A guest may praise a server while criticizing the wait, or describe an issue that belongs to another department. The useful output preserves that distinction instead of assigning a vague positive or negative label.
Keep service recovery with a person who understands the guest promise and has authority to respond. Drafted replies need review for accuracy and tone. Do not allow an automated message to promise a refund, accommodation or operational change that the manager has not authorized.
How should purchasing and inventory recommendations be checked?
Start with the records. If recipes, units of measure or product substitutions are inconsistent, a confident recommendation can still be based on the wrong information.
Ask the vendor to demonstrate how the system handles an unavailable ingredient, an unusual event and an incorrect count. Examine whether the manager can see the reason for a suggested order and correct the underlying record.
The purchasing decision also depends on quality and supplier realities. A lower-cost substitute is not necessarily suitable for the approved recipe. A suggested order may be sensible only if a delivery arrives as expected. Keep approval responsibility clear and make exceptions visible to the people who need to act on them.
What needs care in reservations and customer relationship management?
Customer relationship management, or CRM, organizes guest information and interactions. AI may help summarize permitted notes or identify demand patterns, but the usefulness of that information depends on its accuracy and appropriate use.
Ask which records the system combines and how access is controlled. Guest notes should remain relevant to service. A manager should be able to correct inaccurate information and understand what is visible to different teams.
The Federal Trade Commission’s guidance on AI privacy commitments explains why vendor promises about data use matter. Ask specifically about retention, model training and deletion. A verbal assurance during a sales call is not enough to understand how connected business data will be handled.
How do you test a tool without handing it too much authority?
Use a bounded test with a named manager and a decision about what happens next. The tool can propose an output while the existing approval process remains in place.
- Record the current process and its recurring problems.
- Agree on the information the tool may access.
- Choose ordinary and difficult examples to evaluate.
- Define what counts as a useful result before reviewing outputs.
- Record corrections, review effort and unexpected consequences.
- Decide to continue, revise the test or stop.
The NIST AI Risk Management Framework is a voluntary reference for considering AI risks during use and evaluation. For an owner, the practical application is to keep responsibility, testing and review attached to the specific use case. A vendor’s general assurance of accuracy cannot answer how a tool behaves in your operation.
What should you ask about rollout and ongoing ownership?
Identify who will maintain the information after the launch discussion ends. Menus change, suppliers change and managers move. A useful system needs someone accountable for those changes and a clear route for reporting a problem.
Discuss integration work, permissions, staff training and support with the vendor and your technical providers. Confirm what can be exported and how the business would continue if the tool became unavailable. Consider the full effort of using it, including reviewing recommendations and correcting errors.
As COO at Kiki Group, I built corporate infrastructure across functions including HR, reservations and development. That background shapes the questions I ask about responsibility between teams. A software recommendation is incomplete if nobody owns the handoff.
What should make an owner reject a tool?
Be cautious if a vendor cannot explain its data requirements, show source records behind an answer or describe what happens when the system is wrong. A product that needs managers to maintain duplicate records may create more work than it removes.
Do not expand a test because the demonstration impressed ownership. Expand when the evidence supports the decision and the people using the system understand its limits. Sometimes the right result is improving the current process without another tool.
For restaurant and bar owners, that judgment should connect to a specific operating problem, such as those in the restaurant turnaround playbook. For hotels, the F&B owner’s guide helps place technology decisions within the wider guest stay and outlet responsibilities.
What else should you know?
Where should a restaurant start with AI?
Choose a recurring task with a clear owner and a result you can check, such as reviewing a demand forecast or organizing guest feedback. Confirm that the underlying records are usable before selecting software. A defined problem is a better starting point than a broad instruction to adopt AI.
Can AI make staffing decisions for a hotel or restaurant?
A tool can help forecast demand and propose schedules, but a responsible manager should review the result against skills, availability, service needs and applicable requirements. Recommendations should be explainable enough to challenge. Do not treat a generated schedule as proof that the right people are in the right places.
How do advisors help select and implement hospitality AI tools?
An advisor can define the operating requirements, compare vendor claims and establish decision rights and acceptance criteria. The owner’s leaders and technical providers use that brief to plan configuration, training and rollout. Review should focus on whether the tool improves the intended decision in practice.
What data should we share with a potential vendor?
Begin with the minimum information needed to evaluate the use case and confirm that you are authorized to share it. Review access, retention, model-training use and deletion terms before connecting business systems. Avoid real guest or employee information in an initial demonstration when synthetic or de-identified examples will do.
How do we know an AI tool is worth keeping?
Compare its decisions and the time required to review them with the previous process. Include corrections, duplicate work and service consequences in the assessment. Keep it when the evidence supports continued use, rather than because the team has already spent time adopting it.
To discuss a decision for your business, start a conversation with Daniel.
About this guidance
This article provides general operating guidance, not legal, tax, accounting, or investment advice. Requirements and business conditions vary by location and project. Verify current requirements with the relevant agencies and qualified professionals before acting. Contact us about a correction.