Tessa AI Troubleshooting Checklist for Workplace Operations

When Tessa AI is not producing useful workplace operations answers, the problem is rarely solved by simply asking the question again. The likely causes are more specific: an unclear request, incomplete source data, permissions that limit context, an undefined human review point, or a mismatch between what Tessa is designed to support and what the user is asking it to decide. This diagnostic checklist helps workplace, facilities, employee experience, and operations teams isolate the cause before changing prompts, policies, or workflows.

Start by treating the issue as an operating-system problem, not only an AI problem. Confirm what Tessa was asked to do, identify the records and rules it relied on, and determine whether a person still needs to review the result. For governance context, see this Tessa AI policy template for workplace operations.

Start with the employee or operator symptom

Describe the failure in observable terms before investigating the cause. “Tessa is not working” is too broad to diagnose. A useful symptom statement identifies the user, task, output, and point of failure.

• No answer: Tessa does not respond, cannot find relevant information, or indicates that the request is outside its available context.
• Incomplete answer: The response addresses part of the request but omits a location, date, team, resource, or operational constraint.
• Incorrect answer: The response conflicts with a known assignment, availability rule, policy, or workplace record.
• Unclear answer: The response may be reasonable, but the user cannot tell which information supports it or what action to take next.
• Unsafe or overconfident answer: The response appears to make a decision that should remain with an authorized workplace operator.

Capture the original request, the response, the relevant date and location, and the expected outcome. Preserve the exact wording where possible. Small details such as “today,” “near my team,” or “available for visitors” can materially change the operating context.

Check whether the request defines an operational task

Tessa is more likely to provide useful assistance when a request states the task and its boundaries. A vague question can produce a vague answer even when the underlying workplace data is accurate.

Clarify the decision or action

Separate requests for information from requests for recommendations and requests for execution. These have different risk levels and review needs.

• Information request: “Which meeting rooms are available for this group on Thursday afternoon?”
• Analysis request: “What capacity issues appear in this location based on current bookings?”
• Recommendation request: “Which available room best fits this meeting’s requirements?”
• Action request: “Create or change this booking,” subject to the permissions and controls of the connected workplace system.

Include the location, date, time range, group or resource requirements, and any relevant constraints. Ask Tessa to identify missing information rather than silently assume it. If the request involves a sensitive employee matter, access decision, accommodation, disciplinary issue, or other consequential judgment, route it to an authorized human reviewer.

Verify that the source records are current

An AI response cannot be more reliable than the workplace information available to it. Check the underlying records before changing the wording of the request.

Review the records behind the answer

• People records: Confirm that employee names, teams, locations, roles, and active status are current.
• Space records: Check that rooms, desks, zones, capacity details, accessibility information, and resource attributes are accurate.
• Booking records: Look for cancellations, duplicate reservations, recurring bookings, expired holds, and reservations that were created outside the expected workflow.
• Assignment records: Confirm that assigned seating, team neighborhoods, move changes, and exceptions have been updated.
• Policy records: Verify that booking windows, eligibility rules, approval requirements, and escalation routes reflect the current operating policy.

Pay particular attention to timing. A recently changed assignment or room status may not be reflected in every connected source at the same moment. Record the time of the test and identify which system owns each field. If no system owner is clear, the data problem will likely recur regardless of how Tessa is prompted.

Separate availability from capacity and utilization

Many apparent Tessa errors come from treating related workplace concepts as interchangeable. They are not.

• Capacity describes how many people or resources a space is designed or approved to support.
• Availability describes whether a resource can be booked under the current rules for a particular time.
• Utilization describes how the resource is actually used over a period, based on the measurement method.
• Eligibility describes who may book or use the resource and under what conditions.

For example, a room can have capacity for ten people but be unavailable because it is reserved, under maintenance, or restricted to a particular use. A desk can be available in the inventory but ineligible for a user because of location or accessibility rules. When investigating a response, ask which of these concepts the request required and whether the source data measures it directly.

Inspect permissions, identity, and access boundaries

Tessa may provide an incomplete answer because the requesting user is not allowed to see all of the relevant information. That is a control working as intended, not necessarily a data failure.

Confirm that the user is signed in with the correct identity and has the expected role, location access, and administrative permissions. Check whether the request crosses a boundary between offices, teams, private assignments, or restricted visitor information. Test the same operational question with an appropriately authorized administrator only when doing so follows your privacy and security procedures.

Do not broaden access simply to make an answer appear more complete. Instead, define what the user should be able to see, what Tessa should withhold, and how the user should request an exception. A useful response may need to say that a resource exists but that details require an authorized operator.

Use this Tessa troubleshooting decision matrix

Match the symptom to the most likely diagnostic path. The matrix is a starting point, not proof of a root cause. Validate the result against the original request, source record, and user permissions.

Match the failure pattern to the next check

• No answer or “not found”: Check terminology, date and location filters, source connectivity, and whether the user has access to the requested information.
• Answer conflicts with a known record: Verify record ownership, update timing, duplicate entries, and the definition of the field being compared.
• Answer is too broad: Add the decision boundary, required constraints, and desired output format. Ask Tessa to state assumptions and missing inputs.
• Answer omits an exception: Check whether the exception is documented in the relevant policy or record, rather than relying on informal knowledge.
• Answer recommends a consequential action: Require human review, confirm authorization, and document the final decision separately from the AI-assisted analysis.

Test the prompt without creating new ambiguity

Prompt testing should make the task more precise, not encourage users to work around controls. Change one variable at a time so you can identify what affected the result.

First, restate the request with explicit dates, locations, resources, and constraints. Next, ask Tessa to list the information it used and identify assumptions or missing fields. Then ask for a bounded output, such as a short list of eligible options with the reason each option qualifies. Avoid asking for certainty when the records are incomplete. Ask for a confidence limitation or escalation recommendation instead.

Keep a small test set of representative requests. Include a straightforward lookup, a request with an exception, a request involving multiple locations, and a request that should be escalated. Test after a source-data change, policy change, integration change, or permissions change. This makes troubleshooting repeatable and helps distinguish a one-off prompt issue from a systemic problem.

Check integrations and update ownership

When Tessa relies on connected workplace systems, identify the path information takes from its source to the response. Check whether the integration is active, whether the relevant object or field is included, and whether synchronization has a documented timing expectation.

Assign ownership for each layer: identity data, employee and team records, space inventory, bookings, policies, and Tessa configuration. The owner should know how to confirm freshness, report a failed update, and communicate a temporary workaround. Avoid maintaining a second unofficial spreadsheet of corrections. That can hide the real source problem and create conflicting answers.

If a workplace team needs a broader operating foundation for connected data, review Tactic’s workplace management platform, which brings together workplace resources and operational workflows in one environment.

Confirm the human review and escalation path

Some requests should not be resolved by an AI response alone. Define the point at which Tessa can inform, summarize, or suggest, and the point at which an authorized person must decide or act.

Escalate when the request concerns safety, accessibility, privacy, security, employment status, a disputed assignment, an exception to policy, or an action with material operational impact. The escalation record should include the original request, relevant source records, Tessa’s response, the reviewer’s decision, and any follow-up correction. This creates a feedback loop without treating the AI output as the official record.

Review the issue after the immediate fix

Once the answer improves, retest the original request and at least one related case. Confirm that the fix did not expose information to the wrong audience, bypass a booking rule, or create a new inconsistency in another location.

Track recurring failure themes rather than only individual incidents. Repeated missing fields may indicate an inventory or integration gap. Repeated ambiguity may indicate unclear policy language. Repeated escalation may indicate that the workflow needs a human approval step or a better-defined request form.

Schedule a periodic review of prompts, source ownership, permissions, escalation rules, and representative test cases. Tessa troubleshooting is most effective when it becomes part of normal workplace operations governance, not an ad hoc exercise after an employee reports a bad answer.

Frequently asked questions about Tessa troubleshooting

Should the first response be to rewrite the prompt?

No. Capture the original request, then verify source records, permissions, definitions, and integration timing. Rewrite the prompt after you know the underlying information is available and the user is authorized to receive it.

How can a team tell whether the problem is data or AI behavior?

Compare the response with the authoritative source record using the same date, location, and definition. If the source is wrong or incomplete, correct the data path. If the source is accurate and accessible but the response consistently misinterprets the request, investigate prompt structure, configuration, or escalation design.

Should Tessa make workplace decisions automatically?

Use human review for consequential decisions, policy exceptions, sensitive information, and actions that materially affect employees or workplace access. Tessa can support analysis and coordination without becoming the final decision owner.