Skip to main content

​
Frequently asked questions

Access questions are covered in User Management, the allowance in Credits. This page covers using the agent day to day.

​
Getting good answers

​
Why did the same question give me a different answer the second time?

It depends on the kind of question. Recall questions are stable. “How many employees have this skill” or “what percentage of time for this job is automatable” are calculations over your data: ask twice, get the same answer. If a recall answer did change, the most likely cause is that your data changed in between. The agent reads live data, so validated profiles, edited jobs or a completed integration run move the numbers with them. Generative questions will vary. “What are the role redesign opportunities for this job” asks the agent to reason and propose. Two runs produce two well-founded but differently worded answers, the same way two analysts would. This is intended behaviour.

​
How do I get more consistent results?

  • Be specific about scope. Name the job family, organizational unit, peer group or time period rather than leaving it open.
  • Say which data to use, for example only validated skills, or only a named set of jobs.
  • Start a new conversation for each new analysis. Earlier context helps within one piece of work and confuses across two.
  • Attach the exact list. If you want a specific set of jobs or employees, upload it rather than describing it.

​
The agent says it cannot answer my question. Should I rephrase it?

Once, yes. Repeatedly, no. When the agent says it cannot do something, that is usually accurate: the question needs an analysis the underlying data products do not support, and pressing it into attempting one produces a long wait or an unreliable result. If you believe the question should be answerable, raise it with your TechWolf representative; knowing which analyses people want shapes what we build.

​
Data and conversations

​
Why can a colleague see something I cannot?

The agent shows you exactly what your Console access allows. See What determines what a user sees.

​
What languages can I use?

You can hold the conversation in any language the agent’s multi-lingual models support: prompting in French returns an answer in French. The surrounding interface is English only.

​
How is my data handled?

The agent reads only from your organization’s TechWolf data, within your deployment region. See Responsible AI Principles and Security. For the Copilot surface specifically, see How It Works.