> ## Documentation Index > Fetch the complete documentation index at: https://docs.nofire.ai/llms.txt > Use this file to discover all available pages before exploring further. # Analytics > How long investigations take, whether your team trusts the answers, and what they cost. Analytics answers two different questions about NOFire AI: whether it is making on-call faster, and whether it is being used. Open it from **Analytics** in the left sidebar. Every panel needs a connected source and recorded investigations. On a new account the page says so rather than showing an empty chart. ## Mean time to conclusion How long the agent takes to reach a conclusion. This is the one figure on the page that answers whether NOFire AI is making on-call faster, rather than whether it is being used. Three rules decide what goes into it, and each one exists because the number could otherwise mislead: * **Only investigations the agent finished.** One still running has a runtime that means "so far". Averaging it in drags the figure toward whatever happens to be in flight when you open the page. * **Only a positive runtime.** A zero is a row that was never timed, not an investigation that took no time. Where no runtime was recorded, the elapsed time between when the investigation was created and when it was last updated stands in. * **The sample is shown next to the mean.** The mean is computed over a page of the investigation list, not over the whole account, so the number it is averaged over is printed rather than hidden. Finished by the agent is not the same as concluded by a person. The agent finishing writes a postmortem draft. A conclusion is recorded only when somebody runs the conclude flow. Read this metric as how long the agent took, not as how long your team took to close the loop. ## Trust score A score built from the ratings your team leaves on investigations, weighted so that recent ratings count more than older ones. * A falling score means recent answers were rated lower. Use **By feedback type** to see which kinds of investigation are rated worst. * A steady high score means the reasoning is holding up for your environment. * An account with few rated investigations shows a more conservative score until enough ratings arrive. **Response rate** is the share of investigations that got any rating. A low response rate makes the trust score less reliable, because it is built from fewer opinions. ## Investigations Volume and origin over the window you select. | Panel | What it tells you | | -------------------------- | ----------------------------------------------------------------- | | How investigations started | Whether an alert, a person, or a workflow began the investigation | | By source | Which connected system the investigation came from | ## Weakest entities Where the reasoning is least reliable, broken down by the services involved. | Panel | What it tells you | | -------------- | ---------------------------------------------------- | | Highest impact | The entities where a better answer would matter most | | Emerging risk | The entities whose ratings are getting worse | Use this to decide where to add [Knowledge](/knowledge/organizational-knowledge). Adding context for a weak entity is the most direct way to raise its ratings. ## Active users Who is using NOFire AI, and who is not. | Panel | What it tells you | | ---------------- | ------------------------------------- | | Utilisation | How many people used it in the window | | Not using it yet | People on the account who have not | ## Tokens What the reasoning costs. | Panel | What it tells you | | ----------------- | --------------------------------- | | Spend by model | Which model the budget goes to | | Per active person | Cost per person actually using it | ## Next steps Add the context that raises ratings for your weakest entities. The scheduled and triggered runs that appear in these numbers.