Before you start, connect your assistant: see What is Pensero MCP, and how do I connect it?. For basic questions, see Use Pensero MCP to ask questions about your engineering data.
You don't need to name the tools. Ask your question in plain language and your assistant selects the correct tool. The tool names are shown in brackets so you can check which tool your assistant used: ask it "Which Pensero tools did you call, with which arguments?".
Find why a metric changed
When a metric goes up or down, ask your assistant why. It compares the current period with the period before it (explain_metric_change) and breaks the change into the parts that make the metric.
Try:
"Why did our rework rate go up in the last 30 days?"
"Our delivery dropped in September compared with August. Why?"
"Why did AI cost per delivery point go up for the Platform team?"
After you look at the Signals dashboard: "Why did that signal change?"
What you get
The value in each period and the change. By default, the current period is the last 30 days and the baseline is the 30 days before it. You can give your own periods.
The parts of the metric. For example, total delivery = human + AI-assisted + agentic delivery, and rework rate = rework delivery ÷ PR delivery. For each part, you see how much of the change it causes. The parts add up to the full change.
Joiners and leavers. The part of the change that comes from people who did work in only one of the two periods.
Data problems. Failed syncs, items that are not scored yet, accounts that are not linked to a person, and days with no AI usage data. Your assistant tells you which parts of the metric these problems can change, so that a data gap is not reported as a change in the work.
Go deeper, step by step
Check the data first. If your assistant reports a data problem, fix it before you act on the change. For example, link unlinked accounts in Data health, then ask again.
Follow the largest part. Ask "Go deeper on the part that changed most". Your assistant calls the tool again on that part, until it reaches the raw data.
Find who and what. Ask:
"Who caused this change?" (by person)
"Which teams caused it?" (by team)
"Which PRs caused it?" (by item)
Limits
Some metrics show values but no split. AI cost per delivery point, tokens per delivery point and agent cost per delivery point count different groups of people above and below the line. You see how each part changed, but not how much of the change each part causes.
The person and team breakdowns work for groups of 50 people or fewer. For a larger group, ask for one team, or ask for the items.
Team shares can overlap. A person in two teams counts in each team.
Item lists exist only for delivery, PR quality rates and AI spend. For other metrics, your assistant tells you that no item list exists.
Failed syncs and missing AI usage days are checked for the whole organization, not only for the group you ask about.
SWOT for a person or a team
Get the strengths, weaknesses, opportunities and threats of one person or a group (get_swot).
Try:
"Give me a SWOT for Alice for the last quarter."
"Run a SWOT for the Backend team."
For a person, Pensero compares them with comparables: people with a similar work mix, repositories, technologies and team. Opportunities show where comparables clearly do better. Threats show metrics that are going down and knowledge that only one person has. A group with no comparables is compared with its own previous period.
Every point comes with its numbers. Ask your assistant to show the evidence for any point.
Head-to-head: compare two people on one metric
Find why two people differ on one metric and what can change it (get_head_to_head).
Try:
"Why is Alice's defect rate higher than Bob's?"
After a SWOT: "Compare Alice with the comparable that does better on cycle time."
You get both values and the gap, the metric week by week, what the metric is made of for each person (for example review findings, roadmap vs off-roadmap work, cycle time samples, defects and rework), and recommendations with example PRs and review comments. You can compare only people that you can see in Pensero.
Compare and rank
Put people, teams and groups side by side, or rank them (compare_metrics).
Try:
"Compare the Backend, Frontend and Platform teams on delivery, rework rate and cycle time."
"Who are the 5 people with the best AI cost per delivery point in Engineering?"
"Rank my reports by review help, worst first."
You can compare up to 25 groups at one time. A ranking returns up to 25 people in one direction. Ask again for the other direction.
Build your own groups (cohorts)
Make a group of people with filters such as tenure, employment type, level or team, and use it in any question.
Try:
"Compare full-time engineers with more than one year of tenure against everyone else on defect rate and cycle time."
"Which filters can I use to build a group?"
"Use my saved cohort 'Senior backend' and show its delivery trend."
Your assistant can first show how many people match a filter and some of their names, so you can check the group before you use it.
Compare with the industry
Compare your organization with other companies (get_benchmark). You see your value, your percentile and the industry median and other percentiles. If your organization has peer filters turned on, you can compare with companies of a similar industry, size or country.
Try: "How do we compare with companies of 100 to 500 employees on delivery per engineer and AI cost per delivery point?"
Pensero doesn't show data from other companies, only your percentile and the industry statistics. When too few companies match a filter, the industry values are not shown.
Add context with annotations
Annotations record leave, onboarding, role changes or incidents. They explain changes that metrics alone can't explain.
Try:
"Before you explain the drop in delivery, check the annotations for the same people and period."
"Add an annotation for Bob: on leave from 1 to 14 October."
Managers can add annotations for their reports, and org admins for anyone. If your organization allows it, people can add annotations for themselves. Your assistant can also delete an annotation if you have the same permission. It asks you first, because you can't undo a delete.
Test an AI budget before you save it
Check how many people an AI budget rule would flag, without saving it (preview_ai_governance_policy).
Try: "How many people would go over a budget of $300 per month on AI tools?"
Find how a metric is calculated
Ask "How is AI cost per delivery point calculated?" or "What makes up rework rate?". Your assistant reads the formula of each metric and its parts, down to the raw data.
Tips for better answers
Give the period and the group, for example "the Platform team, July compared with August".
Ask for the data check: "Can a data problem explain this?"
Ask which tools it used when an answer looks wrong, and send us the question and the tool calls.
Ask it to go deeper instead of starting a new chat. Your assistant keeps the context of the conversation.
What you can see
Advanced tools follow the same rules as the Pensero web app. Executives see the whole organization, managers see their reports and individual contributors see their own data. Unlinked account names and emails are shown only to executives.