Agile metrics are supposed to help a team or a train change its behaviour. Most dashboards help a steering group decorate a slide. Measure flow, predictability, and outcomes — and refuse vanity figures that punish the people who told you the truth.
Flow before theatre
Start with four flow measures you can explain in a corridor. Work in progress (how much is started). Throughput (how many items finish per period). Cycle time (how long from start to done). Flow efficiency (active time versus waiting time), if you can measure wait without lying. These tell you whether the system is clogged. A burn-down that hits zero because you split stories on the last day tells you the team can edit Jira.
Plot ageing, not just averages. An average cycle time of six days can hide an item that has been 'in review' for a month. Ageing work is where dependencies and fear live. Pull the oldest blocked item into the weekly conversation before you discuss velocity. Velocity without ageing is a speedometer with the handbrake on.
Compare a team to its own history, never to the team next door. Different work, different size of story, different definition of done. Cross-team velocity league tables are how you train people to inflate estimates and refuse hard work. If someone asks for a 'normalised story point' across the ART, they are asking for a fiction that will be used in a performance conversation. Say no, and offer throughput of similarly sized slices instead.
Predictability is say/do, not heroics
Predictability at team level is: of what we forecast for the sprint or PI, what did we finish, and what did we learn? At ART level, SAFe's PI predictability measure (planned PI objectives versus achieved) is useful if objectives were honest. If teams sandbag to hit 100%, or leadership stuffs the plan after the vote, the number is a costume.
Uncommitted objectives exist so you can be ambitious without lying. Use them. A train that always hits 100% of a timid plan is not predictable in a way customers care about. A train that hits 80% of a stretching plan and can explain the miss is a system you can steer. Write the miss as a cause, not a vibe: dependency slipped, scope grew, capacity was fiction.
Forecast with ranges and recent throughput, not a single date extracted from a Gantt that nobody updates. 'Based on the last three PIs, this feature size usually lands in one PI; the integration risk makes two more likely' is a professional sentence. 'Go-live is 14 November' when the program board is mostly red is how PMs become narrators of a disaster they could see from June.
Value, and numbers that become weapons
Output is not value. Stories completed, points burned, and features launched are activity. Pair them with an outcome: conversion, cycle time for the customer, incidents, revenue, or a leading indicator you named in the epic hypothesis. If nobody can name the outcome, you are funding motion. Motion photographs well. It does not survive a benefits review.
Never use sprint velocity as a performance metric for individuals. Never use escaped defects as a stick without looking at test environments and last-minute scope. People will game whatever you hang a bonus on. They are not immoral. They are adapting. Choose measures that stay useful after they are gamed — customer outcomes and flow — or accept that you are running a theatre of targets.
Show a short set, every time: flow (WIP, ageing, throughput), predictability (say/do with reasons), and one or two outcomes. Kill the twenty-metric dashboard that nobody can act on. If a number did not change a decision this month, stop collecting it. Measurement is a cost. Unread measurement is a cost with a colour palette.
Course complete — what's next
Measure flow (WIP, throughput, cycle time, ageing), predictability as honest say/do, and outcomes rather than points. Do not compare team velocities. Do not weaponise the dashboard.
You now have a working picture of agile at scale: what SAFe is for, how PI Planning creates a shared baseline, how dependencies are reduced and owned, how a portfolio kanban limits bets, and which numbers deserve airtime. Use one of them on a live train this week — a cleaner program board will teach you more than another framework diagram.