Metrics teach a team bad decisions: what a 15-metric leaderboard and 4 screens taught me

Any metric you put on the wall changes how people behave, and not always in the direction you wanted. Last week brought three MarTech pieces on this and a Camunda study of banks, where 87% of executives see a productivity gain from AI and 69% of employees do. Since late 2024 I've been running a leaderboard for a service company in the US, so I'll tell you what in these pieces matches my experience.

A metric creates an incentive whether you want it or not

MarTech, October 20, broke down how marketing metrics teach a team bad decisions. The team hit its lead target, cost per lead went down, and the pipeline isn't growing, because when people are judged by the number of leads they pick cheap audiences and offers that convert well into a form submission and poorly into a customer. The author cites Dreamdata's 2026 data: a click on LinkedIn costs $6.70 on average against $1.70 on Meta, but the cost per company reached is $78.58 on LinkedIn against $144.24. One more figure: from the first LinkedIn ad impression to revenue took 281 days on average, so judging a campaign by 30-day revenue will stop it before the result. The question the author suggests asking of every KPI is one I'm taking for myself: how can a person improve this number while making the real result worse?

MarTech, October 23, wrote about the questions a dashboard can't answer. Marketing lives on fast data like clicks and ROAS. The CEO asks about revenue and profit. A campaign can bring in more attributed revenue and, after discounts and returns, less profit, and the dashboard will show a win. The author splits data into fast, for today's decisions, and slow, for understanding who buys and why.

And MarTech, October 22, came at it from the documents side: if a metric's definition lives in five presentations and each one is different, the numbers in the reports stop matching. The author keeps the definitions in one file with a last-verified date and an owner, and every other document points there. The piece is about reference material for AI agents, but the rule “a fact has one home and a date” works for people too.

When the adoption number says one thing and employees say another

Camunda, October 23, published a study of banks: 87% of executives responsible for processes believe AI has made their teams more productive; among employees, 69% think so. And 45% of employees admitted they had used AI to tick the box when there was no need, and adoption dashboards counted that as progress. Three quarters of executives say process problems have already sunk AI initiatives, at an average of $1.73 million per organization. Camunda sells process redesign, so the study is built around its product, but the gap between the number of logins and work actually closed is familiar to me.

The shortest illustration comes from Forrester, October 20: every week the author returns bread and fruit from an online grocery delivery because a warehouse robot puts a five-kilogram bag of cat litter on top. I don't have the full text, so I won't go beyond this example.

Why this matters to a business owner

Take a car rental where managers get a bonus for the number of cars handed over per day. Within a month handovers speed up, the inspection becomes a formality, scratches stop making it into the report, and a quarter later repair costs go up, and nobody connects them to the bonus. So a quantity metric needs a quality metric next to it, or the team will honestly optimize what you showed them.

What I saw on my own leaderboard

For a PDR company in the US, where I've been the technical partner since 2021, I started a leaderboard at the end of 2024 and developed it all through 2025: leads and sales by manager, percentages, competitions and team splits. It tracks 15 quantity and quality metrics, among them leads, money, calls and SMS. In 2025 I put a TV board on four televisions, each with its own dashboard: for the owner, for production by orders, for the front desk and for the sales team.

A metric on a screen changes behavior, and I've seen it: when a number hangs on the wall, people start working toward it. That's why we keep quantity next to quality, each board has its own viewer, and each one shows workload and the movement of work to the people who need it.

And second: a metric only works if the data under it is honest. Every call is recorded, and AI sorts out who was calling, a lead, an existing customer or an insurer, and how the employee answered. A lead from an IVR call is created in the CRM on its own, so the lead count on the leaderboard can't be drawn by hand. Since 2021 the company has grown from 20 to more than 30 people, every real lead goes through the funnel to the end, and conversion to sale has grown by 30%. That number came from the whole chain of telephony, CRM, app and screens, and the leaderboard is one part of it.

What I would do

For every metric on the screen, write down how it could be improved to the detriment of the business. If the answer comes within a minute, it needs a second metric next to it.

Keep each metric's definition in one place, with a date and an owner.

Split the screens by viewer. The owner and the sales team need different numbers, and one shared screen for everyone helps no one.

Take data from the system, not from manual marks. A call that creates a lead on its own is more honest than a “called” checkbox.

On the rules, statuses and permissions that honest metrics rest on: business logic: rules, statuses and permissions. By area: the business as a single system and spreadsheet automation and reporting. Cases: calls, CRM and the team's work and technical partner of a PDR company. On the same topic: how I read numbers from vendor case studies and how much observability a single-server system needs; earlier I wrote about how to tell a new metric from lost sales. If you already have metrics and don't like how the team behaves, tell me what's on the screen: Discuss your task.