This piece is for the managing director who has to put AI on one page for the board, and keep it there.
Four blocks, same order, every quarter
1. Value, job by job. List each job handed to AI agents on its own row. For each, report only the value types that apply: time, errors, revenue, software spend. Give three columns: the baseline before go-live, last quarter and this quarter. Name who measured it. A row that cannot be filled in stays on the page, marked blank, with the reason.
2. Running. Is it live and in use? For each job, show how much of the work now runs through the platform and how much still goes around it. Work that goes around it is the early warning: people do not trust the output, or the process changed and nobody told the agents.
3. Control. What did people approve, and what did the agents do on their own? Note any change in what the agents may do alone, and the record that justified it. This is earned autonomy on paper: scope widens only when the evidence says it should. List every incident, however small.
4. The decision. End with one question for the board: keep a job as it is, widen it, or stop it. A scorecard that never asks for a decision becomes a status report, and status reports get skimmed.
Three rules that keep the page honest
- Set the baseline before anything goes live. Count how long the job takes, how often it goes wrong and who does it. Without that, every later number is a guess.
- Activity is not value. Tasks completed and messages sent belong in an appendix, if anywhere.
- Keep the page fixed. Same blocks, same order, same jobs until the board drops one. Change the layout and directors lose the thread between quarters.
Where the method breaks
There is no baseline. If a job went live unmeasured, say so. Use this quarter as the baseline and report the change from next quarter.
The value lands somewhere else. Faster quotes may show up as sales revenue, not as time saved in the sales team. Agree with the finance lead where each value type is counted, once, before the first report.
Some value resists a number. Better decisions are real and hard to count. Report them as evidence, such as a decision made earlier because the signal was clear, and never dress them up as a figure.
One number gets gamed. If a single measure decides a job's future, people will steer it. Read value, running and control together.
What stays with people
Agents can gather the figures and prepare the draft. The owner of each job signs off its row, and the finance lead confirms anything touching money. The board makes the decision in block 4. No agent reports on its own success without a person checking the record.
One page, read in two minutes
A good scorecard is boring to make and quick to read. Directors should see in two minutes what changed, what it is worth against the baseline and what they are asked to decide. Clear signal, not noise.
Futuvara is a creative technology agency. We build, implement and run AI platforms for our partners. If you want a second view on what belongs on your board's page, Start an Agentic conversation.
