Dashboards that drive better decisions

Operational data becomes useful at the point where a number has a threshold, an owner and an agreed response attached to it, and it stays decorative everywhere else.

Most operations already hold more data than anybody reads. Orders, tickets, shifts, invoices and delivery times sit across four or five systems that were never meant to be compared.

The gap is not collection. It is the step between seeing a figure move and knowing which action that movement calls for, and who is expected to take it.

Leading and lagging indicators

Revenue, churn and monthly margin lag. By the time they move, their causes are weeks behind. They settle arguments well and steer badly, because nothing done today shows up in them today.

Leading indicators move first: quote response time, backlog age, the share of jobs rescheduled. Choosing them is a short exercise, run once for each outcome the team is judged on.

1. Write down the lagging outcome the team is actually measured on.
2. Trace it back to the operational events that produce it.
3. Pick the earliest of those events that somebody can influence this week.
4. Confirm against existing history that it moves before the outcome does.
5. Give it an owner, or leave it off the view.

Keep the two kinds visually separate. Mixed into one row of tiles, a figure nobody can influence directly gets treated as a target, which produces effort in the wrong place.

Thresholds worth an alert

A figure read only when somebody remembers to look will be missed in the week it matters. Decide in advance the level at which it stops counting as normal.

Set thresholds from the recent distribution rather than a round number: a limit chosen for looking tidy either fires constantly or never fires. Two levels suffice, one colouring the tile and one messaging a named person.

Alerts need a budget. Every alert that fires without requiring action trains its recipient to ignore the next, so one dismissed several weeks running should be retuned or switched off.

Segmentation before averages

An average across the whole business hides the thing worth acting on. A stable mean delivery time can contain one region getting steadily worse and another that improved enough to mask it.

Cut by dimensions that map to a lever: region, team, product line, channel, shift. Where nobody can act differently on a segment, the split gives an interesting chart rather than a decision.

Distribution matters as much as the cut. The median and the upper percentiles describe the experience of the worst-served customers far better than the mean, and those customers leave first.

A cadence for review

Numbers get acted upon when a recurring meeting exists whose agenda is the numbers. Each rhythm should match the speed at which its figures can move, and each slot needs an owner.

• Daily exception check: overdue work, stuck queues and anything that crossed a threshold overnight.
• Weekly operating review: leading indicators by segment, with one owner answering for each trend.
• Monthly business review: lagging outcomes, read against what the leading indicators implied a month earlier.
• Quarterly metric audit: definitions, thresholds and alert volumes, revisited now the business has changed shape.
• Post-incident review: the figures that should have flagged the problem sooner, and why they did not.

Acting on the number

The step most often skipped is recording what was decided and returning to check whether it worked. A short note holding the reading, the action and the effect expected makes a run of meetings something to learn from.

It also exposes measures that lead nowhere. A figure reviewed for months that has never changed what anybody did belongs in an archive rather than the weekly pack.

• engineering