Skip to content

OrganizationPractitioner8 min read

6. Strategic steering

Translate the technical into executive indicators, design a dashboard that can be read in two minutes, assess AIOps without illusions, lead change with ADKAR and spot the weak signals of a project in trouble.

For: team managers · executives and CIOsPrerequisites: Have read lesson 1 of the course.

Reading mode

The CIO does not need the same indicators as the operations team. An airline pilot watches a handful of dials; the mechanic on the ground opens the cowlings. Knowing how to translate the technical into business impact is what separates a project that survives a change of leadership from one that gets cut at the first budget trade-off.

Executive indicators and technical indicators

Section titled “Executive indicators and technical indicators”

Five indicators are enough. The first is service availability as seen by the customer: the share of time the service is fully usable for the user, not for the infrastructure. A server that responds while the order page will not load counts as an outage. The second is the cost per incident, mean and median, counting lost margin, remediation and penalties.

The third is resolution time as seen by the customer, from the first complaint to the actual return to normal. The fourth measures how well contractual commitments are met: the share of service levels honored over rolling 30 and 90 days. The fifth is the cost of observability, relative to the IT budget or, better, to business activity (see lesson 4).

The teams track finer measures. They watch their SLOs and the burn rate, the speed at which the error budget is being used up, that margin of failure you allow yourself on an objective; consumption that runs too fast triggers the alert before the objective is missed (see the error budget simulator). They also track MTTD and MTTR (mean time to detect and to recover) by severity level, the error rate per service and per route, latency in percentiles (p95, p99, meaning the time within which 95% or 99% of requests complete) and instrumentation coverage of critical services.

Each executive indicator must link to at least one technical measure. That link is what lets the committee ask precise questions when a figure drifts.

I suggest five principles. First, five indicators at most on the home page; the rest is available in detail but not shown up front. Next, three colors, green, amber and red, with written thresholds. Ten-level gradients survive neither the projector nor a change of tool.

Every figure comes with a comparison, against the target or the previous period: a figure on its own says nothing. Every indicator offers a path to the detail and opens onto the technical measures behind it. Finally, the dashboard updates continuously and the IT department comments on it in a monthly review. Without a review, it becomes wallpaper.

The interactive executive dashboard applies these principles to Helinord’s fictional data: click an indicator to drill down to its technical measures.

AIOps, the use of AI techniques in operations, is one of the topics most heavily wrapped in marketing. Think of an assistant who sorts your mail: useful if trained on your mail, useless if handed envelopes with no address.

On good data, AIOps does real things. It reduces alert noise by spotting anomalies, groups alerts that belong to the same incident, picks out recurring patterns and anticipates some short-term saturation.

It does not replace defining SLOs: with no objective, it has nothing to optimize. It does not make up for poor data, does not fix an architecture and does not decide alone on a major incident.

It has prerequisites: several months of clean history, consistent tags across services and environments, controlled cardinality and structured post-mortems that make it possible to label past incidents.

Before buying, ask the supplier five questions. Which models does it use and can it explain their results? Is it compatible with OpenTelemetry? What does it cost and is that cost capped in the contract? What results does it give on your data, in a proof of concept rather than a demo? What is the exit plan?

The ADKAR model, published by Prosci, describes five stages a person goes through to adopt a change, a bit like learning to drive: understand why, want to, learn the rules, practise, then keep the habit. Applied to observability, it looks like this:

StageWhat the person experiencesLevers
Awarenessthey understand why change is neededcost of incidents quantified, comparable cases
Desirethey want to take partinternal champions, visible quick wins, recognition
Knowledgethey know how to do ittraining, written procedures, pairing, community of practice
Abilitythey can do it day to dayplatform in service, SLOs in production, self-service, up-to-date documentation
Reinforcementthe change lastsblameless post-mortems, indicators presented to the committee, adjusted roadmap

A project does not collapse all at once; it gives warning signs, like a car that starts making an unusual noise. The thresholds below are benchmarks I suggest for opening a discussion, to be adjusted to your context.

SignalWarning thresholdWhat it means
Departures from the teamseveral departures in a year from a small teamexpertise and context leave with the people
Alert noisethe vast majority of alerts require no actionthe real signals are drowned out
Trace coveragea minority of critical services instrumenteddiagnosis happens blind
Post-mortemsnone in six monthsthe same mistakes keep repeating
Alert debthundreds of rules never reviewedobsolete rules mask the real anomalies

When several of these signals appear at the same time, the project deserves a reset led directly by the CIO.

Allow 45 minutes, alone.

  1. Choose five indicators for your executive committee.
  2. Link each one to at least one existing technical measure, or note that it is missing.
  3. Set a target and green, amber and red thresholds for each one.
  4. Write the commentary you would give at the next monthly review.
  5. If an indicator concerns a business unit, plan a workshop with it before locking it in, as Laurent did with Jean-Paul.

To finish, write down your next action: which one, by what date, with whom?