Learning probes for service teams

Why a learning probe beats a confident plan

Why do uncertain situations so often produce confident, oversized plans? Because committing feels like progress. But under real uncertainty, a large commitment usually hides risk rather than reducing it. A learning probe lets the work teach you something before failure hardens into loss, because it is small enough that a wrong answer is cheap and informative.

A simple action rhythm

The core move is to act on the conditions producing a problem, not on "issues." A learning probe follows a rhythm you can repeat: observe, interpret, act, then observe again.

1. Name what is visible in the work or the stories.
2. Propose the mechanism most likely producing it, the condition, not the symptom.
3. Choose one small, reversible change to that condition.
4. Decide now when you will revisit the same view to see whether anything moved.
5. Close the loop: tell people what you heard, what you changed, and what you will check next.

Define the probe once, run it in many teams

A learning probe does not have to be reinvented on every board. In ScrumDo a probe can be captured as a definition, a work type with its own task map: the ordered steps and default tasks the probe should follow. Define it once at the portfolio level and it can be provisioned into every team that needs it, so the same experiment runs the same way in many places. Teams that adapt their copy keep their changes; the ones that have not diverged stay in sync when the definition improves.

Let governed agents run the probe

Because a probe is defined as concrete tasks on a card, governed agents can support it. ScrumDo routes the work through a governed loop, a maker agent does the work, a verifier agent checks it, and a human approves the plan where policy requires, so the probe can move through the same card record without anyone losing the thread.

• A governed loop routes steps to specific agents, maker, verifier, and more, instead of one opaque run.
• Token and cost budgets cap how much a probe can consume before it dispatches.
• Allowed connectors, required skills, and approval gates are pinned before it starts.
• Every connector call and outcome stays on the card as evidence.

Test for contribution, not attribution

In a complex system you cannot prove a single cause, and pretending to wastes trust. A learning probe asks a more honest question: did this change plausibly shift things in the direction we intended? Small, reversible changes give clear signals; large, bundled changes blur them until you are left telling stories about causation you cannot support.

Learning at scale

When the same learning probe runs across many teams, the results become comparable instead of anecdotal. ScrumDo rolls each run’s evidence and outcomes up to the portfolio, confidence, coverage, and evidence gaps by release and value stream, so leaders can see what held across instances, not just what happened on one board. That is how a hundred small, safe-to-fail experiments become one honest picture of what is working.