Staying can be a decision, too.
Darko Tushev ·
An argument within the Polaris framework, with a fictional teaching example. No company findings or measured results are claimed.
If every AI signal leads us to a new North Star, we have made changing direction the goal. I think the framework should leave room for a less dramatic answer: the current direction still deserves our commitment.
That answer needs reasons. “We have already invested too much” protects a past decision. “The outcome still matters, the route still works, and the proposed alternative has not earned the disruption” makes a claim we can examine.
Signal: notice the possibility without making the commitment.
Imagine a retailer considering an AI assistant that drafts refund replies. The demonstration is promising. The team's existing North Star is durable issue resolution, and its current process already carries a case through to someone who can resolve it.
For this invented discussion, suppose a small review of completed cases finds no recurring handoff failure, while the demonstration has not yet shown how it handles missing payment records or exceptions. Those are scenario assumptions, not research evidence. They make Stay plausible. They do not prove the existing process is best.
Shift: ask what changed in the work.
The demonstration changes what may be possible in drafting. It has not established that the customer's need changed, or that the proposed workflow resolves refunds more reliably. The existing changing-work evidence review offers a related caution about comparisons: check what work enters the measure before interpreting a gain. That research does not settle this fictional retailer's decision.
Keep curiosity alive through an offline comparison of representative cases, including exceptions. Compare the full path to resolution, staff effort, and error handling. A faster draft could still be useful. The question is whether its benefit survives the rest of the work.
Star: say exactly what we are staying with.
Stay, here, means retaining durable resolution as the governing outcome and keeping the current customer-facing workflow for now. It does not require stopping investigation. The commitment under review is a rollout of the proposed assistant, not whether the team may learn about AI.
Adjust becomes more persuasive if comparable cases show that a different route improves resolution or reduces effort without unacceptable errors. Shift requires a stronger argument: evidence that the governing outcome no longer captures the value people need. Dissatisfaction with one tool does not establish that the North Star should move.
System: give Stay an owner and a review trigger.
In this fictional decision, the support lead records the scope, assumptions, and reasons for deferring rollout. The payments partner checks whether the case review misses unresolved refunds. Review in four weeks, or sooner if repeat contacts, exception delays, or a relevant capability change challenge the reasoning.
Agree before the comparison what counts as resolution, which cases are comparable, and which errors would make rollout unacceptable. If there are too few cases or refunds have not settled, extend observation. An average reply-time improvement alone cannot answer the resolution question.
What would change my mind?
Repeated unresolved cases would weaken the assumption that today's route works. A credible comparison showing better resolution at acceptable effort and risk would strengthen Adjust. Customer evidence that exposes an unmet need beyond the current outcome would reopen the Star itself.
Missing evidence deserves a different sentence: “We do not know enough yet.” A bounded pause can be sensible when the decision is reversible and delay is tolerable. It should also name the cost of waiting and the evidence we will gather. Otherwise, Stay can become a comfortable way to avoid deciding.
For the next conversation, try this: What are we choosing to preserve, what would make that choice wrong, and when will we look again?
Record the decision in a North Star Brief →
Provenance: this is a Polaris argument applying SIGNAL → SHIFT → STAR → SYSTEM and Stay/Adjust/Shift. The retailer, case observations, comparison, and review schedule are fictional. For a different decision using the same setting, read the dashboard-versus-outcome example. For the full framework, read the white paper.