Start from the real constraint
Every engagement begins with the constraint that actually hurts: a compliance window, a brittle handoff, a cost curve, or a product promise the current stack cannot keep. We write that constraint in plain language with the people who own it. Tools and architectures come after. If the constraint is vague, the first deliverable is making it sharp enough to judge.
Define outcomes you can judge
A goal is not a slogan. We turn it into an outcome with a clear owner, a time box, and evidence that would convince a sceptical engineer. That might be latency under a load profile, a completed cutover with rollback unused, or a team shipping without us in the loop. If you cannot say what "done" looks like on a Tuesday morning, the goal is not ready.
Separate milestones from vanity metrics
Story points closed and slide decks shipped are weak signals. We prefer milestones that change risk or capability: environments reproducible, secrets out of chat, a documented decision, a monitored path in production. Vanity metrics can still appear in status updates, but they do not decide whether the outcome landed.
Name the decision owners early
Ambiguous ownership is a common failure mode on multi-team programmes. We agree who can change scope, who accepts risk, and who signs off the handoff before the first meaningful build. That sounds bureaucratic; in practice it shortens debates and keeps the technical lane clear when pressure rises.
Scope the smallest useful release
Large transformations still need a first useful cut. We look for the thinnest slice that proves the outcome direction without betting the whole estate. That slice has acceptance criteria, a rollback path, and a plan for what comes next. Big-bang goals without an intermediate proof tend to hide risk until it is expensive.
Revisit goals when reality moves
Constraints shift: vendors change, audits land early, traffic patterns surprise. We schedule short goal reviews instead of clinging to an outdated brief. Changing a goal is not failure if the evidence is written down. Quietly drifting away from the original outcome without a decision log is.