gantt title Target state — 18 months dateFormat YYYY-MM-DD axisFormat %b %y section Foundation Landing zone and connectivity :done, f1, 2026-01-01, 60d Identity federation :done, f2, after f1, 45d Observability baseline :active, f3, after f1, 75d section Decouple Strangler facade at edge :active, d1, 2026-03-15, 90d Extract order domain : d2, after d1, 120d Extract pricing domain : d3, after d2, 90d Retire shared order database :crit, d4, after d3, 60d section Data Ingest to raw zone : a1, 2026-04-01, 75d Conformed customer entity : a2, after a1, 90d Retire nightly extract jobs : a3, after a2, 45d section Resilience Multi-zone for tier 0 : r1, 2026-06-01, 60d Automated failover rehearsal : r2, after r1, 45d Second region for tier 0 :crit, r3, after r2, 90d
What it is
Architectural change plotted against time, grouped into a small number of themes, with dependencies drawn as dependencies rather than implied by position. Each bar should be justifiable in terms of an outcome — a risk retired, a cost removed, a capability unlocked — because that is what the funding conversation is about.
When you produce it
When asking for multi-period funding, when a target state has been agreed and someone reasonably asks how you get there, and at the start of each planning cycle to re-cut it against what actually happened.
Who reads it
Executives and finance, who fund it. Delivery leadership, who resource it. Engineering teams, who need to see their work in a wider sequence. Vendors, who should see only the part relevant to them.
What good looks like
- Four or five themes at most. A roadmap with eleven lanes is a backlog.
- Dependencies are explicit, so the critical path is visible.
- Retirement bars exist. A roadmap that only adds things is a wish list; the saving comes from the decommissioning.
- Bars get vaguer further right, and the diagram admits it rather than implying the same confidence at month eighteen as at month two.
- Each theme has an outcome sentence attached in the accompanying text.
Common mistakes
- Technology names as bars. "Adopt Kubernetes" is not an outcome; "reduce environment provisioning from six weeks to one day" is.
- No decommissioning. Then the estate only grows and the business case never lands.
- False precision in the far future. Fixed-width bars in month twenty tell the reader you are guessing, but with confidence.
- Ignoring the delivery capacity that has to run alongside the change.