Can you still migrate before the renewal?
The most expensive sentence in IT is "we ran out of time, renew for another three years." Pick your migration path and renewal date; the planner back-schedules every phase from the deadline and gives you an honest verdict, plus calendar events you can drop straight into the team's calendar.
Pick your renewal date to see whether a ~15-week Azure OpenAI → Llama (Meta) migration still fits before it, and what has to start when.
Why plan backwards from the renewal?
Migrations that race a renewal date fail in a predictable way: the assessment and design phases run long because nobody scheduled them against a deadline, and by the time the pilot proves the runbook, the only options left are renewing anyway or cutting over without one. Back-planning inverts that: it fixes the finish (your renewal, minus a safety buffer), sizes the phases from the same per-path effort estimates our migration guides use, and surfaces the real decision, the required start date, while you can still act on it.
The production-migration phase deliberately gets the largest share of the timeline: waved cutovers with verification between waves are where estates actually spend their weeks. If your verdict comes back "tight" or "late", the corresponding migration guide covers what compressing each phase actually costs in risk, and the calculator models whether a short-term renewal beats a rushed migration on 3-year cost.
The date that matters is not the renewal date
Almost every enterprise agreement carries a notice period: a number of days before renewal by which you must give written notice, or the term rolls over automatically. Thirty, sixty, and ninety days are all common, and the clause is usually several pages away from the date everyone has memorised. That notice deadline, not the renewal itself, is your real finish line, and teams routinely discover it a week after it has passed.
So before you trust any plan, including this one, go and read two things in the contract: the notice window, and whether notice must be given in a particular form to a particular address. Then plan to the earlier date. If the renewal is 31 March and notice is due 90 days prior, your deadline is 31 December, and a plan that finishes in February was never a plan. The Renewal Redline scanner exists partly to surface exactly these clauses from text you paste in.
One further subtlety worth building into your dates: giving notice is not the same as having migrated. Notice is reversible in most commercial relationships, and serving it early to protect your optionality, while continuing to negotiate, is a normal and often sensible move. Deciding not to serve it because the migration might slip is how a three-year renewal gets signed by default.
Reading the verdict honestly
A comfortable verdict means the phases fit with buffer left over. The risk here is complacency: a plan with slack tends to consume it early, in the assessment phase where nothing appears to be at stake. Set the start date anyway and treat it as real.
A tight verdict means the arithmetic works and nothing can go wrong, which is not the same as workable. The usual response is to reduce scope rather than compress phases: move the awkward twenty per cent of the estate into a second, later wave and migrate the straightforward eighty per cent before the deadline. A partial migration that removes most of the licence count is worth considerably more than a complete migration that misses the date entirely.
A late verdict is genuinely useful information, and it is better to have it now than in month four. It usually points to one of three moves: negotiate a short extension rather than a full multi-year term, which is a far easier ask when you can show a credible dated plan; split the estate and migrate what fits; or accept the renewal and use the extra year properly, with the start date already in the calendar. What a late verdict should not produce is a compressed plan that pretends the phases are shorter than they are, because the phase that gets silently cut is almost always validation, and that cost lands after the project is declared finished.
Whichever verdict you get, the estimate rests on a per-path effort model rather than on your specific estate, so treat it as a planning skeleton. If your inventory is unusually clean, or unusually full of raw device mappings and clustered workloads, adjust accordingly, and use the estate analyzer or your own discovery to find out which of those you are before you commit to a date.