Most large organisations replatform their content management system roughly every five to seven years. Ask why and you will get an answer about changing requirements. Look at the actual triggers and a different pattern emerges: the majority of replatforms are not chosen. They are forced.
An end-of-life date arrives. A vendor gets acquired and the roadmap changes. A major version turns out to be a rebuild rather than an upgrade. The platform hits a ceiling at multi-brand scale that was not visible during the pilot. In each case, the organisation is paying seven figures to end up roughly where it already was.
Breaking that cycle is a procurement problem more than a technology problem, and it is worth being systematic about it.
The four forcing functions
Forced end-of-life. This is the most common and the most predictable. One major enterprise vendor moved all extended-support versions to paid support and security patches from June 2026. Another ends managed-services support in August 2026, with core support for its long-standing on-premise version ending February 2027 and no in-place upgrade path to its cloud product. A widely deployed open-source platform reached end of life in January 2025 with much of its installed base still running on it. In every case the customer’s choice is to pay, rebuild, or run unpatched.
Acquisition and roadmap drift. When a content platform is acquired by a larger company, its roadmap starts answering to a different set of priorities. That is not necessarily bad for the acquirer’s existing customers; it is rarely good for customers who bought a focused best-of-breed product and find it folded into a suite. The relevant procurement question is not whether a vendor is currently independent, but what happens to your roadmap if they stop being.
Suite accumulation. Several enterprise platforms are federations of acquired products presented as one. One vendor’s stack was assembled from four major acquisitions over eleven years. Another has made eight acquisitions since 2015 and is owned by a private equity firm. The pattern shows up in the reviews: customers report that integration between the marketing and content products is “very basic and limited,” or that connecting two modules of the same platform requires developer-configured OAuth and webhooks. You are buying integration risk inside a single vendor.
Architectural ceilings. The quietest one. A platform performs perfectly for one site, one market, one brand, and requires custom architecture at fifteen sites, twelve markets, and four brands. Nothing has failed; the design simply assumed a smaller problem.
What this costs
Published figures are worth having in front of you when someone argues that replatforming is a normal cost of doing business.
Traditional enterprise DXP licensing runs $200,000 to $500,000 a year and upward before implementation. Implementation itself commonly runs $250,000 to $500,000, with large programmes reaching seven figures. Median annual contract values sit around $61,000 for one major vendor and $83,000 for another across sampled deals, with enterprise arrangements several times that. Five-year total cost of ownership above $800,000 is routine, and that is before the rebuild the end-of-life notice will require. Systems integrator rates for the relevant specialists run $160 to $300 an hour.
Against that, the cost difference between platforms is usually a rounding error compared to the cost of choosing one that forces you to do this again in five years.
Buying criteria that actually reduce the risk
Rankings and roundups of the best enterprise CMS platforms are a useful starting point for building a shortlist, but they compare capabilities at a moment in time. Durability is what you are actually buying. Six criteria predict it better than a feature grid.
Is the content portable? If your content is stored as structured, typed data accessible over open APIs, leaving is an export. If it is stored as rendered markup in a proprietary repository, leaving is a project. Ask to see the export before you sign, not after.
Is the front end genuinely decoupled? A platform that does not own your presentation layer cannot trap you in it. This is the single most effective anti-lock-in measure available.
Is governance included or upsold? Single sign-on, audit logs, approval workflows, environment permissions, and accessibility compliance are core enterprise requirements. At several vendors they sit behind the top tier. Compare the tier you would actually buy, and confirm the certifications the vendor holds in its own right — ISO/IEC 27001, SOC 2 Type II, HIPAA where relevant — rather than inherited from a cloud provider.
Is the pricing legible? Metered API calls, per-repository charges, per-locale limits, and consumption-based content requests all create the same problem: the bill scales with your success in ways you cannot forecast. One widely used platform caps localisation at eight locales even on a $675-a-month tier. Model your third-year volumes, not your first-year ones.
How large is the talent pool? Platforms built on a single proprietary stack limit hiring to specialists in that stack, at specialist rates. Platforms that leave the front end to you draw on whatever your team already knows.
Does the upgrade path involve a rebuild? Ask directly what happened to customers during the vendor’s last two major version transitions. The answer is a matter of public record and is more informative than any roadmap slide.
Where agentic capability fits into this
There is a temptation to treat AI capability as a separate evaluation track. It should not be, because it is now one of the main sources of ceiling risk.
A platform whose content is stored as page components will support AI assistance well and operational automation poorly, because agents reason over structure rather than layout. A platform where the content model can only be changed in code will cap what automation can do for business teams. And a platform whose agents orchestrate only across the vendor’s own product suite will deliver value in proportion to how much of that suite you buy — which is the suite-accumulation problem arriving through a new door.
Comparing platforms on the best agentic CMS dimension is therefore not a separate exercise from the enterprise evaluation; it is a test of whether the underlying architecture will still be adequate in three years. The things that make an agentic layer work — structured content, permission-bound execution, full attribution and reversibility, an open protocol layer for reaching other systems — are the same things that make a platform survivable at enterprise scale generally.
The uncomfortable conclusion
Most replatforming programmes are not caused by a bad platform choice. They are caused by a choice made on the wrong criteria — features and demos rather than portability, pricing legibility, governance inclusion, and upgrade history.
The organisations that stop replatforming every five years are not the ones that picked the most capable system. They are the ones that picked a system they could leave, and then found they did not need to.
