Why did enterprises move to composable?
For a long time, an enterprise website meant buying a large suite. One vendor delivered the CMS, search, personalisation, analytics and often commerce as a single product, with a licence to match. Upgrades were projects in their own right, and replacing one part meant replacing all of it.
Composable was the answer to that. Instead of one suite, you pick a specialised product for each job (a CMS, search, a media library, personalisation, hosting) and connect them through APIs. For many of our clients it delivered what it promised: faster releases, and the freedom to replace one part without rebuilding the rest.
By 2025 that choice had stopped being unusual. Gartner made composable architecture a requirement for inclusion in its 2025 Magic Quadrant for digital experience platforms. The question for our clients has moved from whether to go composable to how many vendors they want to compose with.
What does a stack of many vendors cost?
37% of 600 large enterprises named integration complexity as a barrier to their AI projects (MACH Alliance, 2026).
The number matters, because each vendor bills on its own unit. The CMS counts users and API calls, search counts requests and records, the media library counts credits, the customer data platform counts tracked users, personalisation counts active profiles, and front-end hosting counts function invocations and bandwidth. Every one of those needs its own forecast, and has its own price when the forecast is wrong.
The invoices are only part of the cost. Every connection between two products is code someone has to build, monitor and update when either side changes its API. In the MACH Alliance’s 2026 survey of 600 large enterprises, 37% named integration complexity as a barrier to their AI projects, and 34% traced AI failures or delays back to implementation and integration.
Can composable come from one vendor?
Some of our clients have reacted by going back to an all-in-one suite. That solves the invoices and the integration work, but it brings back the problem composable was meant to fix: one part can’t be replaced without the rest. What they want is one contract for the parts that belong together, without giving up the APIs between them.
That is possible, because composable describes how the parts connect, not how many companies you buy them from. What matters is that each part has its own API and can be swapped out on its own. Several of those parts can come from one vendor, under one contract, as long as that stays true. We call that composable inside a platform.
Which parts should you buy, and which should you integrate?
That makes buy or integrate a decision per connection, not per stack. Some parts only need a feed from the others. Search is the clearest case. Published content goes into the index, and nothing has to come back. A connection like that is cheap to build and rarely changes. Other parts need each other’s data all the time, in both directions. Those connections cost the most to build and maintain, and they are the ones we would rather buy.
What are the three levels of personalisation?
Personalisation is the clearest example of the second kind. It needs three things to work together: the content, the segments, and the customer data the segments are built on. How many of those a client buys together depends on the product, and we see three levels.
- 1
The first is a CMS on its own. It holds the content, and the segments and the customer data come from other products. Every connection between them is integration work.
- 2
The second is a CMS with personalisation built in. Prepr is our choice at this level for mid-market programmes. Its segments are built on what a visitor does on the site, such as the pages they viewed and where they came from, and editors pick the content for each segment in the CMS. Personalisation is part of its €299 a month plan. Contentful Personalization does the same at enterprise scale. It tracks visitors itself and is sold as an add-on, billed on monthly active profiles.
- 3
The third level includes the customer data, and this is where Contentstack sits.
When is a CMS with personalisation built in enough?
At the second level the connection between content and segments is bought, but the customer data is not. That is fine for a client that wants to show returning visitors something different from new ones. Once the segments need orders, CRM data or email behaviour, that data has to come from a customer data platform somewhere else.
Prepr says so itself, and advises clients to start with Prepr and add a customer data platform where it adds value. Contentful offers connectors that bring profiles in from products such as Segment. Either way it is an integration again.
What does a platform with customer data add?
At the third level, Contentstack launched the combination in February 2025 as EDGE, which it called an adaptive DXP. In June 2026 EDGE became AXP, short for agentic experience platform. The parts are largely the same. What changed is who is expected to do the work. A DXP assumes that editors write the content and marketers build the segments. AXP puts AI agents at the front of that work, with people setting the rules and approving what goes live.
AXP comes in three parts, under one contract and each with its own API. Content Cloud holds the CMS, the assets and the front-end hosting. Data Cloud holds personalisation and Lytics, the customer data platform Contentstack bought in January 2025. Agent OS holds the AI agents and the automations. Lytics takes in data from the website, from email and commerce systems and from data warehouses, and matches an anonymous visitor to a known customer. Its audiences sync into personalisation automatically, and each personalised experience shows up in the CMS as a set of variants of the normal entry. Nobody builds a sync job or a mapping table between the three.
Contentful is moving to this level by another route. Salesforce bought it in 2026 and plans to combine Contentful’s content with its own customer data platform and agents. How that will work has not been published yet.
AI agents raise the stakes. An agent that writes a page, picks an audience and publishes has to cross every vendor boundary between the content and the customer data, and each one is a place where it can fail. Contentstack’s agents start on events in the CMS, and its assistant works with the same rules and permissions as the person using it.
What does one platform cost you in flexibility?
Buying all three together has a price. Each part has its own API, but the parts depend on each other. Personalisation takes its audiences through Lytics and is billed on the events Lytics records, so the CMS, the data platform and personalisation are in practice one decision. A client that already runs another customer data platform can only connect it through Lytics. Personalisation and Lytics are enterprise options, and some of what we describe here is still in early access.
What should you keep integrating?
For the connections that only need a feed, we still integrate, and so does Contentstack. Its marketplace and automation recipes connect it to commercetools for commerce, Algolia for search and Cloudinary for media. We think that is the right split. Search and commerce are parts where a specialised product does more than a built-in module, and where the connection is cheap to own.
So which level fits your organisation?
Our advice is to buy together the parts that need each other’s data all the time, and to integrate the rest. Which level that means depends on the data the personalisation needs. If it only needs behaviour on the site, a CMS with personalisation built in is enough, and a platform would mean paying for a connection the client doesn’t use. If it needs customer data from several systems, or AI agents working with it, the content and the customer data belong on one platform.
The choice then follows where that data already lives. A client that runs on Salesforce has it there, and Contentful is the content side of that platform. A client that doesn’t can buy both from Contentstack. A client with no plans for personalisation can choose its CMS on its own merits.
Book a call with us