Building a martech stack: layers, choices and vendors

A martech stack is the set of tools a marketing team uses to know its customers, reach them, make what it says to them and learn what worked. This piece helps with one recurring decision: what to add, replace or remove, and on what terms.

• Four layers Data (who the customer is and what they agreed to), engagement (the channels that send and receive), content (what is said and shown) and measurement (what worked). Consent and preferences run through all four.
• A crowded market The 2025 landscape published by Scott Brinker and Frans Riemersma counted 15,384 solutions, 9% more than the 14,106 of a year earlier, while the previous generation of vendors consolidates and AI-native firms arrive^1, 2.
• Use lags spend Gartner's 2025 Marketing Technology Survey put the utilisation of stack capabilities at 49%, and only 15% of organisations qualified as high performers³.
• Integration and ownership decide value more than any feature list: who owns each layer, which system holds the reference copy of each piece of data, and how you would leave.

Data. The records everything else depends on. A CRM holds the people and accounts that sales and service work with. A customer data platform, in the definition the CDP Institute has kept since 2016, is packaged software that builds a unified, persistent customer database that other systems can use; the Institute added its RealCDP checklist in 2019 so buyers could test whether a product calling itself a CDP could do the expected jobs⁵. Many companies also keep a data warehouse that finance and analytics already trust. The choice between them is the subject of first-party data and CDPs.

Engagement. The channels that send and receive: email, messaging, social, paid media, on-site and in-app personalisation. Each has its own rules and its own primer: email, messaging, social, digital advertising. Engagement tools also collect data, and that collection is regulated: in its guidelines adopted in October 2024, the European Data Protection Board treats a tracking pixel in a web page or an email as falling within Article 5(3) of the ePrivacy Directive, the rule on storing or reading information on a person's device¹⁰.

Content. The website and its CMS, the digital asset library, design tools and the brand's templates. This layer keeps the brand consistent across every channel; see consistency and adaptation.

Measurement. Web and app analytics, attribution, marketing mix modelling and experiments. What each method can and cannot tell you is in measuring effectiveness.

The directory classifies every platform it lists by category, so you can see a layer's options side by side: customer data platforms, CRM, marketing automation, web content and DXP, asset management and web analytics.

A stack is worth what moves between its layers. For every connection, write down three things: what moves (events, profiles, audiences, consent), in which direction and how often; which system holds the reference copy of each field; and who owns the connection when it breaks.

Three kinds of connection deserve particular care.

• Collection Tags in the browser send events to each vendor. With server-side tagging, the browser sends one request per event to a server the company controls, which then forwards what each vendor needs and can remove personal data first; cookies the server sets on the company's own subdomain are first-party cookies⁶.
• Consent signals A choice made in a banner must reach every tool. Google's consent mode lets its tags change behaviour according to the consent state⁷, and in European advertising IAB Europe's Transparency and Consent Framework is the common standard for passing people's choices to vendors⁸. Details are in consent, identity and cookies.
• Collaboration A data clean room lets two companies match and analyse their data for agreed purposes. The IAB Tech Lab's guidance, released as version 1.0 in July 2024, sets out common principles, functions and privacy-enhancing technologies, along with their limitations⁹.

Write the split down with the integration RACI; how marketing operations runs it day to day is in marketing operations and governance and on the marketing operations hub.

There are four ways to fill a layer.

• Buy a suite one vendor covers several layers. Fewer integrations to run; more dependence on one roadmap and one contract.
• Buy best-of-breed the strongest tool for each job, joined by integrations you maintain.
• Compose use the company's own data warehouse as the data layer and connect tools to it, often called a composable or warehouse-native approach. It suits companies that already run a warehouse and have engineers to own it.
• Build write your own software. Brinker describes a fast-growing "hypertail" of custom software¹, and AI makes small internal tools cheaper to write; they are not cheaper to maintain.

Whatever the route, count all of the cost: licences, implementation, integration, the people who run the tools, and the cost of leaving. People are often the binding constraint. In Gartner's survey of 413 martech leaders in mid-2025, half said their organisation lacked the technical and data readiness for AI agents, and half cited a shortage of technical talent⁴.

Check that the vendor will still be there. With more than fifteen thousand products on the market and a previous generation consolidating^1, 2, acquisitions and retired products are part of the risk. Ask who owns the company, what happened to the customers of products it merged or withdrew, and what the contract says if the product is sold or discontinued. A short notice period and a complete data export protect you better than any reassurance in a sales meeting.

1. Write three to five use cases as jobs, with the data each needs: "send a basket reminder within a day to people who agreed to email". Features follow from jobs, not the other way round.
2. Shortlist two to four. Find your stack asks three plain questions and explains its matches; Compare puts two to four platforms side by side, for example two customer data platforms.
3. Check the six criteria the directory publishes for every listing (transparency, interfaces, delivery, consent, data and commercial terms), described in the methodology.
4. Ask for a demonstration of your use cases on your own data, not a scripted demo. Treat AI claims with care: 45% of martech leaders piloting or running vendor AI agents told Gartner the agents fell short of the promised business performance⁴.
5. Read the exit terms before the price: data export in a usable format, notice periods, renewal and who owns derived data.

An illustrative business, not a real company: a regional retailer with a CRM, an email tool, a website on a CMS and web analytics. It wants to send messages based on what people browse and buy, online and in store.

• The jobs (1) A basket reminder by email within a day, only to people who agreed to email. (2) One view of each loyalty member's online and in-store purchases. (3) A monthly report that links campaigns to sales.
• The audit Two tools hold contact data, with different consent fields; the analytics tags fire before the banner is answered; nobody owns the CRM-to-email sync.
• The decisions Consent first: the CRM becomes the reference for consent, and tags wait for it. For job 2, the team compares the CRM vendor's add-on with a CDP, shortlists two CDPs in the directory, and runs a time-boxed pilot on real data against the three jobs. Job 3 goes into the measurement framework before any tool is chosen.
• What the team does not do It does not buy a new email tool: the existing one can handle job 1 once consent is fixed.

• Buying features, not jobs. A long feature list with no written use case is how stacks end up under-used.
• Two systems of record for the same field, especially consent, so that a person who unsubscribed in one tool is still mailed from another.
• Ignoring the exit. A tool that cannot hand back all your data in a usable form owns your customer record.
• Forgetting the people. A tool without an owner, training and time decays; Gartner's 49% utilisation figure is the result³.
• Adding a tool to fix a process. If briefs, approvals or data entry are broken, a new tool automates the problem.
• Judging a tool by its licence alone. Implementation, integration and the people who run it usually cost more over its life, and they belong in the comparison from the start.

1. Draw your stack on the four layers and mark the system of record for customer data, consent, assets and results.
2. List every connection: what moves, in which direction, how often, and who fixes it.
3. Name one owner per layer in the integration RACI.
4. Retire or fix one under-used tool before buying another.
5. For the next purchase, write three to five use cases, shortlist with Find your stack and Compare, and test on your own data.
6. Put the cost and the expected effect in a business case, with the exit terms beside the price.

• The size and shape of the market: Brinker and Riemersma's 2025 landscape¹ and CMSWire's report on it².
• How stacks are used: Gartner's 2025 Marketing Technology Survey³ and its survey of martech leaders on AI agents⁴.
• The building blocks: the CDP Institute's definition⁵, Google's guides to server-side tagging⁶ and consent mode⁷, IAB Europe's consent framework⁸, the IAB Tech Lab's clean room guidance⁹ and the EDPB's guidelines on tracking techniques¹⁰.
• Next in this track: first-party data and CDPs, consent, identity and cookies and AI in marketing.