The stack in layers
The simplest useful model has five layers. Each does a job, each is served by one or more tool categories, and data moves upward from collection to activation and back down as results.
| Layer | Job | Typical tool categories | Owned by |
|---|---|---|---|
| Collection | Record events from the site, app, email, ads and sales process | Tag manager (GTMTag management system.), analytics tags, server-side endpoints, software development kits | Marketing operations with engineering |
| Data | Store events, resolve them to people, enforce definitions and consent | Warehouse, customer data platform (CDPUnified first-party customer profiles.), consent management | Data or marketing operations |
| Decisioning | Segment, score, test and automate what happens next | Marketing automation, experimentation, personalisation rules | Lifecycle or growth marketing |
| Activation | Deliver messages and offers through channels | Advertising platforms, email and messaging, on-site personalisation, customer relationship management | Channel owners |
| Measurement | Report what happened against the plan | Analytics (GA4Google's event-based analytics platform.), business intelligence, attribution | Analytics |
Few stacks are this tidy. A marketing automation platform usually spans decisioning and activation; an analytics property is both a collection tag and a measurement tool; a customer relationship management system is a data store, an activation channel and a source of truth for sales all at once. The layers are a way of asking what each tool is for, not a rule about how many tools there should be. When two tools sit in the same cell of the table for no stated reason, that is the first thing to question.
Design, data flow, tags: the core guides
The core guides are best read in order.
Designing a marketing stack is the method. It starts from the use cases the business runs or wants to run, maps each to the layers it touches, and only then asks which tools fill the gaps. It covers the build-or-buy decision for each layer, the principle of selecting for integration before features, reference shapes for teams of different sizes and a roadmap for getting from the current stack to the designed one without a big-bang migration.
Data flow and integrations follows the data. Sources and event definitions, how identity is resolved across them, the complementary roles of a warehouse and a customer data platform, the reverse flows that push segments and conversions back to channels, and the integration patterns that break most often. It is the most technical of the core guides and is written for the marketer who has to specify what engineering builds.
Tag governance takes the collection layer as an organisational problem. Tags sprawl because adding one is easy and removing one is nobody's job; the guide sets out the inventory, consent gating, performance budgets, vendor review and change control that keep the layer under control. The Google Tag Manager section covers the container itself; this guide covers the rules around it.
The tool categories section is the companion: once the stack design says which layer needs a tool, that section explains the categories and how to choose within one.
Further guides
Further guides turn the design into routines: auditing what the stack holds, writing down what it is for, reviewing who can reach it, recording how consent and data move through it, testing changes before customers see them, and keeping copies the business can restore from:
- Running a marketing stack audit — a tutorial for finding every tool, recording owners, costs and connections, mapping tools to use cases and deciding what to keep, fix or retire.
- Writing the marketing use-case list — a tutorial for collecting the jobs the team runs, writing each as an outcome, recording what it needs from data and consent, mapping it to layers and tools, reading the gaps and overlaps and keeping the list current.
- Reviewing who can access your marketing tools — a tutorial for the access register, owners and deputy admins, agencies and former staff, shared logins and what to remove.
- Mapping consent across the marketing stack — a tutorial for each purpose: where permission is captured and stored, which tools read it, how withdrawals travel and the tests.
- Documenting marketing integrations — a tutorial for one register of each connection, with its fields, identifiers, schedule, consent filter, owner and checks.
- Deciding how long marketing tools keep personal data — where copies sit, periods by purpose, suppression exceptions and deletion that reaches every copy.
- Test environments and test data for marketing tools — how to test changes before customers see them: where tests run, preview and test accounts, invented test records, reports and clean-up.
- Backing up marketing tool data and settings — which records and settings deserve a copy outside the tool, where copies live, when to take them, expiry under the retention rules and restoring without bringing back opted-out people.
Identity and consent across the stack
Two questions run through every layer and are decided nowhere in particular unless someone decides them deliberately.
The first is identity. A person visits the site on a phone, opens an email on a laptop, buys in a shop and calls support. Each system records a fragment with its own identifier: a cookie, an email address, a loyalty number, a phone number. Whether the stack knows these are one person depends on where identifiers are captured, where they are joined and which rules decide that two records match. The decision has to be made once, in the data layer, and respected everywhere else; a stack in which each tool resolves identity its own way produces segments that disagree and reports that cannot be reconciled.
The second is consent. What a person has agreed to decides what may be collected, stored, joined and used for each purpose, and the answer differs by jurisdiction and by purpose. The stack has to carry consent state from the point of collection to every tool that acts on data, which in practice means a consent management layer that tags, the data layer and activation tools all read from. The consent mode guide covers how tags respond to consent state in Google's tooling, and the privacy and consent guide sets out the general direction without being legal advice. Both questions shape the stack design before any tool is chosen, which is why the design guide treats them as requirements rather than settings.
The further guides make the consent question practical across tools. The consent map records, purpose by purpose, where permission is captured, which system holds it, which tools read it and how withdrawals and objections travel. The retention guide decides how long each tool keeps personal data, keeps suppression records on purpose and makes deletion reach every copy.
Ownership
A stack without owners decays. Tags accumulate, integrations break silently, contracts renew by default and the person who understood the warehouse schema leaves. Ownership is the simplest control and the one most often missing.
The useful unit of ownership is the layer, with a named person for each tool inside it. The collection layer belongs to marketing operations, with engineering as the partner for anything that touches the site's code. The data layer belongs to whoever can change a schema and answer for a definition, which in a small company is often the same marketing operations person and in a larger one is a data team. Decisioning and activation belong to the people who run the campaigns, because they feel the consequences first. Measurement belongs to analytics, with the KPIA metric tied directly to an objective. definitions written down in a metrics layer that the business intelligence section describes.
Ownership means four things in practice: the owner can log in with administrative rights, knows what the tool is connected to, holds the contract and renewal date, and is the person a change request goes to. An owner who can do none of those is a name on a list. Where a tool has no plausible owner, that is evidence it should not be in the stack.
The stack audit records an owner, cost, renewal date and connections for every tool and ends with a decision on each. The access review checks that each tool has an owner and a deputy administrator, that agencies and former staff no longer hold access, and that shared logins are replaced.
A travel company assigns owners
A travel booking company runs an inventory of its stack and lists 24 tools. For each it records an owner, an administrator login, the integrations and the renewal date. Nine tools have no current owner: three were bought by a marketing director who left, four are agency-managed with the agency holding the only administrative access, and two are free tiers nobody remembers signing up for. The company assigns the agency-managed tools to internal owners and requires administrative access to be transferred within 30 days, retires the two free tiers after confirming no tags or integrations reference them, and finds that one of the departed director's tools, a survey platform costing £410 a month, has not sent a survey in 14 months. The inventory becomes a shared document reviewed each quarter, and every new tool request now names an owner before procurement starts.
Common stack problems
Most stack problems arrive as a symptom in a report or a campaign rather than as a stack problem. The table maps the usual ones to the guide that diagnoses them.
| Symptom | Usual cause | Guide |
|---|---|---|
| The same metric differs between tools | No shared event definitions; identity resolved differently per tool | Data flow and integrations |
| A segment built in one tool cannot be used in another | Activation depends on exports; no reverse flow from the data layer | Data flow and integrations |
| Pages are slow and nobody knows which scripts are needed | Tag sprawl without an inventory or performance budget | Tag governance |
| Tags fire before consent is given | Consent state not carried to the tag layer | Tag governance, Consent mode, Mapping consent across the stack |
| Two tools do the same job for different teams | Stack grown by purchase rather than design | Designing a marketing stack |
| Nobody can say which marketing jobs the stack is for, so requests and renewals turn on opinion | No written use-case list or layer map | Writing the marketing use-case list |
| A new use case needs a tool nobody can choose | Requirements not written; evaluation by demo | Choosing and evaluating tools |
| Integrations break and nobody notices for weeks | No monitoring on data flows | Data flow and integrations, Documenting marketing integrations |
| Test orders are counted as revenue or sent to ad platforms | Changes tested in the live tools; test records not marked or excluded | Test environments and test data |
| A bad import or bulk edit damages records and there is nothing to restore from | No copy taken before one-way changes; copies held only inside the tool | Backing up marketing tool data and settings |
The marketing technology hub sets the stack in the context of the whole discipline, and the customer relationship management fundamentals guide covers the system that most stacks treat as the record of people.
