CMO playbook & advanced tactics for MarTech stack architecture

Your predecessor signed a three-year enterprise agreement. It renews in seven months, the account team has already offered a double-digit discount for adding two modules, and three regional teams are quietly paying for tools that do part of what those modules already do. Nobody in the building can tell you what the whole thing costs, because a chunk of it sits on departmental cards. That is the CMO version of stack architecture, and no framework resolves it for you.

Gartner's CMO spend research has tracked utilisation of purchased martech capability falling to roughly a third, down from well over half in 2020, while martechmartechThe connected set of software tools a marketing team uses to plan, run, measure and automate campaigns across channels.View full definition → keeps absorbing about a quarter of the marketing budget. A stack used at a third is not a procurement accident. It is a run of decisions nobody made on purpose.

The decisions that do not delegate

Your team can mapmapUsing software to automate repetitive marketing tasks and campaigns, enabling personalisation at scale across channels like email, web, and social.View full definition → every tool to the four layers the foundations lesson sets out, and the methodology lesson hands them capability scoring, the total cost of ownership model and a replatform sequence. None of that answers the questions that land on your desk: which vendor gets to be the anchor, what you stop paying for, what you do about the tools you never approved, and whether a consolidation is worth the months of degraded output it costs.

Scott Brinker's martech landscape passed 11,000 products in 2023, so "there is a better tool for this" will always be true. The scarce resource is integration labour and the attention of the people who actually ship campaigns. Every arbitration below is an argument about where that scarce resource goes.

Sub-Concept 1: Suite or best-of-breed, decided on structure not preference

The honest input to this decision is your operating model, not a feature matrix. One brand, one catalogue, one consent regime, a central marketing function: a suite anchor is defensible, because the integration work you avoid is worth more than the flexibility you give up. Lego sells one global brand on roughly DKK 65 billion of revenue through lego.com, more than a thousand branded stores and the Insiders loyalty programme. The customer definition is the same everywhere, so central mandates hold.

Now take the opposite shape. Siemens runs marketing across business units selling industrial hardware, software and healthcare into procurement cycles measured in years, in around 190 countries with different works councils and privacy postures. A single mandated activation platform there is a fiction that regional teams will route around within two quarters. The workable enterprise pattern is asymmetric: mandate the identity and consent layer, the warehouse and the metric definitions; devolve activation to the units, with a floor. A unit running under a dozen campaigns a quarter should not own its own sending infrastructure, because the deliverability risk outweighs the autonomy.

Note who is talking when you hear this argument. Adobe and Salesforce sell the suite; Twilio Segment and Snowplow sell the hub layer; the systems integrator advising you bills for the integration either way. And the real lock-in is not the contract, it is the data model. Exporting your segmentssegmentsDividing a market into distinct groups of customers who share similar needs, characteristics or behaviours, so each group can be served with a tailored approach.View full definition → is easy. Exporting the logic that built them, the ten years of campaign history and the identity resolution rules is a project, not an export.

Sub-Concept 2: Licence rationalisation, and the seat audit that lies

Start with SSO logs, not with vendor-reported usage. Ninety days of logins gives you monthly active seats per tool and a cost per active seat that survives contact with a CFO. Expect to find whole licence blocks bought for a campaign in 2022 and never reassigned.

Then apply the correction, because a seat audit systematically lies about two categories. Consent management, deliverability monitoring, tag governance and data quality tooling have almost no seats and enormous blast radius: cut them and you find out in a regulator's letter or a sender reputation collapse. The second category is the tool with four users that produces the number your board deck runs on.

The trap in the other direction is the renewal ratchet. A discount for adding modules raises the committed floor you renegotiate from next time, and consumption pricing converts your "saving" into overage the first time a campaign works. Two practical rules: price every renewal against the walk-away cost, not last year's invoice, and never let your two anchor contracts renew in the same quarter. Aligned renewals feel tidy and destroy your leverage on both.

Sub-Concept 3: Shadow tooling and the amnesty that beats a ban

Large enterprises typically run a few hundred SaaS applications, and SaaS management vendors consistently find a large share of that spend sitting outside IT. Marketing is usually the worst offender, for a rational reason: the sanctioned path takes six weeks and a ticket, while a card and $49 a month takes an afternoon before a launch.

The exposure is not the money. It is a vendor holding customer email addresses with no data processing agreement, no deletion path when a subject access request arrives, and no connection to your consent record, so people who opted out keep getting messaged from a system your DPO has never heard of.

Bans do not work here; they push the same behaviour into spreadsheets on laptops, which is worse because it leaves no audit trail at all. What works is a 30-day amnesty: register everything, no blame, no budget clawback. Then two gates for anything touching customer data, a signed processing agreement and SSO. A vendor that cannot do SSO does not get personal data, whatever the demo showed.

Sub-Concept 4: What a failed consolidation actually costs

Consolidation business cases are written on the licence line and paid for everywhere else. Price these five before you sign.

• Parallel run. You will pay both vendors for six to eighteen months, because you cannot switch off the old system until the last reporting dependency is gone.
• Implementation labour. Integration and migration fees routinely land at a multiple of the annual licence, and internal engineering time rarely appears in the case at all.
• Deliverability. New sending domains and IPs need weeks of warming; botch it and revenue from email drops during the exact quarter you promised savings.
• Comparability. Move platforms and you lose like-for-like history, so year-on-year reads are unreliable for around a year, which is also the year you are being judged on the decision.
• Key people. Two or three people understood the old system's quirks. Consolidations are when they leave, and their knowledge was never written down.

Then add the calendar constraint nobody raises in the steering committee: a retailer whose sales concentrate in the fourth quarter has one safe migration window, roughly January to March. Miss it and the honest answer is to wait a year, not to compress the plan.

Real-world cases

Case 1: Lego is the shape of company where suite consolidation is the right call. One brand, one product catalogue, direct channels it owns end to end, and a loyalty programme that gives it a first-party identifier across web, app and stores. The constraint is not architecture, it is timing. Toy demand concentrates hard into Q4, so the consolidation calendar is set by the trading calendar, and any programme that cannot finish cutover before autumn should not start.

Case 2: Siemens is the opposite shape, and the methodology lesson follows its selection work. The leadership version of the same problem is federation: you cannot mandate one activation platform across units whose buyers, sales cycles and regulators differ, so you mandate the layer underneath and audit compliance with it. Siemens also sells enterprise software itself, which adds a wrinkle most CMOs do not face: internal platform choices double as reference-customer politics.

Case 3: Twilio bought Segment for about $3.2 billion in 2020, and by early 2024 was running it as a separate business unit while publicly reviewing its options for it. Nothing about that is unusual in software. It is a warning about anchoring your architecture on a single vendor's continuity. Whoever your hub is, contract for portability now, while you still have signing leverage: schemaschemaA schema is the formal blueprint that defines how data is structured, named, typed, and related within a database, file, or message.View full definition → and event history export in a documented format, defined exit assistance, and a written statement of what happens to your data if the product is sold.

CMO action items

• Pull 90 days of SSO logs and build one page: tool, annual cost, monthly active seats, cost per active seat, layer owned. Flag anything under 20% active seats, then manually exempt the low-seat, high-blast-radius tools before anyone starts cutting.
• Build a renewal calendar 18 months out and deliberately stagger your two anchor contracts into different quarters.
• Run a 30-day shadow tooling amnesty, then enforce two gates on anything touching customer data: signed processing agreement and SSO.
• Require every consolidation case to state the parallel-run months, the deliverability plan, and a named rollback point with a date. A programme with no rollback point is a bet, not a plan.

Common mistakes that kill results

Mistake 1: Consolidating to save licence fees. The licence line is the visible cost and usually the smaller one. If the case does not model integration labour, parallel run and lost campaign throughput, you are approving a number that has no relationship to what you will spend.

Mistake 2: Killing shadow tools without funding the job they were doing. The tool existed because a team had a real gap and no route through procurement. Remove it, leave the gap, and the spend reappears within two quarters, or the work moves to spreadsheets where you cannot see it.

Mistake 3: Accepting the bundle discount and never turning the modules on. This is precisely how utilisation drifts to a third: modules committed for a discount, no owner, no launch date, no metric. Make every added module name an owner and a go-live date in the contract approval itself, and treat an unlaunched module at renewal as a licence to hand back.

Key takeaways

• Suite versus best-of-breed follows your operating model, not preference. One brand and one consent regime favour a suite anchor; divergent business units favour a mandated data layer with devolved activation.
• Utilisation around a third is a governance failure, and bundle discounts are its most common cause.
• Seat audits mislead in two directions: they overvalue rarely used consent, deliverability and governance tooling, and they undervalue nothing at all.
• Shadow tooling responds to amnesty plus gates, never to bans. The real exposure is customer data at a processor your DPO cannot name.
• Price the parallel run, the deliverability warm-up, the lost comparability and the key-person risk before you approve a consolidation, and write down the rollback point while you can still use it.