CMO playbook & advanced tactics: building the team that drives revenue

A marketing org that works at 12 people breaks at 60, and it breaks in ways you can predict and price. Twelve people share context by sitting near each other. Sixty need a written model, and every model buys one set of problems to avoid another. Centralise and you get consistency plus a queue. Embed and you get speed plus four versions of the brand. The decision a CMO owns is not which structure is correct; it is which failure mode the business can afford this year, and what it costs to unwind the wrong answer eighteen months later.

What the playbook actually arbitrates

Assume the roles and culture contract the foundations lesson sets out, and the squad, tribe and pod vocabulary the frameworks lesson defines. Assume too the hypergrowth sequencing the application lesson follows through one company. What is left for the CMO is three standing decisions: what sits in the centre and what sits inside product or region teams, how much process load a team of a given size can absorb before the process eats the output, and when a model has failed badly enough to be replaced rather than patched.

The arithmetic behind all three is unglamorous. Communication paths grow as n(n-1)/2. A 12 person team has 66 possible pairs. At 60 people it is 1,770. You cannot manage 1,770 relationships, so you buy structure to cut them, and structure has a price per head. Fred Brooks made this point about software in 1975 and it has not stopped being true about marketing departments.

Centralise or embed: the trade with numbers attached

Microsoft runs a centralised corporate marketing function under a single CMO alongside field marketing inside country subsidiaries and product marketing attached to engineering groups. That shape hardened in July 2013, when Steve Ballmer's "One Microsoft" reorganisation moved the company off product divisions and onto functions, marketing included. The functional spine survived the change of CEO in 2014. What it bought was one brand system, one set of enterprise agency and tooling contracts, and one measurement vocabulary across a company selling to almost every market on earth. What it cost was speed: in a functional org, any decision touching a single product needs a cross-functional forum, which is why Satya Nadella's visible work has been pushing decision rights down rather than redrawing boxes.

The usable rule: centralise what is expensive to duplicate and slow to change. Brand system, martechmartechThe connected set of software tools a marketing team uses to plan, run, measure and automate campaigns across channels.View full definition → contracts, data model, the definition of a qualified lead, legal and crisis authority. Embed what needs product context and weekly iteration. Launch messaging, lifecycle copy, community, partner enablement. The test is simple. If two teams working separately would produce two answers you could both live with, embed it. If they would produce two answers and you would then have to reconcile them in front of the board, centralise it.

The reconciliation cost is where most CMOs get surprised. Four embedded teams each buying their own analytics and email tooling produce duplicated licences, which is annoying, and incompatible event definitions, which is expensive. Once two teams report pipelinepipelineAll active sales opportunities across the stages of the sales process, together with their combined potential value and probability of closing.View full definition → on different attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → rules, every quarterly review starts with a forty minute argument about whose number is real, and the fix is a data project measured in quarters.

How much structure a team can carry

Zappos is the documented extreme. The company adopted holacracy from 2013 and pushed it company wide in 2015; Tony Hsieh's offer of severance to anyone not committed to the model was taken by roughly 18% of employees, about 210 people. By 2020 the company had moved off holacracy and toward teams running as internal businesses with their own budgets. Seven years, one full reversal, and a heavy attrition bill in the middle.

The second-order consequence matters more than the headline number. Attrition under a structure change is not random. The people with the most external options leave first, which means you lose the senior operators who could have made the new model work and keep the people who cannot easily move. A reorg that sheds 18% of headcount typically sheds far more than 18% of institutional knowledge.

Give yourself a structure budget and count it in hours. Take every recurring ritual on the calendar, multiply by attendees. Twenty people in six hours of standing meetings a week is 120 hours, three full-time equivalents spent on coordination. That may be a fair price at 60 people. At 15 it is a third of your capacity, and the team will quietly stop attending, which is worse than not having the ritual because now you have a governance model nobody follows.

Reading the break before it costs you

Spotify's squad and tribe language spread further than the practice ever did. Jeremiah Lee, who worked there, wrote publicly in 2020 that Spotify itself never fully implemented the model and that the matrix left delivery accountability ambiguous, with the lead layer carrying more than it could hold. Companies copied the diagram from a 2012 paper and inherited the ambiguity without the culture that had been absorbing it.

Four signals that your model is failing, before the results show it:

• Two teams announce competing things in the same week. Nobody owns the calendar, which is a symptom of embeddingembeddingAn embedding is a numerical vector that represents data (text, images, or items) in a way that captures meaning, so similar items sit close together in space.View full definition → without a central traffic function.
• Your own calendar fills with tie-breaks. Escalation volume is the cleanest measure of unassigned decision rights.
• Contractor and agency spend rises while headcount is flat. Embedded teams are buying their way around a central bottleneck, at roughly double internal cost.
• Attrition concentrates in one layer rather than spreading evenly. Usually the lead layer, which is where matrix load lands.

The headcount premium of embedding

Embedding multiplies the number of people who can run a plan without supervision. Six embedded pods need six operators of that calibre, not one senior plus five juniors, because there is no adjacent desk to check the work. That premium is real money and it is the constraint most CMOs discover after they have announced the new structure. A centralised team can carry a higher junior-to-senior ratio precisely because supervision is cheap when it is co-located.

So the arbitration runs backwards from budget: if you cannot fund the senior layer, do not embed, whatever the org chart fashion says. The compromise everyone reaches for, a dotted line to the centre and a solid line to the business unit, has its own well-known failure mode. Two managers, one calendar, and the person in the middle optimising for whoever does their review.

Real-world cases with numbers

Microsoft abandoned stack ranking in November 2013, after years of reporting, most visibly Kurt Eichenwald's 2012 Vanity Fair piece, that forced curves pushed employees to compete inside their own teams. The lesson for a CMO is that the performance review system is part of the org model. You can embed pods and preach shared pipeline goals, and a curve that ranks people against their teammates will beat both.

Spotify cut about 17% of its workforce in December 2023, roughly 1,500 people, and Daniel Ek's memo was explicit that too many people were supporting work or doing work around the work rather than producing output. That is a public admission about the ratio of coordination to delivery, from the company whose org diagram everyone else copied.

CMO action items

• Write the centralise/embed list explicitly, item by item, and publish it. Brand, data model, tooling, lead definitions, launch messaging, regional adaptation. Every unlisted item defaults to argument.
• Cost your coordination load in hours per week per person, then decide out loud whether the output justifies it. Cancel one ritual and see who notices.
• Before any restructure, name the two metrics that will tell you within a quarter whether it worked, and the date you will look at them. Reorgs without a review date never get reversed, they get renamed.

Common mistakes that kill results

• Reorganising to fix a measurement problem. If nobody agrees what a qualified lead is, no box on the chart repairs that, and you will have spent six months of disruption to find out.
• Copying a structure from a company with a different constraint. The org that invented the model was solving for engineering autonomy at a particular headcount, in a particular culture, and often was not running the published version itself.
• Treating activity as accountability. Tracking blog posts and campaigns rather than pipeline and retention trains the team to look busy inside whichever structure you pick.
• Leaving no sunset clause. Every model has a headcount range where it works. Say out loud what that range is, so the next change is a scheduled decision instead of a crisis.