CMO playbook & advanced tactics for customer research

Twelve teams, one customer base. Brand wants concept testing for the autumn campaign. Demand genDemand genCreating and stimulating demand for your offer, often upstream of the buying process to generate interest and awareness before prospects are ready to buy.View full definition → wants message testing on four ad variants. The enterprise PMM wants loss interviews on last quarter's three biggest misses. Product wants usability sessions before a redesign ships. Pricing wants a sensitivity study on the new packaging. All five will approach the same few hundred accounts, three will hire different recruiters, and at least two will produce a deck that changes nothing. At any real scale the binding constraint on customer research is not budget. It is customer patience, synthesis capacity, and the number of decisions your organisation can absorb in a quarter. Your job is to ration all three.

The operating model: one queue, one owner

Below about fifty people, research coordination happens in a group chat. Above it you need an intake queue with a named owner sitting inside product marketing, a shared panel, and a repository that anyone can search before they commission anything. The search step alone kills a meaningful share of incoming requests, because the answer already exists in a study from eight months ago that nobody outside the requesting team ever saw.

Unilever sells roughly 400 brands across 190 countries. A brand team in Indonesia and a brand team in Brazil separately buying the same category study is not a rounding error at that scale, it is a structural tax. Unilever's response, from around 2016, was to build People Data Centres that pool consumer and social signal centrally instead of letting each brand and market procure its own. The centralising move buys efficiency and creates a new problem, which is the next thing you have to decide.

Arbitration: whose question gets funded

Every request should carry four things before it enters the queue: the decision it will inform, the person who owns that decision, the date the decision gets made, and the money exposed if it goes the wrong way. Requests that cannot name a decision are rejected. That single rule removes more work than any budget cut, and it changes behaviour faster if you publish the rejected list with the reason attached rather than quietly declining.

Two refinements matter. First, weight by reversibility. A message test on a campaign you can rewrite in a week deserves less of your research capacity than a packaging change you will live with for two years. Second, reserve roughly a fifth of capacity for questions nobody asked. The highest-value study is often the one with no internal sponsor: the churn cohort that belongs to no team, the no-decision losses that never reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.View full definition → a sales debrief because nobody logs a deal that simply went quiet.

The cost of asking, and who you keep asking

A fifteen-interview study with senior enterprise buyers costs tens of thousands of dollars once you count recruiting fees, incentives running to hundreds of dollars per hour of a senior buyer's time, moderation and loaded analyst hours for synthesis. Synthesis, not fieldwork, is usually the bottleneck: transcripts pile up faster than anyone codes them.

Because it is expensive, teams reuse the accounts that answer the phone. Design partners, the advisory board, the eleven logos in every case study. Those customers have been trained by your own team over years and answer in your vocabulary, which makes their agreement worthless as evidence. Cap any account at two research contacts per year across all teams and track it in the CRM where account managers can see it. Skip that and the account team eventually blocks research access altogether, at which point you keep the happy customers and lose the loss interviews, which is the opposite of what you need.

The other lever is method economics. Behavioural telemetry answers what happened at almost zero marginal cost per additional user, and the split the foundations lesson draws still governs which questions each side can answer. Spotify built an internal experimentation platform to run large numbers of concurrent tests and later packaged it for outside customers as Confidence, so treat their account of the practice as a vendor describing its own product. The leadership rule underneath it is cheap: never buy an interview for a question a query can answer, and never accept a dashboard as an answer to a question about motive.

From finding to decision: the activation gap

Demand three artifacts and refuse to fund a study that does not update one of them: a living ICPICPKey Performance Indicator, a measurable value that shows how effectively you're achieving a specific objective, tracked over time against a target.View full definition → with behavioural detail beyond firmographics, a messaging hierarchy that ranks propositions by segment and funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.View full definition → stage, and a competitive positioningpositioningThe mental space you want your brand to occupy in your target customer's mind relative to alternatives.View full definition → mapmapUsing software to automate repetitive marketing tasks and campaigns, enabling personalisation at scale across channels like email, web, and social.View full definition →. Each gets a named owner and an expiry date. Anything past its date comes down, because a stale artifact is more dangerous than no artifact.

The activation gap is where most research operations die. If your research function reports somewhere with no authority over messaging, findings land in a wiki and stay there. The second-order consequence is worse than waste: sales builds a shadow deck from field intuition, and within two quarters you are running two competing versions of the customer truth with no way to adjudicate between them.

What it costs to get wrong

Unilever's centralised model creates an arbitration a CMO has to run personally: global insight against local reality. A pooled finding about a category holds at the aggregate and can be flatly wrong in a market where the shopping trip, the pack size and the retailer are different. Centralise the infrastructure, the panel and the repository; leave the interpretation with the market team that has to sell the thing. Get that boundary wrong in the direction of the centre and you ship a campaign that tests beautifully in aggregate and misfires in your third-largest market.

Notion shows the opposite failure. Its community of template makers and ambassadors is loud, generous and immediately available, which makes it the cheapest research input in the company and the most biased. Those users are power users of a tool they already love; they are not the finance director signing an annual contract. Notion's buyer profilebuyer profileA semi-fictional, research-based representation of your ideal customer: their goals, frustrations, behaviours and decision criteria.View full definition → also moved fast, from individual productivity users around 2020 towards team leads and operations managers as the company pushed into paid workspaces and reached a $10 billion valuation in 2021. A team running 2022 campaigns off 2020 research would have addressed the wrong person with the wrong promise, using evidence that felt strong because it came from the most enthusiastic people in the market.

CMO action items

• Run one intake queue with a named research owner inside product marketing. Every request names the decision, the decision owner and the date. No decision, no study, and the rejection is published with its reason
• Keep a contact register capping any account at two research asks a year across all teams, visible to the account owners, so research access does not get shut off by sales
• Ring-fence about a fifth of capacity for questions with no internal sponsor, starting with churn and with deals that ended in no decision
• Give every artifact an owner and an expiry date, and put a fast lane on the queue: small questions answered inside 72 hours, so teams have no reason to go around you

Common mistakes that kill results

The deck as deliverable. A study that costs thirty thousand dollars and changes no price, no page and no roadmap item has a return of zero, and it is invisible because the team still felt busy. Audit backwards once a quarter: list the decisions made, then find which study fed each one. The studies that appear nowhere tell you what to stop commissioning.

Over-centralising until the queue becomes a bottleneck. When a message test takes six weeks to schedule, teams buy a cheap survey tool and run it themselves, badly, and you now have confident numbers from a leading question asked of the wrong sample. That is worse than the delay you were trying to fix.

Validation theatre. If interviews are run mainly to confirm what the executive team already decided, you have a compliance exercise. The tell is easy to spot in the artifact: research that never once forced a repositioning conversation is not research.

Studying only buyers. Everyone who reached your interview list already engaged with you. The people who evaluated and chose nothing, or never entered the funnel at all, hold the explanation for the flat quarter, and in enterprise pipelines no decision is routinely a bigger loss category than any named competitor. Budget for the hardest recruit you have, because that is where the unexpected finding lives.