Who Decides What Your Customer Hears Next? Coherence at Velocity | October 16, 2026
When marketing, sales, and customer success each deploy their own AI against the same customer, who decides what that customer hears next?
In my upcoming book, Coherence at Velocity, I describe the coherence paradox. The moves that make an enterprise faster, including automation, distributed decision-making, and local optimization, also raise the risk of inconsistency unless leaders scale shared data, rules, and operating logic alongside them. One team tunes a threshold to lift conversion, another reinterprets a policy to cut service load, and the customer ends up as the first place anyone reconciles the two.
In Wednesday's edition I argued that your proprietary context is the asset competitors cannot license. Authors at Forrester, Marketing Week, and McKinsey published on Wednesday and Thursday about the next problem: each function feeding that context to its own agents under its own rules.
Treat this week's reading as one operating-model decision
Laura Cross of Forrester tells RevOps leaders to map the revenue decisions AI now influences and assign decision rights for each one. John Buten, also of Forrester, tells portfolio marketers to turn the judgment behind their battlecards and messaging documents into living guidelines that agents across departments draw on. Josh Stephenson reports for Marketing Week on how Giffgaff's CMO and CTO spent two years learning to start from the problem before choosing a tool. Eight McKinsey authors tell leaders to rank AI investments by economic value and test each business case against higher AI prices.
Taken together, these authors lay out a sequence. Decide who owns each decision that more than one system touches, write down the judgment that owner applies, give technology the problem along with the objective, and price the result at the costs you will pay at renewal.
Map the revenue decisions your AI touches before another team buys an agent
Laura Cross, VP and principal analyst at Forrester, published her case for redefining revenue operations on October 15. She starts from the work AI already performs or assists with in RevOps, including forecasting, reporting, and lead routing, and tells RevOps leaders to stop treating those deliverables as the job. In her account, the job becomes designing how people, data, processes, technology, and AI combine to make and carry out revenue decisions.
Cross describes the coherence paradox from the revenue side. Marketing deploys AI to optimize campaigns and buying-group engagement, sales deploys it to prioritize accounts and recommend actions, and customer success deploys it to predict risk and trigger interventions. Each team can make a sound choice, and the company still ends up with conflicting priorities, inconsistent processes, and a fragmented customer experience. Cross tells leaders to govern the business decisions AI influences along with the technology, and she lists three steps: map decisions alongside processes, define where AI may recommend, act on its own, or require human approval, and redesign the RevOps capability model.
In Marketing Operations 3.0 I describe RevOps as a useful step with a limited scope. RevOps teams aligned the front office around shared pipeline data, and most stopped short of a shared state and shared policy across marketing, sales, service, product, and post-sale work. Teams that follow Cross's three steps move RevOps into that territory. If your leadership team needs a common starting definition, the Agile Brand Guide entry on revenue operations covers the function as most companies run it today.
Count your agents before you build the decision map. On the same day, three of Cross's Forrester colleagues, Joe Stanhope, Rusty Warner, and Jessica Liu, published five criteria for what qualifies as a marketing AI agent: marketing-focused, autonomous, responsible, goal-oriented, and able to execute multiple tasks in an adaptive workflow. They wrote it because vendors have blurred the term. Run the five criteria against every tool sold to your teams this year as an agent, and expect a few of the ones with friendly first names to fail the autonomy test.
Take a hypothetical $400 million B2B software company. Marketing runs an agent in its automation platform that scores buying-group engagement and enrolls contacts in sequences. Sales runs an agent in the CRM that ranks accounts for reps each morning. Customer success runs a churn model that triggers retention plays. The marketing operations lead lists every revenue decision those three systems influence and finds 23. Seven involve the same accounts and at least two of the systems, and the team can name a single owner for three of the seven.
One of the unowned decisions covers what to offer an existing customer next. A mid-market account sits 60 days from renewal. Its support tickets doubled last quarter, so the churn model scores it as high risk and the success playbook sends a 15 percent retention discount. Three of its users attended a product webinar, so the marketing agent enrolls them in an upgrade sequence. The CRM agent places the account in the rep's top 10 expansion targets, and the rep calls to pitch the premium tier.
The customer's operations lead now holds a discount offer, an upsell sequence, and a sales call from the same vendor, all dated the same week, and each team configured its system correctly for its own target.
The marketing operations lead can fix this decision in one meeting. Assign the next-offer decision for accounts inside the renewal window to the customer success lead. Let the marketing and sales agents recommend for those accounts and let only the success playbook act. Require a person to approve any discount above 10 percent. The team has applied Cross's second step to one decision, and six more remain on the list.
My take: Cross writes for RevOps leaders, and the decision map belongs to whoever in your company can convene marketing, sales, and success leaders in one room. If nobody holds that authority today, put the question on the next CMO and CRO agenda.
Write down the judgment behind your marketing deliverables so every agent uses the same version
John Buten, principal analyst at Forrester, published his assessment of portfolio marketing's next AI phase on October 15. He cites Forrester's July 2026 survey of portfolio and product marketers in its B2B Marketing Online Community: 95 percent use AI to create marketing and sales content, 83 percent use it to gather competitive intelligence, and 78 percent use it for buyer persona research. Buten calls most of those productivity gains defensive, because they mainly help teams keep pace with faster launch schedules and more requests for segment-specific content.
He moves portfolio marketers from producing battlecards, personas, and messaging documents toward maintaining the expertise and hard choices those documents represent, in a form agents can use. A sales rep asks an agent to help compose an answer to a difficult buyer question. A demand team refines campaign materials with a messaging agent. Both draw on guidelines the portfolio team keeps current, and Buten tells portfolio leaders to build those living guidelines with RevOps, sales leadership, and demand teams.
Buten picks up where Cross's decision map ends. Once you assign a decision to an owner, someone has to write down the owner's judgment in a form every agent that recommends or acts on that decision can read. In Marketing Operations 3.0 I call this knowledge engineering: organizing the company's proprietary truth, from value propositions and pricing to brand guidelines, into a semantic layer agents can query, so buyers get specific, verifiable information that adds decision value.
It reminds me of what Chris Taylor, Head of Brand Experience at DDN, said when I interviewed him on The Agile Brand podcast: "An archetype, you know, focuses more on the operational aspect of a brand. They tell you how to make decisions, which is really helpful, whether they're large or small… They're not really rules to follow. That's much more scalable than just being the brand police…"
Write your living guidelines the way Taylor describes his archetype, as a statement of how the brand decides, which a person or an agent can apply to cases the authors never listed.
Commentary: Pick the three deliverables your sales team requests most often. Write down the reasoning behind each one in plain language, assign an owner and a review date, and point every agent that drafts customer-facing copy at that document.
Bring your technology partner the problem, and keep the objectives in marketing
Josh Stephenson reported for Marketing Week on October 15 on a Festival of Marketing session where Giffgaff CMO Georgina Bramall and CTO Steve McDonald described two years of bringing their functions closer. Giffgaff marketers now bring McDonald's team the problem before they name a solution, and McDonald often challenges the brief before anyone starts building. The company created a team it calls Mar-Tech, which pairs marketing specialists with engineers, data scientists, and data engineers. The team reports into technology, and marketing sets its objectives.
The arrangement took six to nine months to work well. Early on, people left the same meeting with conflicting accounts of what the two sides had agreed, so Bramall and McDonald started attending meetings together to resolve misunderstandings and to show their teams how to handle disagreement. Stephenson cites Marketing Week's 2026 Career & Salary Survey, in which 4.6 percent of respondents named the CMO-CTO relationship as one to nurture, placing it last among common C-suite relationships. That ranking matches most martech intake queues I review, where requests arrive as a vendor name and a deadline.
Giffgaff applied the same approach to AI. Before commissioning any custom tools, the company gave staff access to several large language models and asked each person to spend five hours a week for a month experimenting, with tutorials from the technology team. McDonald wants staff to understand where the tools fall short, and he notes that teams get more reliable results from conventional automation on some tasks. McDonald is applying the rule behind the mechanism axis in my AI Capability Framework: use the lightest mechanism that clears the task's accuracy bar, and expect to pay for anything heavier twice, in compute and in verification.
The joint meetings put Agile Brand Principle 3 into practice: make adaptivity and collaboration part of the culture. Bramall and McDonald modeled the behavior for six to nine months before their teams picked it up.
My take: If you move your martech team under technology, write marketing's authority to set its objectives into the team charter, along with who settles disputes between the two leaders. McDonald says he has watched technology teams spend significant time on tools that made little business difference, and Giffgaff wrote the objective-setting rule to prevent it.
Rank your AI use cases by value, and price them at higher costs before you renew
Lieven Van der Veken, Sandra Durth, Tanguy Catlin, Tarik Alatovic, and Tom Isherwood of McKinsey, with Dhanush Thirumoorthy, Dmitrii Kazantsev, and Shaahid Nanabhay, published The AI value curve: The economics of applying AI to existing work on October 14. They open by observing that many organizations still measure AI adoption, counting users, copilots, and tokens, and leave value unmeasured.
Their value curve compares the cost of completing a task today with the full cost of completing it using AI, including model and token costs plus any human review, validation, and rework the output requires. They put the annual global opportunity at $2.52 trillion, counting only tasks where AI already costs less than people at current prices and wages. They then stress-test the estimate: in an illustrative scenario, a 30 percent rise in AI costs shrinks the global pool from $2.52 trillion to $1.67 trillion, a 34 percent drop, and pulls $320 billion off a $960 billion US baseline. The authors tell leaders to map value at the task level, rank AI investments by area under the curve, and run AI FinOps as a standing discipline.
On the same day, Gartner analyst Will Sommer gave buyers a reason to take the higher-cost scenario seriously. Gartner analysts estimate that AI providers could face impairment on more than $3.5 trillion of infrastructure spending by the end of 2030. To cover planned spending, financing costs, and a 12 percent return on invested capital, data center owners would need about $8.5 trillion in cumulative revenue, and Gartner analysts forecast no more than $3.6 trillion in AI revenue attributable to data centers. The analysts also expect more than 80 percent of neocloud providers to stop operating independently by 2030. For a marketing buyer, that adds up to pricing pressure and vendor consolidation in the layer underneath your martech contracts.
Run the McKinsey stress test on the software company from earlier. Suppose its three agents cost $300,000 a year in combined model and platform usage, and the team spends a third of that on each system reconciling the same duplicated and conflicting account records every time it runs. A 30 percent price increase at renewal adds $90,000, and $30,000 of the increase goes to cleanup the company could fix once in a shared data layer.
It reminds me of what Ben Schein, Chief AI and Analytics Officer at Domo, said when I interviewed him on The Agile Brand podcast: "Clean upstream and you fix a definition once, in a pipeline, deterministically, and every run afterward inherits it free. Clean downstream and you pay a probabilistic model to guess at the same mess on every run, forever. One is a repair. The other is a subscription to your own sloppiness."
My take: Add a cost-sensitivity line to every AI business case in your 2027 plan, and list the reconciliation spend each system carries. Use the McKinsey team's 30 percent figure as the minimum increase you test.
Take these actions before your next agent goes live
Count your agents. Apply the Forrester five-criteria test to every AI tool marketing, sales, and success run, and record the vendor and the team that own each one.
Map shared decisions. List the revenue decisions those systems influence and flag every decision that two or more systems touch.
Name one owner per shared decision. Specify where each system recommends, where it acts, and where a person approves.
Write the judgment down. Turn your most-requested deliverables into living guidelines with an owner and a review date, and point every customer-facing agent at them.
Brief technology with the problem. Send the next martech request as a problem statement with the customer outcome attached, and leave the tool choice open.
Stress-test renewal prices. Re-run each AI business case at 30 percent higher cost and separate out the spend on reconciling bad data.
Answer the opening question by name, one decision at a time
As a photography major I printed in a darkroom, and I learned to write exposure time, paper grade, and dodging notes on the back of any print I wanted to make again. Without those notes I could not reproduce my own work a month later, and nobody else in the lab could reproduce it at all.
In Coherence at Velocity I write that agents operate more literally than people. A service rep who sees a loyalty record that looks wrong checks a second system and uses experience to reach a reasonable outcome. An agent acts on the data, rules, and permissions it receives. Your teams cover the gaps between the marketing, sales, and success agents with informal repair work today, and that cover ends once those agents act without a person in between.
Before your next agent goes live, answer the opening question in writing. List the revenue decisions your AI systems touch, name one owner for every decision more than one system influences, and have each owner write down the judgment they apply so every agent works from the same notes.
Featured this cycle:
AI Won't Save RevOps — It Will Redefine It. Laura Cross, Forrester, October 15, 2026
Portfolio Marketing's AI Honeymoon Is Over — Now The Real Work Begins. John Buten, Forrester, October 15, 2026
'Start from the problem': How Giffgaff brought marketing and technology closer together. Josh Stephenson, Marketing Week, October 15, 2026
The AI value curve: The economics of applying AI to existing work. Lieven Van der Veken, Sandra Durth, Tanguy Catlin, Tarik Alatovic, Tom Isherwood, Dhanush Thirumoorthy, Dmitrii Kazantsev, and Shaahid Nanabhay, McKinsey, October 14, 2026
Also referenced: So You Think You Have A Marketing AI Agent?, Joe Stanhope, Rusty Warner, and Jessica Liu, Forrester, October 15, 2026; Gartner Says Over $3.5 Trillion of AI Infrastructure Spending Could Be Exposed to Impairment by the End of 2030, Gartner, October 14, 2026.