What instruction would you give an advertising agent that a competitor could not copy into theirs?
My upcoming book, Coherence at Velocity, argues that brand values have to become operating logic. Customers experience a value through pricing rules, eligibility logic, exception handling, and automated decisions. A value that never reaches a rule ends up praised in the strategy deck and contradicted at the point of sale, and the customer believes the point of sale.
On November 10, analysts at Forrester and a Harvard Business School professor published four pieces that put that argument under commercial pressure. Advertisers can now hand planning, buying, and optimization to agents that work against the objectives they state. Anything an advertiser leaves unstated is open to trade-off once the agent starts optimizing.
Treat the agent brief as the place your strategy now lives
Nikhil Lai and Melissa Bongarzone describe the redesign media teams need before agents pay off. Emily Collins and three Forrester colleagues report Amazon's pitch to sell outcomes in place of ad products, and tell advertisers to set the boundary of Amazon's optimization authority. Kasia Jakimowicz and Zachary Stone find that most organizations have AI policies, while only 42 percent of AI decision-makers say governance strategy is part of their official AI strategy. Stefan Thomke, writing in HBR, tells leaders to identify the capabilities that make their company distinct before they change how those capabilities show up.
Taken together, the authors describe one job for leaders. Before an agent takes over a decision, someone has to write down the thresholds, the limits of the agent's authority, and the parts of the brand that no optimization target can override.
Codify your planning decisions before you automate them
Lai, a principal analyst at Forrester, and Bongarzone, a senior researcher, published Agentic Advertising's Productivity Paradox on November 10. They open with the Solow paradox, the observation that heavy IT investment in the 1980s and early 1990s produced little measurable productivity gain, and cite research from 2000 estimating that for every dollar a company spent on computer hardware, it may have needed up to $9 in process redesign, restructuring, and retraining. Their line for media teams: "Layering AI onto broken processes merely automates inefficiencies." They describe three shifts. Media planning becomes continuous decisioning, and "Planners' value will shift from constructing plans to defining how and when plans should change." Buying moves from platform interfaces to stated intent. Optimization moves from campaign KPIs to lifetime value, margin, incrementality, and profitability. Their prescription includes dollar-based approval limits, brand safety constraints, frequency caps, kill switches, and first-party data made portable across platforms, because "An agent that only sees data inside one platform will optimize for that platform's yield, not advertisers' return."
Marketing Operations 3.0 gives this work a job title. Chapter 6 describes the Bot Manager, who designs the commercial policies that serve as guardrails for agent behavior and audits the logic behind an agent's action as well as the output. In my AI Capability Framework, this is the Orchestration pillar's supervision dial: a team earns each notch of autonomy by showing it can write down the rule the agent will follow. Most media plans I review still store those rules in a senior planner's head, which works until the senior planner takes a vacation.
My take: Lai and Bongarzone's $9 figure is the budget line most agent business cases leave out. Before your next agent pilot, list the ten budget moves your planners made last quarter and write the threshold that triggered each one. Whatever you cannot write down, the agent will decide for you.
Decide where your retail media partner's optimization authority ends
Collins, Lai, Brad Haag, and Fiona Swerdlow published Amazon Ads unBoxed 2026: Amazon Makes Its Pitch For Outcome-Based Advertising on November 10. Their first takeaway: "Amazon wants advertisers to buy outcomes, not ad products." Amazon Ads Agent, DVA+, and Full-Funnel Campaigns reduce the work of translating a goal into campaigns, channels, audiences, and creative. The authors warn that the same automation can move control to Amazon and reduce advertiser visibility into why budget and inventory choices were made, and they advise advertisers to define where Amazon's optimization authority starts and ends. They note that Amazon's long-term sales metric assigns value to intermediate actions such as detail page visits and video views, and does not control for outside factors or non-Amazon advertising, so advertisers should reconcile it against independent incrementality testing, media mix models, lifetime value, and first-party sales data. They also describe Branded Conversations, which lets brands supply product information that Alexa for Shopping can surface, and advise treating AI shopping as "a new consideration surface."
When the same retail media network operator runs your budget and grades the result, plan on checking the grade yourself. The Branded Conversations advice connects to Brand Visibility for Agentic Commerce: a shopping agent can only compare what you encode as structured, verifiable attributes, so "built to last" in ad copy carries less weight than a lifetime repair policy in a product field.
It reminds me of what Dave Simon, President of In-Store Marketplace (ISM), said when I interviewed him on The Agile Brand podcast: "If you buy from three different retail media networks…the methodology for determining ROAS is different by each three of those. And by the way, none of those three line up to your own internal measurement scheme… from my perspective, the brand is the one that should say, okay, guys, here's how I'm measuring you. Here's exactly how it works."
My take: Write a one-page charter for every outcome-based buy: the objective, the spend ceiling, the categories and SKUs the agent may not promote, and the holdout you will use to check the platform's own numbers. Send it to your account team before the next budget release.
Put governance inside your AI strategy, including the AI you buy
Jakimowicz, a senior analyst at Forrester, and Stone, a senior data researcher, published AI Governance Has A Strategy Problem on November 10. Citing Forrester's most recent data, they report that 42 percent of AI decision-makers say governance strategy is included in their organization's official AI and agentic AI strategy. Policies set acceptable use and responsibilities. They leave leaders without a repeatable way to decide which use cases deserve investment, which risks to accept, what suppliers must demonstrate, who approves a deployment, and where human oversight is mandatory. The authors recommend defining authority, access, oversight, and escalation before agents enter critical workflows, applying the same requirements to supplier AI, governing at each lifecycle stage, and treating AI literacy as a governance capability. "Governance cannot be a compliance check bolted onto the end of an existing process," they write.
Put their supplier point next to the unBoxed piece. Amazon Ads Agent is supplier AI acting on your budget, and in most marketing organizations nobody has run it through the governance review the internal tools received. The Identity and Permissions layer of the AI Capability Framework asks two questions of any agent: on whose behalf it acts, and how much authority it holds. Ask both of every agent a media partner operates for you.
My take: Add a column to your AI inventory labeled "operated by a partner." For every row with a yes, name the internal owner who approved its authority and the escalation path when it moves money you did not expect it to move.
Name what makes your brand distinct before an optimizer erodes it
Thomke, the William Barclay Harding Professor of Business Administration at Harvard Business School, published As Your Company Evolves, Hold On to What Makes It Unique in HBR on November 10. He opens with Ferrari's first electric vehicle, the Luce, unveiled May 25, 2026, with interior and exterior design by Jony Ive and LoveFrom. Reaction to the styling was negative, former chairman Luca di Montezemolo said the company risked "destroying a legend," and Ferrari's stock fell 8.4 percent the next day before recovering within about a week. Thomke's argument is that a company's essence sits in capabilities competitors cannot easily replicate and customers value, and that leaders "must discover their company's essence through careful questioning and experimentation" and "distinguish enduring strengths from the practices that embody them."
That distinction is the one an outcome-optimizing agent cannot make on its own. A discount, a channel, or a creative format can be a practice that expresses the brand or a habit that dilutes it, and the agent sees only which one scored higher last week. The second of the Agile Brand principles asks teams to know when evolution is needed and when revolution is, and Thomke's test is a practical way to make that call before a model makes it by default.
It reminds me of what Emily Ward, VP of Global Marketing at Turnitin, said when I interviewed her on The Agile Brand podcast: "You have to honor both the heritage of the brand and the nuances of every market. It won't work without both."
My take: Ask your leadership team to name three capabilities a competitor could not copy within a year. Then check whether any of the three appears as a constraint in an agent brief, a bidding rule, or a product data field. If none does, your brand essence exists only in slides.
Run the same quarter twice at an outdoor apparel brand
Take a hypothetical outdoor apparel company with $600 million in revenue. It sells about $80 million a quarter through Amazon, and its lifetime repair program is the capability its leadership team agrees no competitor matches. Full-price sales carry a 55 percent gross margin and promotional sales 35 percent. This quarter the CMO moves $12 million of media into Amazon's outcome-based campaigns.
In the first version, the team sets the objective as long-term sales and approves the default settings. The agent shifts spend toward promoted SKUs, where intermediate signals are strongest. Amazon's long-term sales metric reports $54 million in attributed value, and the quarterly review celebrates the number. The full-price share of Amazon revenue falls from 72 percent to 61 percent, which takes blended gross margin from 49.4 percent to 47.2 percent, about $1.76 million in gross profit gone on the same $80 million in sales. Nobody ran a holdout, so nobody can say how much of the $54 million would have happened anyway. The repair program never appears in product data, so Alexa for Shopping compares the brand's jackets with cheaper ones on price and fabric weight.
In the second version, the media director and the head of e-commerce spend two working sessions on a charter before the money moves. Promoted SKUs are capped at 25 percent of spend. Any reallocation above $50,000 waits for a planner's approval within one business day. Ten percent of target regions are held out as a control. The repair program goes into Branded Conversations and product attributes as a structured warranty field. The long-term sales metric reports a lower $46 million, the holdout test shows $19 million in incremental sales, and full-price share holds at 71 percent. The CMO reports the holdout figure to the CFO and treats the platform metric as a directional signal.
Both versions used the same budget, platform, and products. The second team wrote its brand essence into the rules before the agent started spending.
Take these actions this quarter
Write the planning thresholds down. List last quarter's budget moves and the trigger behind each one, then turn the triggers into rules an agent can follow and a person can audit.
Charter every outcome-based buy. State the objective, the spend ceiling, the protected SKUs and categories, and the approval limit for reallocations.
Hold out a control group. Reserve a share of regions or audiences on every agent-managed buy and report incremental results next to the platform's metric.
Govern partner-operated agents. Run supplier AI through the same authority, access, oversight, and escalation review as your internal agents.
Encode what makes you distinct. Put warranties, service programs, and certifications into structured product data where shopping agents can read them.
Name your three uncopyable capabilities. Check each one against your agent briefs and bidding rules, and add the missing constraints.
Measure what the platform cannot report about itself
In my consulting work with Fortune 500 marketing organizations, the hardest meeting usually came late, when two senior leaders had to agree in writing on which number the business would believe once the platform's report and the finance team's report disagreed.
Coherence at Velocity names measurement as the third enabler: measurement that shows leaders whether the customer experience is improving, kept separate from how efficiently the organization reports its own activity. Lai and Bongarzone's redesign, the unBoxed charter, Jakimowicz and Stone's supplier requirements, and Thomke's essence test each depend on that kind of measurement, because an agent working against a platform-defined score will report success in that platform's terms.
Answer the opening question before your next budget release. Write the instruction your competitor could not copy, put it in the agent brief as a constraint, and set up the holdout that will tell you whether the agent followed it.
Featured this cycle:
Agentic Advertising's Productivity Paradox. Nikhil Lai and Melissa Bongarzone, Forrester, November 10, 2026
Amazon Ads unBoxed 2026: Amazon Makes Its Pitch For Outcome-Based Advertising. Emily Collins, Nikhil Lai, Brad Haag, and Fiona Swerdlow, Forrester, November 10, 2026
AI Governance Has A Strategy Problem. Kasia Jakimowicz and Zachary Stone, Forrester, November 10, 2026
As Your Company Evolves, Hold On to What Makes It Unique. Stefan Thomke, Harvard Business Review, November 10, 2026