What Does Your AI Know That Your Competitors’ AI Doesn’t? Coherence at Velocity | October 14, 2026
When every competitor can run the same model on the same public data, what does your AI know that theirs does not?
In my upcoming book, Coherence at Velocity I write that as the marketing stack moves toward identity, decisioning, orchestration, and AI-mediated action, companies gain advantage by governing context, policy, and next action across their systems. Four sets of authors published on that question between Monday and Tuesday. Read together, they give you a test for every AI program on your 2027 plan: what proprietary context feeds it, and what evidence shows it changed results.
On Saturday I wrote about who records the outcome when an agent completes a purchase for your customer. Those outcome records are part of the context these authors describe, and today's edition covers what you do with them.
Read the four arguments as one budget question
Five McKinsey authors propose an enterprise "AI operating system" and tell CEOs to own their context while renting or replacing the underlying technology. Richard Shotton reviews research showing that basic AI prompts push creative ideas toward the most common human answer. Giovanni Gavetti and three coauthors argue in HBR that AI gives every competitor the same analysis, so strategists have to find knowledge their rivals cannot copy. Ian Whittaker tells marketers to prepare the evidence behind every return figure before the 2027 budget meetings that start in the next few weeks.
All four describe the same exposure. A team that feeds a shared model shared inputs gets an average answer, and then has to defend the spend on that answer to a CFO who can buy the same model.
Fund one shared context layer before you fund another use case
The AI operating system: A new way for CEOs to run the enterprise, published October 13 by Asutosh Padhi, Brian Elliott, Daniel Swan, Kim Baroudy, and Vik Sohoni of QuantumBlack, starts from a gap in McKinsey's 2026 State of AI survey. Forty percent of respondents at large companies report scaling AI agents. Six percent attribute significant financial impact to AI, which McKinsey defines as at least 5 percent of EBIT, roughly the same share as the year before.
The authors' answer is an enterprise system that holds business context, reasons across functions, recommends or executes decisions, and learns from outcomes. Their governing rule is to own the context and rent or replace the technology. They tell leaders to limit the number of these systems to between one and three, because a separate stack for every function or use case breaks the integrated view the system exists to provide. Their first actions include picking three to five high-value domains, funding one shared system across use cases, and redesigning decision rights so everyone knows which decisions agents make, which need human judgment, and when an issue escalates.
Put one pair of figures from the piece on your next steering agenda. Token unit costs have fallen more than 95 percent, and AI spend at many companies has risen more than three times. Your CFO will put those two numbers side by side before you do.
I read the 40 and 6 percent figures as an absorptive capacity problem. In Coherence at Velocity I define absorptive capacity as the enterprise's ability to recognize new information, build it into how work gets done, and apply it in a way that changes performance. Teams that scale agents faster than they redesign workflows and decision rights produce activity, and their EBIT stays where it was. Molly Innes reported in Marketing Week on October 13 that marketing leaders at top-performing businesses are four times more likely to restructure their teams for AI and twice as likely to work from a centralized AI roadmap. I read that as the same finding from inside marketing.
It reminds me of what Emma Robinson, Head of B2B Marketing at Canva, said when I interviewed her on The Agile Brand podcast: "You can consider generic AI a bit like having an intern with no context, whereas brand-aware AI is more like a coach who knows your playbook and then constantly learns from it."
My take: In my AI Capability Framework, Memory and Context is a layer every capability depends on, and the McKinsey authors make the same call at the CEO level. Count the context stores your marketing AI programs draw from today, and if the number exceeds three, start consolidating before you add a fourth use case.
Change how your team prompts before you judge AI creative work
Richard Shotton's October 12 column, Why It Doesn't Work: AI and the 'beige trap' of creativity, opens with an exercise. Ask ChatGPT to pick a number between one and 10 and you will probably get seven, and so will your competitors.
Shotton cites a 2024 study by Oliver Hauser of the University of Exeter and Anil Doshi of UCL. Among writers who scored lower on creativity, evaluators rated stories written with AI help 27 percent better written and 23 percent more enjoyable, and the researchers found those stories 9 to 11 percent more similar to each other. He then cites University of Pennsylvania research led by Lennart Meincke. Basic prompts produced ideas 55 percent less diverse than a group of students brainstorming on their own. Telling the model to adopt a persona barely moved the score, and asking it to think like Steve Jobs made the ideas 9 percent less diverse than the basic prompt. A chain-of-thought prompt brought diversity within 5 percent of the human group.
Shotton concludes that how your team prompts matters as much as whether it uses AI, and that AI output works best as a starting point.
My take: You can run a cheap version of the Meincke experiment. Have one team run your next concept round with a chain-of-thought prompt and another with your current default, then score both sets for distinctiveness against your three closest competitors' recent work.
Move your strategists' time from dashboards to customers
In What Sets Great Strategists Apart in the AI Era, published October 13, Giovanni Gavetti, Hart Posen, and James Siderius of Dartmouth's Tuck School and Zia Zaman of Dalhousie University start from Michael Porter's premise that strategy means being different. Every competitor can now pull the same analytics, frameworks, scenario models, and data from AI tools, so competitors reach similar conclusions and lose their edges quickly.
The authors argue that distinctive strategy will rest on what they call sticky knowledge: insight that people find hard to articulate, transfer, or separate from its setting. They cite Starbucks, Apple, and MetLife, and they tell leaders to move time from analytical reviews into direct work with customers, employees, and unfamiliar settings.
Most quarterly business reviews I sit in contain more slides about customers than minutes spent with them.
It reminds me of what Ann Davis, Chief Revenue Officer at Crunchbase, said when I interviewed her on The Agile Brand podcast: "You have to assume everybody is going to be using that same public data. So find what your edge is and how you can drive more pipeline and conversion into revenue by seeking out data sources others don't have or haven't learned to use effectively."
My take: Gavetti and his coauthors write about corporate strategy, and the same logic applies to the inputs your marketing AI uses. Sticky knowledge reaches your models only when someone writes it down in a form a system can read, so assign that work to a named person in marketing operations.
Show finance your working before they ask for it
Ian Whittaker spent 20 years as an equity analyst. In The 5-55 rule: Marketers must show their working to ensure credibility, published October 13, he describes investor meetings where he spent five minutes on his conclusion and the other 55 defending how he reached it. He tells marketers to prepare for the same ratio in the 2027 budget meetings happening over the next few weeks.
His illustration starts with a slide claiming £4 in return for every £1 spent. If the £4 is revenue and the business runs a 40 percent gross margin, the return becomes £1.60 of gross profit, and after the original £1 you have 60p left before any other costs. He cites Adidas, which put 77 percent of its budget into performance media under last-click attribution and later found through econometrics that brand activity drove 65 percent of sales. He also points to Kraft Heinz, which wrote $15.4 billion off the value of its brands and goodwill in February 2019, to argue that cuts to advertising deserve the same scrutiny as increases.
Apply Whittaker's arithmetic to an AI program. Take a hypothetical specialty retailer whose team reports that an AI personalization program returned $5 in revenue for every $1 of its $800,000 annual cost, or $4 million. At a 45 percent gross margin, that revenue becomes $1.8 million of gross profit. A holdout test shows that 40 percent of those customers would have bought anyway, which leaves $1.08 million of incremental gross profit. Subtract the $800,000 and the program earns $280,000, or 35 cents per dollar. The team can defend a 35-cent return in a budget meeting.
Now suppose the retailer's two largest competitors license the same personalization vendor and feed it the same third-party segments. Expect their holdout results to converge on the retailer's, and expect the retailer's CFO to ask what the company gets for its $800,000 that competitors do not. The answer has to come from first-party context, such as purchase history, service records, and returns data, which those competitors cannot license.
My take: Run an incrementality test on every AI program you plan to renew, and bring the margin-adjusted figure to the budget meeting yourself. In my book, Marketing Operations 3.0 I frame return on agents as an executive question: is the system creating enough commercial value to justify its cost, complexity, and governance burden? Use Whittaker's arithmetic to answer it.
Take these actions before your 2027 budget closes
Inventory your context. List every data source your marketing AI programs draw from, and mark the ones a competitor could license from the same vendor.
Cap the number of context stores. Following the McKinsey authors, aim for one to three shared systems, and route new use cases into them.
Test your prompt defaults. Run Shotton's comparison on your next creative round and score the output for distinctiveness.
Book customer time for strategists. Move one analytical review per month into direct customer or frontline contact, and require a written note from each session.
Margin-adjust every return claim. Convert revenue to gross profit, subtract the baseline from a holdout, and present the remaining figure before finance calculates it for you.
Answer the opening question with evidence
I have recorded more than 900 podcast conversations with senior enterprise leaders. The detail I use most often in client work comes from the part of a conversation where a guest explains how a decision actually got made inside the company. Nobody publishes that detail, and a model trained on public text has never seen it. Gavetti and his coauthors would call it sticky knowledge, and every company holds some in its customer records, service logs, and frontline experience.
In Coherence at Velocity I describe measurement as the way leaders separate real improvement from well-documented activity. Apply it to the opening question. Before your 2027 budget meeting, write down what your AI knows that your competitors' AI does not, name the data source behind each item, and attach a holdout result to each one. If you cannot fill the list, fund the context work before you fund another use case.
Featured this cycle:
The AI operating system: A new way for CEOs to run the enterprise. Asutosh Padhi, Brian Elliott, Daniel Swan, Kim Baroudy, and Vik Sohoni, McKinsey, October 13, 2026
Why It Doesn't Work: AI and the 'beige trap' of creativity. Richard Shotton, Marketing Week, October 12, 2026
What Sets Great Strategists Apart in the AI Era. Giovanni Gavetti, Hart Posen, James Siderius, and Zia Zaman, Harvard Business Review, October 13, 2026
The 5-55 rule: Marketers must show their working to ensure credibility. Ian Whittaker, Marketing Week, October 13, 2026