What Frogs And Socks Can Teach Us About AI Implementation

Full analysis

Rebecca Sykes, a partner at The Brandtech Group, spent an hour on AdExchanger Talks with host Allison Schiff making a case most operators already half-believe but rarely act on: the AI model is not the problem, the workflow around it is. The frog is fine. The sock is the issue. What's actually being decided here is where agencies and the vendors who serve them place their chips over the next year. On cheaper creative production, or on the plumbing that connects creative to media buying.

This is easy to undo at the level of any one operator. Nobody has to commit to Brandtech's pitch today. But the direction Sykes describes, creative production collapsing in cost while the integration layer becomes the moat, is hard to reverse once it's underway. There's no hard deadline, though Sykes's "connect creative to media buying later this year" roadmap claim puts a soft clock on at least one piece.

The Market Analyst Follow the money through the agency P&L. Ad resizing, asset swaps, video remixing. Those are billable hours and vendor fees today. Brandtech's Scribble agent is built to vaporize them. When a holdco agency tells you it's investing in AI creative, read it as a margin play on its own cost base before you read it as a client benefit. The more interesting line is "Generative Engine Marketing," what Jellyfish calls Share of Model. Brands are starting to fund a new budget line for how they show up inside ChatGPT, Gemini, and Perplexity answers. For an informed outsider: companies are now paying to manage their reputation inside AI chatbots the way they once paid for search ranking. That budget comes from somewhere, and paid search is the obvious donor.

The Skeptic Sykes is talking her book, and the summary flags it correctly. "There's no such thing as bad or messy data" is a sales line for a data consultancy that just rolled up 500 people into a practice selling data readiness. Garbage in, garbage out did not get repealed by large language models. For attribution, compliance, and personalization, clean data still decides whether the output is right or a confident fabrication. Watch the 70% figure too. "70% of transformation effort should go to people and process" is a consultant's framing, because people and process is what consultants bill for. The frog-and-sock fable is charming. It is also a pitch for selling change management, not technology.

The Operator The one claim here that survives contact with a Tuesday morning is the SLA story, and it's the most useful thing Sykes said. A client saved 70% of time on an individual AI-assisted task and saw zero reduction in the four-day turnaround. Why? The surrounding steps across six-plus systems never changed. Every operator who has deployed AI into a real workflow has lived this. You speed up one box in the flowchart and the deliverable arrives on exactly the same day as before, because the handoffs, approvals, and copy-paste between systems are the actual constraint. For a non-specialist: making one worker faster doesn't help if everyone else in the chain works at the old pace. This is why pilots dazzle and rollouts stall. The pilot picks one clean box. The rollout inherits the whole mess.

The Customer / End User Put yourself in the seat of a brand marketer or the DSP serving them. Scribble hands you judgment and orchestration and takes away craft execution. Fine, until you ask who learns the craft now. Sykes admits the problem and has no answer: the ability to cut through generic AI output is "the preserve of people in their 40s who had to do all of this stuff without the tools," and there's no structural substitute yet. So the brands buying into agentic creative today are quietly spending down a talent reserve they didn't build and can't replace. Her differentiation prescription is sound, though. If every brand runs the same tooling on no proprietary data, outputs converge into what she calls a sea of sameness. The brands that win encode their identity so tightly that any employee answers brand questions identically, then use cheap AI to go deliberately off-brand for exploration.

The CFO Two budget lines move if Sykes is right. Creative production fees compress, because resizing and remixing stop being a line item and become a background task. And a new Generative Engine spend appears, competing directly with paid search and SEO. The uncomfortable part: the second is unmeasured. Share of Model repeatedly queries the models and tracks how answers shift, but there's no accepted currency, no audited match rate, no agreed KPI. Operators will be asked to fund a channel whose return nobody can verify yet. That's the GEO category today. Real demand, no measurement spine, every vendor claiming the crown.

Where the council splits. The Market Analyst sees a genuine budget migration into Generative Engine Marketing. The CFO sees a channel with no agreed way to prove it worked, which caps how fast that budget can grow. The Skeptic sees most of the episode as a sales narrative for a freshly assembled consultancy. Those three are arguing about the same fact from different ends: there is real movement toward managing brand presence inside AI answers, but the measurement to govern that spend does not exist, and the people loudest about the category are the ones selling into it.

What it hinges on. Whether anyone builds a trustworthy, independent way to measure brand presence inside LLM answers. Without it, Share of Model and its competitors stay in the experimental-budget bucket, funded out of innovation lines, not pulled from paid search at scale. With it, the migration the Market Analyst describes actually happens and search budgets feel it. For the broader ecosystem, the workflow-interoperability point is the durable takeaway. The vendor that automates across the six-plus systems captures the value Sykes says is currently stranded. The one with only a clever single AI feature does not.

Direct programmatic impact: low. This is brand marketing operations and creative AI. If you run an SSP, a bidder, an identity graph, or measurement plumbing, there's a framing worth stealing here and not much to act on.

Prediction: No independent, audited measurement standard for brand presence inside AI chatbot answers (the category Jellyfish calls "Share of Model" and the industry calls Generative Engine Optimization) will be adopted as cross-vendor currency by the 2027 upfront negotiations in May 2027.

Confidence: Medium. The demand is real but the measurement scaffolding is years behind the pitch.

Why: GEO budgets are appearing now, but every tool in the category, including Jellyfish's Share of Model, works by repeatedly querying the models and reporting its own numbers, with no third-party audit and no agreed definition of a "surfacing." That is the same unverified, self-reported setup that kept earlier attention and influencer metrics out of currency status for years. Buyers do not let sellers grade their own homework when real money moves, and the model providers themselves change answer behavior constantly, which makes any fixed yardstick unstable. For an agreed currency to exist by the 2027 upfront, an independent measurer would have to solve both the audit problem and the moving-target problem in roughly 18 months, with no such body currently in the field. The opposite outcome, a settled standard that fast, is what would be surprising.

Revisit by 2027-05-31: We're right if GEO / Share-of-Model measurement is still vendor-reported with no independently audited cross-vendor standard cited in 2027 upfront deals. We're wrong if a recognized measurement body or a buy-side consortium publishes and advertisers transact against an audited AI-answer-presence currency before the 2027 upfront.