Every marketing leader has sat through this meeting. The team has shipped AI workflows, decks are getting drafted faster, campaign turnaround has shrunk — and then someone asks the question that stalls the room: what has this done for revenue?
A piece published on MarTech argues that the answer usually goes soft not because teams lack results, but because they lack shared language. The proposed fix is a three-layer framework that separates AI foundations, working systems and business outcomes — so you can report honestly without claiming revenue you cannot prove.
Base, builder, beneficiary
The framework splits AI work into three stages:
- Base: the foundation — clean and consistent data, documented brand and policy guidance, stable platforms, and a record of decision logic. The test is simple: is your offer terms document, brand guideline or content asset stored in one place everyone pulls from, or scattered across five near-identical versions in folders, decks and inboxes?
- Builder: the systems built on that base — workflows, agents, routing logic and automations. Think an audience segmentation workflow deciding who gets which email, or an agent deciding which support ticket needs a human.
- Beneficiary: where the business actually feels it — faster turnaround, higher throughput, lower cost-to-serve and, sometimes, incremental revenue.
Base enables builder; builder scales beneficiary. And the cycle repeats as platforms evolve and data improves. The discipline is in naming the layer: a base update is about stability, a builder update is about reliability and scope discipline, and only a beneficiary update reports business outcomes.
Why the revenue question is genuinely hard
Productivity gains from AI are well documented. A National Bureau of Economic Research study found generative AI assistance lifted issues resolved per hour by customer support agents by an average of 14%. GitHub reported developers using Copilot finished a coding task 55% faster. Stanford’s 2026 AI Index puts productivity gains at 14% in customer support, 26% in software development and 50% in marketing output.
Revenue is a different story because it is downstream, multicausal and slow to prove. Marketers know this problem intimately — it is the attribution debate all over again. First touch introduces, retargeting reminds, email brings the cart back, paid search closes. AI is now one more contributor in that chain, and rarely the only one.
The enterprise data backs up the caution. MIT’s 2025 State of AI in Business report found 95% of generative AI pilots at large companies produce no measurable return on the P&L. McKinsey’s State of AI in 2025 survey found 88% of organisations use AI somewhere, but only 39% can point to a measurable bottom-line effect, and just 5.5% attribute more than 5% of EBIT to it. BCG found 75% of C-suite leaders rank AI in their top three priorities while only 25% report significant value.
What to invest in first
The unglamorous work comes first: core documents, data definitions and a confirmed source of truth that every future agent will draw from. Isolated team knowledge can sit in lightweight retrieval; anything serving multiple teams or apps needs a shared context layer, or you will rebuild the same knowledge base five times.
Crucially, set a rule for which data source wins when two disagree, and build traceability — a context graph or equivalent record of why a decision was made. Skip it and you get outputs that sound confident but behave inconsistently, which is far harder to debug than to prevent.
As you scale, treat agents like products with owners, versioning and a replacement plan. And resist generalist tools: a narrowly scoped agent that does one thing well beats a broad one doing everything. As the article frames it, an agent that is 80% right across five domains is harder to trust than one that is 98% right in a single domain.
The lab-factory gate
To avoid choosing between showing progress now and building properly, the piece suggests a lab-factory split. The lab optimises for fast learning without production standards; the factory optimises for reliability with strict ones. Define what must be true before an initiative graduates — a stability threshold on the base, a validated builder pattern, or a sustained accuracy level. Without that gate, teams either experiment forever or push half-baked systems live.
Run both at once on different things. Adobe’s content supply chain work — unifying brand guidelines, metadata and review workflows — is base-layer investment. Coca-Cola’s Create Real Magic platform, built with OpenAI and Bain & Company, is builder-layer construction. Duolingo sits at the beneficiary layer, having said AI helped scale course content from roughly 7,100 units a quarter to over 20,500, and tying that to double-digit revenue growth in investor communications.
Why it matters for marketers
Depth beats breadth. BCG’s From Potential to Profit study found companies focused on an average of 3.5 AI use cases generated 2.1 times more ROI than those spreading effort wider. For CMOs, the practical takeaway is not to defend a number you do not have, or apologise for it. State which layer the current work sits in, and what must happen next for it to reach revenue.
Source: MarTech
