AI productivity is easy to demonstrate. Revenue impact is harder. When revenue questions come up in leadership updates, the answers often turn vague, focusing less on numbers and more on signals. A three-layer framework helps you show what AI delivers at each stage and what must happen before its impact reaches the revenue line.
The same meeting is happening at many companies right now, where one senior leader briefs another on where the team stands with AI. The team has shipped things with AI, and work is visibly moving faster. But revenue is a lagging indicator, and attribution can be difficult. Rather than forcing every AI initiative into a revenue metric, it’s more helpful to talk about what’s happening at each stage of the AI workflow.
The framework uses three layers: base, builder, and beneficiary. Base establishes the foundation, builder reflects the AI systems and workflows built on it, and beneficiary captures the business outcomes. This framework won’t claim revenue that isn’t provable yet or undersell real gains.
A 3-layer framework to discuss AI outcomes
Base forms the foundation
This AI workflow layer includes consistent data, documented brand and policy guidance, stable platforms, and a clear record of decision logic. Think of an offer’s terms, a brand guideline, or a piece of content. It’s stored in one place everyone pulls from, not scattered across five slightly different versions in campaign folders, decks, and inboxes.
It determines whether anything built with it still holds together six months later. Skip it, and the builder-layer agent you ship next quarter might look perfect in the demo, but start giving confidently wrong answers the first time it runs into a problem.
Builder reflects AI systems
This layer is where the construction happens via workflows, agents, routing logic, and automations. Think of an audience segmentation workflow, identifying who gets which email, or a routing agent deciding which ticket needs human oversight.
An update on the builder layer confirms whether the workflow’s getting more reliable and whether the scope is staying disciplined. It doesn’t cover revenue, but it’s still important to the business.
Beneficiary captures outcomes
The beneficiary layer is where the business feels the impact of AI via shorter turnaround, higher throughput, lower cost-to-serve, and sometimes incremental revenue. It answers the ROI question.
10X your SEO with Semrush for Enterprise.
The world’s most powerful SEO platform, purpose-built for Enterprise.
Request demo
How the framework elements connect
Base enables builders, while builders scale beneficiaries. The sequence is ongoing, as platforms evolve, data improves, expectations shift, and you cycle through all three layers again on the next initiative.
By naming which layer a given update belongs to, you effectively communicate the nature of the initiative.
- A base update is about stability.
- A builder update is about reliability improvements and scope discipline.
- A beneficiary update reports on business outcomes.
Why AI revenue questions are so challenging to answer
AI is a strong, proven productivity lever. Teams are shipping, drafting, and reviewing faster — especially with coding tasks. AI tools generate more accurate output that requires less human oversight.
A National Bureau of Economic Research study found that access to a generative AI assistant raised the number of issues that customer support agents resolved per hour by an average of 14%. GitHub found that developers using GitHub Copilot completed a coding task 55% faster than those without it. Stanford’s 2026 AI Index asserts that productivity gains range from 14% in customer support to 26% in software development to 50% in marketing output.
Yet measuring revenue gains from AI remains challenging. Revenue impact is downstream, multicausal, and slower to prove.
Revenue attribution challenges echo marketing attribution limitations
Marketing has been navigating a similar issue for years. Standard attribution tools and models can’t always settle the question of which channel deserves credit for a sale.
Often, several channels contribute in an order that’s hard to reconstruct after the fact.
- First touch gets the introduction.
- Retargeting gets the reminder.
- Email gets the customer back to the cart.
- Paid search closes the deal.
AI is now a contributor to revenue-generating outcomes. But it’s rarely the only one.
AI contributes little measurable revenue at many enterprises
AI’s revenue problem surfaces in every major study of enterprise AI spend. For example, the MIT 2025 State of AI in Business report found that 95% of generative AI pilots at large companies produce no measurable return on the profit and loss statement (P&L), even as adoption climbs.
The McKinsey State of AI in 2025 survey found that 88% of organizations now use AI somewhere in the business, but only 39% can point to any measurable effect on the bottom line. Just 5.5% attribute more than 5% of their earnings before interest and taxes (EBIT) to it. BCG research found that 75% of C-suite leaders rank AI among their top three priorities, but only 25% say their organization realizes significant value from it.
I’ve watched a lot of leadership updates on AI, and only a handful of the CMOs in the room have explicitly drawn the line between productivity improvements and revenue attribution. This often reflects a gap in shared language rather than a lack of understanding.
How to get the resources you need for AI initiatives to work
This resourcing conversation can sound like asking for time and budget with nothing to show yet. It comes down to a short list of unglamorous investments that matter more than it might seem.
Where to invest first
Establish core documents and data definitions first. Confirm the source of truth that every agent and workflow will eventually draw from.
Be deliberate about where you store information. Lightweight retrieval works fine for information that’s isolated to one team. Anything meant to serve multiple teams or apps needs a shared context layer, or you’ll end up rebuilding the same knowledge base five different ways.
Distinguish between load order and authority for context. Build in traceability via a context graph or an equivalent record of why you made a decision.
Include a clear rule for which data source wins when two disagree. Skip this and you’ll get outputs that look confident but behave inconsistently, which is a much harder problem to debug than it is to prevent.
Where to invest as you scale
Treat agents like products as you scale AI implementation. They need owners, versioning, and a plan for what happens when you replace one, like any other piece of software.
Resist the pull toward generalist tools. A narrowly scoped agent that does one thing well will consistently outperform a broad one tasked with doing everything. It’s similar to how you’d rather call a plumber for a leak than a generalist handyman, even though the handyman might be cheaper and available sooner.
Specificity enables a system to be both explainable and trustworthy. A generalist agent that’s 80% right in five different domains is harder to trust than a specialist that’s 98% right in one.
How to incorporate the lab-factory model
Using AI can feel like a choice between showing progress now and building the right thing slowly. But you can address sequencing issues using the lab-factory model.
- A lab is optimized for learning fast without holding outputs to production standards
- A factory is optimized for reliability, where standards should be strict
Decide what must be true before an initiative graduates from the lab to the factory. For example, that might be a stability threshold on the base, a validated pattern at the builder layer, or an accuracy threshold you maintain for a defined stretch of time.
Without the gate, teams often default to one of two extremes. Everything stays an experiment indefinitely, or everything gets pushed into production before it’s ready. This can decrease trust across your team and negatively impact your key performance indicators (KPIs).
Real-world examples of the lab-factory model in action
To make this model sustainable, run lab and factory work simultaneously on different things, rather than sequentially across the whole org. That’s how you get visible directional wins like a shipped workflow, faster process, or pilot with real numbers. Meanwhile, the less-visible base work keeps advancing, rather than everyone waiting for the foundation to be finished before anything ships.
You can see this pattern play out across several enterprise companies:
- Adobe has spent the past couple of years building a content supply chain with brand guidelines, metadata, and review workflows unified into a system that agents can draw from. This is squarely base-layer work, done before scaling any generative workflow across the enterprise.
- Coca-Cola, working with OpenAI and Bain & Company, built a generative platform, Create Real Magic, which lets creators and internal teams produce on-brand content through a live, working pipeline. This was a real builder-layer construction.
- Duolingo has publicly said that AI enabled it to scale course content from roughly 7,100 units a quarter to over 20,500. It’s tied that acceleration directly to double-digit revenue growth in its own investor communications.
In the From Potential to Profit study, BCG found that companies focusing on an average of 3.5 AI use cases generate 2.1 times more ROI than those spreading their efforts across more use cases. Depth beats breadth, and addressing the base before the builder, before the beneficiary makes this depth possible rather than just aspirational.
Reframing the revenue question
Leaders are right to keep asking the ROI question. But the next time you field a question about what AI is doing for revenue, the answer isn’t to defend the number or apologize for not having one. Instead, say plainly which layer the current work addresses and what has to happen next for it to reach the revenue layer.
