The Three Layers of AI That Actually Matter

AI Strategy • 9 min read • by Girish Koliki
The Three Layers of AI That Actually Matter

Most companies have people using ChatGPT and call it an AI strategy. Real AI maturity means working three distinct layers: people enablement, process automation, and product AI. The companies getting results are doing all three, in order.

Every company has an AI strategy now. Or at least, every company says it does.

In most cases, "AI strategy" means a handful of people use ChatGPT, someone bought a Copilot licence, and there is a slide deck with the word "transformation" on it. That is not a strategy. That is a starting position.

The companies actually capturing value from AI are working three distinct layers, and they are working them in order. People enablement first. Business process automation second. Product AI third. Skip a layer and the whole thing stalls. Get the sequence right and the layers build on each other.

~5% Companies generating real value from AI at scale (BCG, 2025)
78% Companies using AI in at least one core business function
The gap Not adoption. Depth.

§ Layer 1: People Enablement

This is where almost everyone starts and where most get stuck.

People enablement means every person in your organisation can use AI effectively, whether they write code or write invoices. It is not about hiring data scientists. It is about raising the baseline so your existing team knows how to get useful results from AI tools, check the output critically, and fold AI into their daily work.

BCG's 2025 research is direct about this: the companies generating real value from AI are not the ones with the most tools. They are the ones that built "integrated, enterprise-wide enablement systems that embed AI into how people think, work, and lead." The rest are still running pilots.

IKEA is the clearest example. When their AI chatbot started handling 47% of routine customer enquiries, they did not cut 8,500 call centre jobs. They reskilled those people as interior design advisors. The result was $1.4 billion in additional revenue. Walmart is taking a similar approach at larger scale, offering free AI training through Google's certification programme to all 1.6 million US and Canadian employees.

For engineering teams specifically, the tool landscape has expanded fast. AI-native IDEs like Cursor, CLI-based agents like Claude Code, and open-source alternatives give developers more capability than ever. But tools alone do not create enablement. What matters is whether your people know how to direct them, check the output, and fold them into how work actually gets done.

Invest in your people before your tools. That is what gets you past the pilot stage.

What people enablement looks like in practice

  • → Every role has a clear picture of how AI fits into their daily work
  • → AI literacy training is systematic, not optional or self-directed
  • → People check AI outputs critically rather than copy-pasting blindly
  • → Domain experts are treated as AI assets, not AI casualties
  • → Curiosity and experimentation are rewarded, not just tolerated

§ Layer 2: Business Process Automation

Once your people know how to work with AI, the next question is: which business processes should AI run?

This is different from people using AI as a personal productivity tool. Business process automation means AI is embedded into how work flows through the organisation. Order processing. Customer support. Invoice reconciliation. Compliance checks. The repetitive, high-volume work that eats capacity without creating value.

Danfoss automated their order processing entirely and cut customer response times from 42 hours to near real-time. Atmira's AI platform handles roughly 114 million monthly requests, with recovery rates up 30 to 40% and operational costs down 54%. These are not experiments. These are production systems.

On the technical side, open protocols like MCP (Model Context Protocol) are reducing the integration tax that used to make AI automation prohibitively complex. Instead of writing a custom connector for every tool and every AI model, teams can write one MCP server and any compatible agent can use it. That changes the economics of automation significantly, especially for mid-sized teams that cannot afford to maintain dozens of custom integrations.

The prerequisite matters, though. AI automation breaks the moment it hits messy data, undefined ownership, or workflows held together with manual exceptions. If your process depends on someone noticing an email and copying values into a spreadsheet, AI will not make it smoother. It will make the edges sharper. Every unowned handoff becomes a failure point.

This is why the sequence matters. Layer 1 gives your people the literacy to identify which processes are worth automating and the judgement to spot when automation is going wrong. Without that foundation, you are automating confusion.

Signs your processes are ready for AI automation

  • → The workflow has a clear owner, not a committee
  • → Data definitions are consistent and the system of record is explicit
  • → You can measure cycle time, quality, and cost today
  • → Exceptions are documented, not stored in someone's head
  • → There is a feedback loop that improves the process over time

§ Layer 3: Product AI

This is where AI stops being an internal tool and starts touching your customers.

Product AI means embedding intelligence into the things your customers actually use. Recommendations. Personalisation. Predictive features. Conversational interfaces. AI that does not just help your team work faster, but helps your customers get more value from your product.

Amazon, Shopify, and Instacart have all embedded AI agents into their customer operations. These agents resolve order queries, handle delivery changes, and manage scheduling autonomously in most cases. Wagestream uses Google's Gemini models to handle over 80% of customer enquiries without human involvement.

At the infrastructure level, the plumbing is maturing too. MCP connects agents to tools. A2A (Agent-to-Agent Protocol) connects agents to each other. Together, they mean your product AI does not have to be a monolith. You can compose specialised agents that each handle a different part of the customer experience, coordinating through open standards instead of custom wiring.

But product AI is also where the stakes are highest. When AI touches your customer directly, the tolerance for errors drops to near zero. A wrong internal summary is annoying. A wrong customer-facing recommendation is a trust problem. Deloitte's 2026 report found that while close to three-quarters of companies plan to deploy agentic AI within two years, only 21% have mature governance in place. The ambition is running well ahead of the guardrails.

Product AI only works if the two layers beneath it are solid. Your people need to understand AI well enough to design responsible customer experiences. Your internal processes need to be reliable enough that AI-driven features have clean data to work with. Skip those layers and product AI breaks in public.

~75% Companies planning to deploy agentic AI within two years
21% Have mature AI agent governance in place
Gap Ambition is ahead of readiness

§ The Sequence Is the Strategy

The three layers are not a menu. They are a progression.

Companies that skip straight to product AI without people enablement end up with features nobody on the team understands well enough to maintain. Companies that try to automate processes before raising AI literacy end up with automated versions of broken workflows. The 5% generating value at scale, according to BCG, did not skip steps.

Most organisations are somewhere in layer one right now. Their people are experimenting with AI tools, but there is no systematic enablement, no clear picture of what "good" looks like, and no plan to move through the layers. That is not a failure. It is a starting point. But only if you treat it as one.

§ Where to Start

Grade yourself honestly.

Layer 1: Can every person in your organisation use AI effectively in their role? Not just the technical team. Everyone. If the answer is no, that is your first job.

Layer 2: Are your core business processes running on AI, with clean data and clear ownership? If you are still in pilot mode, focus here next.

Layer 3: Is AI embedded in your product in ways your customers can see and value? If not, that is fine. Get layers 1 and 2 right first.

The question worth sitting with is not "are we using AI?" Almost everyone is. The question is "are we building something that compounds, or are we just collecting tools?"

A note from fusecup

At fusecup, we help businesses work through all three layers, from people enablement to process automation to product AI. If you are trying to figure out which layer needs attention first, or how to move from one to the next, we are always happy to talk. No agenda, no pitch. Just a practical conversation about what might work for where you are right now.

§ References

  1. BCG, To Unlock the Full Value of AI, Invest in Your People (2025). bcg.com
  2. Steal These Thoughts, How IKEA Reskilled 8,500 Employees and Made $1.4 Billion (2024). stealthesethoughts.com
  3. AInvest / Fortune, Walmart's AI Workforce Bet: Free Training for 1.6 Million Employees (February 2026). ainvest.com
  4. Google Cloud, AI Agent Trends Report 2026: Danfoss, Atmira, Wagestream, Amazon, Instacart case studies. cloud.google.com
  5. Deloitte AI Institute, State of AI in the Enterprise 2026 (January 2026). deloitte.com
  6. McKinsey / AmplifAI, Generative AI Statistics: enterprise adoption. amplifai.com
  7. Medium / Exit Fund, AI In 2025: What Changed, What's Coming In 2026. medium.com