The AI upgrade path.

Talk to enough companies about AI and a pattern emerges. Nobody "adopts AI" in one move. Perhaps there is no such thing as AI adoption in general. Companies adopt AI tools, one by one, for different use cases. We climb a path, and each stage of it is a genuinely different game, with different work, different risks, and different people who can help.

Quite often, AI programs get confused about which stage they are actually in. Companies buy stage-2 help for a stage-3 problem, or chase stage-4 ambitions before stage 2 has landed. Let's try to summarize it as a map.

Four stages

1 ◆ Learn - a small pioneer group starts. Someone curious, or someone worried, pulls a few colleagues into figuring out what this technology actually is. Awareness sessions, first experiments, maybe a policy draft so people stop pasting customer data into random chatbots. The output is not value; it's vocabulary and appetite.

2 ◆ Equip - the universal tools arrive. Chat assistants (probably Gemini, ChatGPT or Claude) are given to everyone. Coding copilots are procured if you build software. Meeting-summary and document-drafting tools follow. There are strong ready-made tools available for all this by now. You add some training to make people actually use them for everyday work, and the governance to do it safely. This stage is largely a procurement, IT and change-management exercise. The tools are off the shelf, and the playbook is increasingly standard.

3 ◆ Transform - AI reaches the core business processes. This is where a transportation company gets an AI-empowered TMS, a manufacturer gets AI-driven planning, an insurer gets AI in claims or customer support. The tools are much more specialized - often derived from off-the-shelf products, sometimes custom-built. The questions get much harder: which processes are actually worth it, what does the data allow, which vendor claims survive contact with reality, and what has to change in the operation itself for the numbers to move.

4 ◆ Productize - AI, or AI-enablers, become something you really sell to create value for customers and, hopefully, extra revenue. A supply chain platform company starts selling its data as a product. A supplier exposes an MCP or agent interface so that customers' AI systems can talk to it directly - quote, book, track, settle, without a human in the loop. AI features become part of the offering, priced and positioned. This is not an IT project anymore; it's next generation product strategy, with everything that implies: discovery, validation, pricing, go-to-market.

A couple of caveats. First, the path is typical, not mandatory as companies do skip stages, run them in parallel, and circle back. A software company might hit stage 4 opportunities while its HR department is still in stage 1. Second, the stages don't end; stage 2 keeps evolving underneath you even while you work on stage 3.

The stall between 2 and 3

Stages 1 and 2 have become table stakes. They do matter, and you can't skip them, but they don't differentiate anyone anymore, because your competitors are buying the same copilots from the same vendors. The productivity gains are real but broadly evenly distributed and it is a basic hygiene factor nowadays. You don't necessarily see new revenue or cost decrease easily just from them, but you may see future downside if you don't do it.

The step from 2 to 3 is where many companies today stall, so it's worth being precise about why. Stage 2 is easy to buy: the tools are generic, the risk is low, the vendor does the heavy lifting. You may even get a lot of it bundled, as part of your company workspace platform package. Stage 3 is a different game entirely, as it demands knowing your own processes and data deeply, redesigning how work flows, betting real money on specific use cases, and telling several convincing vendors "no." The failure statistics that haunt enterprise AI, like pilots with no measurable P&L impact, are overwhelmingly stage-3 failures: companies bring stage-2 habits (buy tool, roll out, hope) and then struggle when they try to tackle stage-3 problems.

Then stage 4 raises the bar again: now you need new product (or major upgrade) judgment. Is there a real customer, a real willingness to pay, a defensible position? Most companies have never productized anything outside their core offering; doing a major upgrade with AI doesn't make it easier.

Every stage-3 and stage-4 move is a triple diamond

The ladder alone tells you where you are. It doesn't tell you how to execute a move. For that, each stage-3 or stage-4 initiative (an AI-empowered core process, a data product, an MCP interface) has to travel iteratively the full three diamonds of AI value: find the right problem, validate the right solution, and prove the value in the P&L. The ladder is the map; the diamonds are the method. Companies that skip the diamonds at stage 3 are the ones producing the pilot graveyards.

Where are you on the path?

• A few enthusiasts experimenting, no shared tooling: stage 1. Get everyone learning; this is the cheapest stage to do well.
• Tools bought, training done, everyday productivity up; but core operations unchanged: the 2→3 stall. This is the decision point that matters most.
• AI pilots running in core processes, results unclear: mid-stage 3, still quite likely stuck in a valley between solution and value.
• Customers asking how they can plug their AI into you: a clear stage-4 pull signal. Take it seriously; someone in your market will answer it first.

A note from the author

Different stages need different help, and no single advisor honestly covers them all; me included.

For stage 1, you very probably don't need a strategy consultant; you need good trainers, courses and communities, and they are plentiful and affordable. For stage 2, the tool vendors and your IT partner will carry most of the weight, and the playbooks are public. If you're there, I'll happily point you to people who do that well. LinkedIn is quite full of freelance AI consultants also.

Where I (as in GoPlex) work is stages 3 and 4. Stage 3 is where independent, vendor-neutral judgment earns its keep: which core processes are worth AI, which are not, and an unconflicted go / no-go before serious money moves.

Know your stage. Buy help for the stage you're actually in. And when you make the move to 3 or 4, the diamonds are your best friends.