A Framework for AI Adoption — And Why Most Budgets Are Funding Hobby Projects Instead

The biggest barrier to introducing new technology in corporate is trying to figure out what problem it will solve for you and what’s the ROI. The enterprise is still stuck at these questions along with many misnomers about AI. Budgets may get allocated but will largely be wasted on hobby projects of an executive or a team. This is a shame because AI has been ready for the enterprise for a while now.

Here is a proposed AI Adoption Maturity framework that you can use to create your AI roadmap.

Start with two framing questions instead of a vendor RFP – If you got some additional funding to hire, where would you allocate these resources? This will tell you where the real bottleneck is.

Second, don’t approach it as a compute problem. Approach it as a prediction or cognitive problem and ask – what would I like my data to predict for me?

Level 0 adoption (Ready now): Pointed AI solutions solving one specific problem.

Address a specific problem that can solved using AI technologies that are proven and tested – NLP, CV, LLMs, etc. These AI tools would easily fit into the enterprise architecture without major rethink or redesign. For example, data extraction, cleaning, transformation and migration across ERP and home-grown systems from unstructured and semi-structured sources (PDFs, Web, etc.).

Level 1 adoption (Ready now and rapidly evolving): Predictive and Generative AI solutions.

Create a cognitive layer (Cognitive Architecture) that overlays on top of your enterprise architecture by introducing predictions within the workflows. For example, as the items are being received at the receiving dock of an electronic store, a prediction model can start predicting sales of each item based on historical trends, a GenAI model can start drafting a marketing campaign, while another GenAI model can propose best placement positions of each item type based on the store layout and other information it may have.

Level 2 adoption (Ready in 6 – 12 months): AI agents as team members.

In the very near future, as AI Agentic workflows become prevalent you should be able to add them to your teams. Not as mere co-pilots, but active team members. Imagine the scenario that you are asked to spin up a team for a new product. Instead of hiring or allocating one PM, one product manager, two developers and a tester; you should be able to add multiple AI agents to the team that will accelerate the product design, development and release with much higher quality and faster time to market.

This is where it also becomes more complex. To get the most advantage from L2 adoption, you must first have L1 completed.

Level 3 adoption (Ready in 18+ months… or earlier): AI first.

The future may arrive sooner than the time horizon that I predict. Consider the scenario where the AI department, or the IT department, can assemble one or more teams of AI Agents for any project on demand. These AI teams become your primary teams to execute the project, and you then add human team members for critical tasks, oversight, governance and quality assurance.

Think how Cloud resources can be provisioned and deployed with a click of the mouse today, instead of raising a request with the IT team and waiting days, if not weeks. The mistake I see most often is how the CxOs try to skip straight to Level 3 or 4 without going through the readiness step of Level 1. In this blog, I have focused on the enterprise. The bigger disruption will be to the established engagement models of IT services providers, like Accenture, IBM, Infosys, TCS and other consulting organizations. That will be the topic of a later post.