AI adoption is easy to observe. Business impact is not.
An organisation can buy licences, run pilots, train employees, launch chat assistants, and report widespread usage without changing its cost base, customer experience, cycle time, quality, or revenue. This is not necessarily a sign that AI has failed. It is often a sign that the organisation has adopted a tool without yet reorganising the system around it.
The adoption trap
Counting users is attractive because it is simple:
- How many employees have access?
- How many prompts were sent?
- How many copilots were activated?
- How many pilots are running?
These numbers can matter. They measure exposure and learning. But they are not outcome measures.
A better question is:
What operational bottleneck changed because AI is now part of this workflow?
If the answer is unclear, the organisation is still in adoption mode rather than impact mode.
Technology moves faster than reorganisation
AI can be deployed in a browser in minutes. The complements required to capture value move more slowly:
- Data access and data quality
- Workflow redesign
- Integration with systems of record
- Clear ownership and accountability
- Policy and control design
- Skills, incentives, and performance measures
- Procurement, security, and regulatory approval
This is why visible capability can arrive long before visible productivity. The technology is present, but the surrounding organisation still reflects the workflow of the previous era.
The productivity J-curve
Many general-purpose technologies create an initial productivity dip. Teams spend time learning, reworking processes, migrating data, resolving exceptions, and maintaining both old and new ways of working. The benefits appear only after complementary investments are in place.
AI is likely to follow the same pattern.
The question is not whether a team can use a model. The question is whether the workflow has been redesigned so that the model’s output reaches a decision point, reduces rework, and changes the economics of the process.
From pilot to production
A useful production workflow has five properties:
- A specific decision or action. The workflow does not end with a summary; it changes what someone does next.
- Trusted inputs. The relevant data and documents are accessible, governed, and connected.
- Exception handling. The system knows which cases it should not handle automatically.
- An accountable owner. Someone is responsible for the outcome, not merely the tool.
- A measurable metric. Cycle time, error rate, recovery rate, conversion, risk exposure, or another business outcome improves in a measurable way.
An example
Compare two AI initiatives.
Initiative A: Employees receive a general chat assistant. Usage rises. People say it saves time. There is no direct measure of quality, process change, or financial impact.
Initiative B: AI extracts contract obligations, compares them with procurement data, identifies approaching notice periods, links each recommendation to source clauses, and routes exceptions to contract owners. The organisation measures missed-renewal risk, review time, and renegotiation outcomes.
Both may be useful. Only the second is structured to demonstrate and compound operational impact.
The AIxtract perspective
AIxtract focuses on the gap between data and action. The goal is not more AI activity. It is a workflow in which fragmented information becomes a traceable decision, a decision becomes an accountable action, and the result improves the next decision.
Adoption is the beginning. Reorganisation is where the value is captured.
Adapted from the book Cheap Thinking: What AI Makes Abundant, What It Makes Scarce, and Who Captures the Difference by Daoyuan Li, PhD.