AI: From Productivity to Business Value
Why do individual AI gains not always become business value? A practical measurement framework for leaders, informed by McKinsey’s 2026 research.
A team prepares presentations faster, automates meeting notes and spends less time on research. Its manager still faces a basic question: what changed for the customer, the cost base or revenue? The connection between AI productivity and business value needs to be designed.
In McKinsey’s May–June 2026 survey of 1,719 respondents across 97 countries, 80% reported improved individual productivity. The share reporting enterprise-level EBIT impact was 37%. These figures answer different questions; their difference is neither a conversion rate nor a calculation of lost value. They do, however, make two levels of measurement visible. Research, pp. 3, 12 and 30.
Where does the saved time go?
Our starting point at Stratify is the business outcome that should change. When a sales team prepares proposals faster, it might reach more customers, improve each proposal or simply spend longer waiting for approval. The same time saving can lead to three very different outcomes.
Build the measurement chain explicitly:
Level | Question | Example measure |
|---|---|---|
Task | Is the work faster or better? | Proposal preparation time |
Workflow | Is the entire process improving? | Time from customer need to approval |
Business outcome | What changes for the organization or customer? | Opportunities won, rework cost |
This is our suggested working framework, rather than a finding from the survey. Assigning a data source and an owner to each measure is as important as choosing the measure itself.
Write a small value agreement
For the next pilot, prepare a one-page note covering the baseline, expected change, full cost, quality threshold and review date. Include human review, learning time and process changes alongside licensing and usage fees. An inexpensive experiment can otherwise create an expensive operational burden.
Consider a system that classifies customer requests. Measuring speed alone misses requests routed to the wrong team and the time required to resolve them again. If a faster first step delays the eventual resolution, the expected value has not materialized. The experiment needs a clear decision rule: what evidence would justify expansion, revision or stopping?
Bring a better question to the management meeting
“How many people use it?” tracks adoption. “Which decision are we making better, and how do we know?” opens a discussion about value. Both matter; one should not stand in for the other.
Our Competing in the Age of AI program examines this connection through strategic priorities and investment choices. Start by selecting one recurring team decision and writing down the business outcome that saved time is supposed to improve.
