88% of companies already use AI in at least one function. But only 6% get a significant impact on their operating income (EBIT) from it. In between lies a chasm: scaling up.

It isn’t a tooling problem. The models are accessible, costs are falling, teams have tried. What is blocking is the method. The “how”, not the “with what”.

McKinsey has identified what sets the best-performing organizations (high performers) apart. The six AI best practices below all follow the same principle: the AI that creates value is the AI anchored in real work. Not AI bolted onto an existing workflow, a topic we already covered in our article on the barriers to AI adoption.

1. Redesign business workflows before deploying AI

This is practice number 1 by impact. Leaders are 3 times more likely than others to have fundamentally redesigned their processes before deploying AI.

Deploying a generic copilot on an existing workflow produces marginal gains. Redesigning the workflow around what AI does well (automating the mechanical steps, putting people back where the decisions are made) changes the order of magnitude of the impact.

In practice: before choosing a tool, map the process, identify the low-value tasks, and redefine who does what between people and AI. This scoping step avoids the worst mistake of all: automating a bad process.

2. Aim for growth, not just efficiency

Eight organizations out of ten set efficiency as the goal of their AI. High performers always add growth and innovation.

That nuance changes everything. AI driven by cost cutting quickly hits a ceiling: you automate a bit more, cut a few hours, and stop. AI driven by growth opens new fronts: services, markets, new uses.

The right question isn’t “how much will I save?”, but “what does AI let me do that I couldn’t do before?”.

It is also a matter of measurement: counting only the hours saved locks you into the efficiency mindset. The 4 metrics to measure the impact of AI go beyond time saved to capture the real value created.

3. Leaders who set the example by using AI tools every day

The leaders of successful companies get involved in AI far more often than their peers. Not as committee sponsors, as users.

A CEO who uses Claude or ChatGPT in their daily work sends a signal a thousand times more powerful than a transformation plan in PowerPoint. Leading by example remains the most underestimated management practice when it comes to AI.

Why it works: a leader who prompts every day understands the real use cases, sees the technical blockers, accepts the model’s imperfections, and makes decisions as a practitioner, not as a spectator. Teams feel it immediately, and adoption follows.

4. Frame human review by level of risk

Inaccuracy is the AI risk companies run into most often. That is no reason to give up. It is a reason to set a framework.

High performers explicitly define, for each use case, where and how a person checks the result. Neither systematic review (which kills the time saved) nor blind automation. A framework based on what is at stake:

  • Low stakes (internal summary, email draft): a person rereads the output.
  • Medium stakes (customer analysis, internal recommendation): a peer reviews it before it goes out.
  • High stakes (customer decision, external communication, legal document): double review and a record of who approved what.

This framework isn’t decreed by the leadership team: it is built with the teams who use the tool. That is the condition for trust to take hold without slowing down use.

5. Adopt an agile organization

AI moves fast. Models change every month. A classic project cycle (6 months of scoping, 6 months of implementation) doesn’t hold up. You have to iterate, measure and adjust.

In practice, there are three shifts to make:

  • Embed AI in business processes (not as a separate tool managed by IT).
  • Track specific KPIs for each AI skill (usage volume, satisfaction rate, time saved, quality).
  • Feed frontline feedback back into the roadmap every 4 to 6 weeks, not once a year.

6. Invest a real budget

More than a third of high performers spend more than 20% of their digital budget on AI. A minority, but they are the ones getting the EBIT impact.

By contrast, most other organizations stay well below this threshold, scatter their investments and confuse experimenting with scaling up. The result: no use case reaches the critical mass needed to generate a measurable ROI.

The unwritten rule: fewer use cases, better funded, rather than a multitude of underfunded experiments.

The condition behind them all: trust the front line

These six levers have one thing in common. None of them works if the use cases are chosen far from the real work.

The classic mistake: choosing use cases in the leadership team, then imposing the rollout from the top. The result: AI stays far from daily work, teams don’t take it up, and the ROI never comes.

The teams who do the work know better than anyone where the pain points worth automating are. Involving them from the start multiplies the chances of choosing the right use cases, and of getting a measurable ROI from them.

This conviction is at the heart of Teamstarter’s approach to rolling out AI: find the use cases on the front line, build working AI skills with the teams, and measure the ROI continuously. To go further, read our complete guide to AI at work.

Conclusion

The six best practices identified by McKinsey have one thing in common. They talk neither about tools, nor models, nor technology. They talk about method, framework, and the trust placed in employees.

The AI that transforms a company is the AI that starts from the front line. Not the AI handed down from a strategic plan. Not the AI bolted onto what already exists.

AI moves forward where teams take the lead.

FAQ: AI best practices at work

What are McKinsey’s AI best practices for companies?

McKinsey identifies six levers among high performers: redesigning workflows before deploying AI, aiming for growth and not just efficiency, leaders who set the example by using AI tools every day, framing human review, adopting an agile organization, and investing a real, dedicated budget. The condition behind them all: trusting the front line to choose the use cases.

Why do most companies struggle to scale AI?

88% of companies use AI in at least one function, but only a third have started to scale it, and only 6% qualify as high performers (an EBIT impact of 5% or more, and significant value). The blocker isn’t technological: it comes from workflows that weren’t redesigned, a scattered budget, leaders who don’t use the tools, and use cases chosen far from the front line.

How do you redesign workflows with AI?

Before choosing a tool, map the existing process, identify the mechanical, low-value tasks, and redefine who does what between people and AI. Leaders are 3 times more likely to have fundamentally redesigned their processes before deploying AI. Bolting a copilot onto an existing workflow only produces marginal gains.

How much budget should a company devote to AI?

More than a third of high performers spend more than 20% of their digital budget on AI, while most companies stay well below that. The unwritten rule: fewer use cases, better funded, rather than a multitude of underfunded experiments that never reach the critical mass needed to generate ROI.

Why is human review essential for AI at work?

Inaccuracy is the AI risk companies run into most often. High performers define a framework based on what is at stake: a simple reread for low-stakes internal uses, a peer review for recommendations, and a double review for external communications or customer decisions. Neither systematic review nor blind automation.

How do you involve the front line in rolling out AI?

The teams who do the work know better than anyone where the pain points worth automating are. Involving them from the choice of use cases onward, building AI skills with them and measuring the ROI continuously multiplies the chances of adoption. It is the condition for McKinsey’s six best practices to actually produce impact.

→ Want to see how to deploy AI where it really makes an impact in your company? In 30 minutes, we show you concretely. Talk to an expert.

Source: McKinsey & Company, The state of AI in 2025: Agents, innovation, and transformation*, QuantumBlack, November 2025. Survey of 1,993 respondents in 105 countries.*