Why most AI integrations fail
Companies buy tools. Lots of tools. According to the Baromètre France Num 2025, the French government's annual survey of digital adoption in small businesses, 26% of French small and midsize businesses now use artificial intelligence. The number is rising fast, but it hides a reality I see in the field: most of that adoption stays superficial.
The scenario is almost always the same. An executive reads an article about ChatGPT. They buy a subscription for their team. A few employees try it, others ignore it. After 3 months, nobody really knows who uses what, for what result, or whether the investment has had any measurable effect.
What's missing every time is a method. Companies treat AI as a standalone tool, when it should come in as a change to roles, skills and metrics. I've documented the variations of this failure in my article on the 3 most common AI mistakes in SMBs.
After working with dozens of marketing teams, I tested several integration frameworks. Most shared the same flaw: they started with the technology and treated people as an afterthought. Then I came across the STEP method.
What the STEP method is
The STEP method was designed by Paul Leonardi, PhD, a professor and researcher at the University of California, Santa Barbara (UC Santa Barbara). It draws on 3 years of academic research with 10 companies. The findings were published in Harvard Business Review France, the magazine's French edition.
STEP is an acronym for 4 complementary pillars:
- Segmentation (segment the tasks): identify what AI must not do, what it can augment and what it can automate.
- Transfer (evolve the roles): reinvest the time freed up in higher-value activities.
- Education (keep learning): set up continuous learning that outlasts a one-off workshop.
- Performance (evaluate differently): adapt performance metrics to the new roles.
Most companies focus on choosing AI tools, when the real challenge is rethinking the way people work with those tools.
Paul Leonardi, Harvard Business Review France
I've tested several AI integration frameworks, and STEP is the only one I kept. It's the only one that puts people at the center and addresses all 4 dimensions of change at once: tasks, roles, skills and measurement. It has become the foundation of all my engagements.
What sets STEP apart from the usual approaches is its sequence. Each pillar prepares the next. Segmentation clarifies roles. Transfer gives the change meaning. Education makes it last. And Performance makes it measurable.
S for Segmentation: segment the tasks
Segmentation is the starting point. Before rolling out a single tool, each team member maps their own activities into 3 distinct categories.
The first category covers the tasks AI must not touch. These are activities that require human judgment, original creativity or a strong interpersonal component. Negotiating with a partner, settling an editorial disagreement, building a long-term client relationship: AI has no place here.
The second category covers the tasks AI can augment. The employee stays in control, but AI speeds up or enriches the work. Writing a first draft, summarizing competitive intelligence, analyzing campaign results: the human leads, AI executes.
The third category identifies the tasks AI can fully automate. Weekly reporting, data entry, scheduling posts, sending standardized follow-ups. These tasks take up time without creating any differentiation.
A real example from an engagement with an 8-person marketing team:
| AI doesn't touch | AI augments | AI automates |
|---|---|---|
| Brand strategy and positioning | Content writing (first drafts) | Weekly KPI reporting |
| Negotiating with partners | Competitive monitoring and summaries | Social media scheduling |
| Customer relations and crisis management | Campaign performance analysis | CRM data entry |
| Art direction and editorial choices | SEO keyword research | Sending standardized follow-ups |
| Team leadership and mentoring | Creating ad variants | Data extraction and formatting |
The segmentation exercise is a team effort. Each employee maps their own tasks, then the team compares results. This is often when disagreements surface, and that is exactly what makes the exercise valuable. Some tasks that management assumed could be automated turn out to be far more nuanced once you listen to the people who do them.
T for Transfer: evolve the roles
Transfer is the most strategic pillar of the method. It answers a question many companies dodge: what do you do with the time AI frees up?
STEP's answer is clear: freed-up time is time to reinvest. Every hour saved on an automated or augmented task must be reassigned to an activity with higher value for the company.
In practice, this means roles evolve. A content writer who spent 70% of their time producing blog posts now spends 40% on AI-assisted writing and reinvests the remaining 30% in editorial strategy, performance analysis or experimenting with new formats. A junior marketing associate who spent their days on reporting can now help shape the acquisition strategy.
Roles are kept and enriched. That is the core promise of transfer.
Transfer is always the part that reassures teams the most. Once employees understand that their job is safe and that AI shifts their day-to-day toward more interesting work, resistance to change drops dramatically. I see this shift in every engagement.
For transfer to work, you have to formalize it. Each team member needs a clear view of their new scope. I recommend documenting role changes in a simple table: activities dropped (handled by AI), activities kept (augmented by AI), new activities (where the freed-up time goes). Writing it down removes the vagueness that feeds anxiety.
E for Education: keep learning
One-off training fails. This is the best-documented finding in Leonardi's research. Companies that run a single initial training day and then consider the topic closed see adoption collapse within a few weeks.
AI is evolving at an unprecedented pace. Models change, tools update, new capabilities appear every month. A static training program becomes obsolete within weeks. That's why STEP calls for continuous education, built around regular rituals.
The formats I roll out with my clients, directly inspired by the Education pillar:
- Weekly AI clinics (30 minutes): a recurring slot where every team member can ask a question, share a discovery or work through a blocker together. This format builds the habit of experimenting and improving as a team.
- Monthly skill challenges: each month, a concrete challenge (for example, automating a specific process or creating a reusable prompt for a recurring task) that pushes the team to explore new uses.
- Internal prompt library: a shared, living document that the whole team adds to, collecting the best prompts proven in practice. It's a powerful accelerator for new hires.
- Peer learning: the most advanced employees train the others. This format works better than external training because it's rooted in the team's real use cases.
According to McKinsey's The State of AI report, organizations that invest in continuous training reach an 80% adoption rate, compared with less than 30% for those that only hand out licenses. I've documented my hands-on lessons on this topic in my case study on training 100+ employees on AI.
Ready to structure AI integration in your team?
The AI Marketing Cockpit: your brand encoded, your tools connected, 54 ready-to-use skills.
Discover the AI Marketing CockpitP for Performance: evaluate differently
When roles change, performance metrics have to change too. This is the fourth pillar of STEP, and probably the one companies neglect most.
Evaluating a writer solely on the number of articles produced makes no sense when AI lets them produce 4 times as many. Measuring a marketer on the volume of leads qualified by hand becomes absurd when AI automates qualification. The old KPIs no longer reflect the reality of the work.
STEP recommends short evaluation cycles focused on new dimensions: the quality of collaboration with AI, adaptability, knowledge sharing with the team, and the quality of the handoff between human and machine.
| Traditional metrics | STEP metrics |
|---|---|
| Number of articles produced per month | Editorial quality and engagement per piece |
| Volume of leads qualified manually | Time from first contact to actionable insight |
| Hours spent on reporting | Experimentation rate (new AI uses tested) |
| Number of campaigns launched | Collaboration score (sharing AI practices) |
| Raw individual productivity | Quality of the human/AI handoff |
Productivity metrics stay. The point is to complement them with metrics that reflect the new reality of augmented work. An employee who shares 3 effective prompts with the team creates more value than one who keeps their techniques to themselves, even if the latter's individual productivity is slightly higher.
I recommend monthly evaluation cycles for the first 6 months of integration, then quarterly ones once the new metrics have stabilized. Evaluation is there to steer and adjust practices. Using it to penalize late adopters defeats its purpose.
Why STEP works better than the alternatives
Most of the AI integration approaches I come across in the field fall into 3 categories.
The first is the “tool-first” approach. The company picks a tool, rolls it out and hopes adoption will follow. It's the most common approach and the least effective. When nobody has thought through tasks, roles and skills, the tool becomes a gadget abandoned within weeks. I cover this in detail in my article on the AI tools to know in 2026.
The second is the top-down mandate. Leadership declares that “everyone must use AI” without providing a framework, training or a clear vision. Teams get the order without the means. Frustration replaces enthusiasm.
The third is laissez-faire. Each employee explores AI at their own pace, with no coordination. Some become very skilled, others never touch the tools. The gap widens, practices aren't shared, and the organization as a whole stands still.
STEP works better because it combines 3 rare qualities. It's people-centered: every pillar starts from the employees, their tasks, their roles and their skills. It's iterative: short cycles allow constant adjustment. And it's evidence-based: 3 years of academic research, 10 companies studied, published and verifiable results.
It's also the only method I've found that addresses all 4 dimensions of change at once. Partial approaches fail because they always leave a blind spot. Segmenting without transferring creates frustration. Training without measuring creates no accountability. STEP covers the whole spectrum. That's why I made it the foundation of my method for structuring first AI projects.
Conclusion
The STEP method is a philosophy of integration: people lead, AI executes. Segment tasks to know where AI creates value. Transfer the time freed up to richer work. Educate continuously to keep pace with the tools. Measure with metrics suited to the new reality of work.
Any company, whatever its size, can apply this method. It works on a modest budget, with no dedicated technical team. It takes a method, a willingness to listen and a commitment to putting employees at the center of the change.
If you want to know where your team stands and where to start, my free AI marketing diagnostic gives you a clear picture in 6 minutes.
