The most subtle mistake in AI adoption isn't technological. It's human. Buy the best platform, integrate the most advanced model, deploy on perfect infrastructure — and then the system doesn't get used. Because the people who should use it don't understand it, don't trust it, or don't know how to integrate it into their daily work.
GRAL sees this pattern with alarming regularity. And the solution isn't what consultants sell — you don't need corporate reorganizations or ten-person internal AI teams. You need targeted preparation of the people you already have.
The Internal Data Scientist Myth
Many companies' first reaction is: "We need to hire a data scientist." For most Italian SMEs, that's the wrong answer.
A senior data scientist costs 60-90 thousand euros per year gross. In major Italian cities, even more. And a single data scientist, isolated in a company without data culture, produces little. Not because they're not good — because they don't have the ecosystem to be effective: accessible data, infrastructure, colleagues who understand what they do.
For the majority of Italian companies, the correct approach is different: build distributed AI competencies, not concentrated ones. Not one expert who knows everything, but an organization that knows enough.
The Three Levels of AI Competency
At GRAL, we distinguish three competency levels, each with a specific role in AI adoption.
Level 1: AI Literacy (Everyone)
Every person in the company should understand what AI can and cannot do. Not at a technical level — at a practical level.
What this means concretely:
- Knowing that AI learns from data, and bad data produces bad results
- Understanding that AI isn't magic — it makes predictions based on patterns, not reasoning
- Being able to distinguish between what AI handles well (repetitive patterns, large volumes) and what requires human judgment (novel cases, ethical decisions, emotional context)
- Having neither irrational fear nor blind trust
Time required: 4-8 hours of training, distributed across 2-3 sessions. Not a master's degree — a practical overview with examples from their own sector.
Who delivers it: can be the AI vendor, an external trainer, or even internal resources after train-the-trainer preparation.
Level 2: Operational Competency (Direct Users)
People who will use the AI system daily need to understand how it works, when to trust the results and when not to, and how to handle cases where the system is wrong.
What this means concretely:
- Being able to use the system interface and interpret results
- Understanding confidence levels — if the system says "80% probability," what does that mean for the decision to be made?
- Knowing when and how to escalate to human supervision
- Being able to provide structured feedback to the system (corrections, error reports)
Time required: 2-4 days of hands-on training, with shadowing during the first 2-4 weeks of use.
Common mistake: training once and then forgetting about it. Operational competency is maintained through periodic refresher sessions, especially when the system is updated or expanded.
Level 3: Management Competency (1-2 People)
You need at least one person — ideally two — who bridges the company and the AI vendor. They don't need to be a data scientist. They need to be someone who understands the business, has basic technical literacy, and knows how to ask the right questions.
What this means concretely:
- Monitoring system performance metrics
- Understanding when performance degrades and why
- Coordinating maintenance and update activities with the vendor
- Translating business needs into understandable technical requirements
- Evaluating proposals for system expansion or modification
Who is this person: often an IT manager, operations manager, or project manager with interest in technology. The key is curiosity and the ability to communicate between different worlds, not pure technical skills.
Time required: 3-5 days of intensive training, plus a mentoring track with the vendor during the first 6 months.
The Training Plan That Works
Based on GRAL's experience with Italian companies of various sizes, here's a realistic training plan.
Month 1-2: Before the AI Project
Goal: create the cultural context for adoption.
- AI literacy workshop for all management (half day)
- Literacy sessions for involved operational teams (2-3 hours each)
- Identify the Level 3 person and begin their training track
- Clear internal communication: why we're adopting AI, what will change, what won't change
The mistake to avoid: skipping this phase because "it takes time." Companies that don't prepare people before the project spend triple the time afterward, managing resistance and misunderstandings.
Month 3-4: During Development
Goal: involve users before go-live.
- End users participate in system testing — not as guinea pigs, but as co-designers
- Collect feedback on interfaces, flows, real use cases
- Level 3 person participates in technical reviews with the vendor
- "What to expect" session to manage expectations: the system won't be perfect on day one
Month 5-6: Go-Live and Support
Goal: smooth transition to operations.
- Intensive hands-on training for direct users (Level 2)
- On-the-job shadowing during the first 2-4 weeks
- Direct channel for reporting problems and feedback
- Weekly check-ins between Level 3 person and vendor
Month 7+: Operations and Continuous Improvement
Goal: growing autonomy.
- Quarterly refresher sessions for users
- Monthly performance reports shared with management
- Additional training when the system is extended or modified
- Annual competency assessment and gap analysis
The Most Common Resistance (and How to Handle It)
"AI will replace me"
The most widespread fear. Address it with honesty, not empty reassurances. In many cases, AI will change the role — less data entry, more oversight and decision-making. This must be communicated clearly, with concrete examples of how daily work will change.
The most effective response GRAL has seen: having people participate in system design. Those who contribute to the solution perceive it as their own, not as a threat.
"I don't trust the results"
Healthy skepticism. The way to overcome it isn't saying "trust it" but giving tools to verify. Transparency on metrics, ability to compare the AI result with their own judgment, a period where AI suggests but humans decide.
Trust is built through experience, not presentations.
"It's too complicated"
Often a signal that the interface is poorly designed, not that the user is incapable. If an operator with thirty years of experience finds the system incomprehensible, the problem is the system.
Enterprise AI must be simple to use for those who use it daily. Complexity belongs under the hood, not in the interface.
"It worked fine before"
Sometimes it's true. Not all processes benefit from AI, and not all benefits are immediately perceivable. The best response is numbers: show pre- and post-adoption performance data. If the improvement is real, the numbers speak.
The Training Budget
Companies typically invest 1-3% of the AI project budget in training. GRAL recommends 5-10%. The difference in adoption and ROI is disproportionate to the additional cost.
A 150,000-euro AI project with 15,000 euros invested in training has drastically higher chances of success than the same project with 3,000 euros of training. The technology is the same. The difference is the people who use it.
Investing in AI without investing in people is like buying industrial machinery and not training anyone to use it. It works on paper. It doesn't work on the factory floor.
GRAL always includes a structured training plan in every project. Not as an option — as an essential component. Because an AI system nobody knows how to use isn't an AI system. It's a cost.