AI leadership: a practical guide for managers

In short: AI leadership is about decisions. Owners and managers do not need to understand algorithms; they need to choose which decisions to base on data, make sure that data is reliable, train their people and stay accountable for the outcome. AI does not replace managers: it shifts their work from gathering information to choosing, checking and leading the team. Without that direction, AI remains a pile of tools with no effect on margin.

Key takeaways

• The main barrier to AI adoption is not technology but missing skills, according to both Eurostat and Istat, Italy's national statistics office.
• A data-driven decision has four parts: a precise question, reliable data, a model that reads it and a person who owns the choice.
• Managers must know where AI performs well and where it is confidently wrong: without that judgement, people end up trusting answers that are wrong but persuasive.
• Accountability stays human: the GDPR restricts decisions taken solely by an algorithm when they significantly affect people.
• Leading change means redesigning roles and tasks, not handing out licences.
• Training the people who use AI has been a duty under the EU AI Act since 2 February 2025, but no certificate is needed: what counts is a role-based programme, recorded internally.

Why leadership is the real bottleneck for AI

Adoption is growing fast, but without direction. According to Eurostat, 19.95% of EU companies with at least 10 employees used at least one AI technology in 2025, up 6.47 percentage points on 2024, and among large companies the figure was 55.03%. Italy shows the same pattern in sharper form: Istat reports adoption doubling in a year, from 8.2% to 16.4%, yet 33.4% of Italian firms using AI did not select any of the business purposes listed in the survey. Istat reads this as early, experimental use that is not tied to a defined business area.

The barriers are organisational too. Among EU firms that had considered AI but did not use it, the most common reason in 2025 was lack of relevant expertise (70.89%), ahead of unclear legal consequences (52.52%) and concerns about data protection and privacy (48.83%). Skills, rules, data and budget are the responsibility of whoever runs the company, not of the software vendor.

Smaller firms feel it most. Eurostat found AI in use at 17% of small enterprises, 30.36% of medium-sized ones and 55.03% of large ones in 2025. In Italy, Istat notes that while the gap between large companies and SMEs is narrowing on most digital indicators, on AI it is widening: from about 20 percentage points in 2023 to 37 in 2025. For a small or mid-sized business, direction from the top matters even more, as the article on the future of AI for small businesses also points out.

For an online store or a B2B company this is practical: if nobody decides where AI must create value, every department buys its own tool and a year later nobody can say what improved.

What changes for leadership with artificial intelligence?

The way a decision is prepared changes, not who makes it. AI cuts the time spent gathering and reading information; the leader still makes the call and owns it.

AI leadership is the ability to steer people, processes and data towards measurable goals, using artificial intelligence for analysis and execution without handing it your judgement. AI can act as a thinking partner, suggesting hypotheses, scenarios and objections; it cannot take responsibility for a choice.

Area	Before	With AI built into processes
Preparing a decision	Monthly reports compiled by hand	Daily data, automatic summaries and forecasts
Manager's role	Gathers and interprets information	Frames the question, checks the answers, decides
Team skills	Industry and customer knowledge	Industry knowledge plus critical use of AI tools
Control	After the fact, on results	Ongoing, on data, models and results
Main risk	Slow decisions	Fast decisions on wrong data

AI speeds everything up, mistakes included: that is why data and model governance is a board-level topic, as covered in the guide to AI governance and explainability in business.

Will AI replace managers?

Not the role of the person accountable for decisions, but yes, part of the work that fills a manager's day. Preparing reports, summarising documents and comparing scenarios are tasks AI can already speed up, provided someone checks the output. Choosing with incomplete information, negotiating with customers and suppliers, motivating a team and answering for the numbers remain human work.

The rules point the same way. Article 22 of the GDPR gives people the right not to be subject to a decision based solely on automated processing, including profiling, that produces legal effects concerning them or similarly significantly affects them. There are three exceptions: the decision is necessary for a contract, authorised by law or based on explicit consent. In the contract and consent cases the controller must at least guarantee the right to obtain human intervention, to express one's point of view and to contest the decision. For deployers of high-risk AI systems, the European Commission also notes that the duty to train staff to ensure human oversight remains in place.

The real risk is not being replaced by an algorithm but simply watching the numbers AI produces without being able to challenge them. In an online store or a B2B company, whoever sets prices, discounts, product range or sales priorities must be able to explain why they followed, or ignored, the model's suggestion.

How do you make data-driven decisions with AI?

Start from the decision, not the tool. A data-driven decision is a choice where the option is selected by comparing verifiable data against a stated goal, rather than on experience or gut feeling alone.

Data now comes from far more sources than before, and decisions need to be both better informed and faster. The key is less the tool than how information flows inside the company: who collects it, where it is stored, who looks at it and how often. With data coming from the online store, CRM, ERP, campaigns and customer service, analytics software can act as a digital adviser that reads performance continuously, flags anomalies and suggests operational fixes, provided someone has decided which numbers matter and who acts on them.

Many SMEs still lack the foundations. In Italy, Istat found that in 2025 48.8% of SMEs used ERP software against 85.9% of large companies, and 21.1% used a CRM against 56.5%. The share of companies running data analysis, with in-house staff or external partners, rose from 26.6% to 42.7% in two years. Without well-organised data, AI has nothing reliable to read.

Two hypothetical scenarios:

• An online furniture store with 300 orders a month. Restocking is decided by feel and 10% of orders stall on out-of-stock items. A demand forecasting model fed with sales history and seasonality proposes reorder quantities every Monday. The buyer approves or adjusts: within three months it is clear whether stockouts are falling.
• A B2B components manufacturer receiving 40 requests for quotation a month. Sales reps answer in order of arrival. Scoring requests by industry, customer size and order history changes the order of work; the sales director sets the threshold and checks whether win rates go up.

In both cases AI does not decide on its own: it prepares the decision and shows which data it relies on. Choosing indicators takes a baseline first, as explained in the guide to KPIs for measuring the value of AI.

Which departments and industries benefit most from AI in decision-making?

The data shows where companies already use it: mainly in marketing and sales and in business administration. Among EU companies using AI in 2025, 34.70% used it for marketing or sales and 31.05% for organising business administration processes or management, while logistics was the least common purpose (6.08%). In retail trade, 48.18% of companies using AI applied it to marketing or sales; in manufacturing the top purposes were marketing and sales (30.41%) and business administration or management (27.05%), according to Eurostat. Italy follows a similar pattern, with marketing and sales (33.1%) and administrative processes (25.7%) in the lead.

Beyond the statistics, the type of decision is what matters: AI helps where a choice is repeated often, relies on data the company already collects and has a measurable effect. That holds from retail to manufacturing. Here is where to start in an online store or a B2B company:

Department	Recurring decision	Data needed	Who stays accountable
Marketing	Budget split across campaigns and channels, segments to target	Cost and conversions per campaign, CRM data	Marketing manager
B2B sales	Priority of requests for quotation, customers at risk of churning	Order history, customer industry and size, response times	Sales director
Online store	Prices, promotions, products to feature	Sales by product, margins, stock, traffic	E-commerce manager
Purchasing and warehouse	Reorder quantities and timing	Sales history, seasonality, supplier lead times	Purchasing manager
Finance and admin	Payment reminders, invoice checks, cash flow forecast	Due dates, payments, customer terms	Finance manager
Customer service	Which requests to handle first, standard replies	Tickets, contact reasons, returns	Customer service manager

For industry examples, see the guides on AI in retail and on AI in the industrial machinery sector.

What skills does management need in the age of AI?

Judgement matters more than technical skill. An experiment with more than 700 Boston Consulting Group consultants, reported by MIT Sloan, shows why: when generative AI worked on tasks within its capabilities, performance improved by nearly 40%; when the task sat just outside them, performance dropped by 19 percentage points on average, because people trusted answers that were wrong but persuasive. The researchers call this boundary the "jagged technological frontier".

For a manager that translates into five concrete skills:

1. AI literacy. Knowing what a model does, what it cannot do and why it can be confidently wrong.
2. Critical reading of data. Asking where a number comes from, how recent it is and what is missing before deciding on it.
3. Breaking down processes. Splitting a process into tasks and working out which sit inside the AI frontier and which do not.
4. Governance and risk. Knowing the basics of privacy and the AI Act; the NIST AI Risk Management Framework, a voluntary framework built around Govern, Map, Measure and Manage, helps structure controls.
5. Change management. Supporting people and redesigning roles.

None requires coding; all require deep knowledge of your own business, which is also the starting point for deciding how generative AI fits into your strategy.

How can managers use AI in their day-to-day work?

The fastest way to build judgement about AI is to use it on your own work, with clear rules. Across the EU, the AI technologies companies use most are text mining (11.75% of all enterprises) and AI generating images, video or audio (9.55%) or written and spoken language and programming code (8.76%), according to Eurostat: exactly the kind of tools that support a manager's work.

• Preparing meetings. Summaries of reports, minutes and long documents, with a list of the decisions to be made.
• Reading the numbers. Asking for an explanation of a sales or margin trend and for the questions worth exploring with the team.
• Comparing scenarios. For example the effect of a discount or a price increase, starting from stated, checkable assumptions.
• Drafting. Team updates, procedures and customer replies, always reviewed before they are sent.
• Preparing negotiations and quotes. Gathering information on a B2B customer and likely objections before a meeting.

Three ground rules: enter customer and employee data only into tools the company has approved, because it is personal data protected by the GDPR; check every number and source before using it; agree when it must be disclosed that a document was prepared with AI. In its FAQ on Article 4 of the AI Act, the European Commission states that even companies whose staff use tools such as ChatGPT to write advertising copy or translate texts must inform them about specific risks, such as hallucinations.

Who should lead AI in a company: the owner, IT or a Chief AI Officer?

Direction belongs to senior management; execution involves several people. On the AI literacy duty under Article 4 of the AI Act, the European Commission makes clear that no specific governance structure is required: there is no need to appoint an "AI officer" modelled on the GDPR data protection officer or to set up a dedicated board. A large company may choose to have a Chief AI Officer; in an SME or an online store, clear roles are enough:

• Sponsor (owner or general management): sets priorities, approves budget and rules, answers for results.
• Process owner: the manager of the department involved, who defines the decision to improve and the KPIs.
• Data lead: knows where the data sits, how reliable it is and who can access it.
• IT or technical partner: integrates the tools and manages security and access.
• Privacy and compliance: checks GDPR, AI Act and supplier contracts, together with the DPO if the company has appointed one.

A short monthly meeting between these people, with the numbers from current projects, can be enough. In a larger organisation, it pays to link these choices to the wider picture of enterprise AI and operational context.

How do you lead change when AI arrives in your company?

You lead it by redesigning work, not by announcing a tool. NBER research by Brynjolfsson, Li and Raymond on 5,179 customer support agents found that a generative AI assistant raised issues resolved per hour by 14% on average and by 34% among novice and less-skilled agents, with minimal effect on the most experienced. AI spreads what top performers do, but someone has to define it and redesign roles.

The researchers quoted by MIT Sloan suggest an onboarding phase where people learn where the tool works, faster learners acting as recognised internal trainers, experimentation across roles and a culture of accountability in which nobody hides behind "the AI said so".

Engaging people matters as much as the tool. If nobody explains what will change in their role, they are unlikely to use it with conviction. Be clear about which tasks change and which do not, involve the people who will do the work every day in the pilot, collect their doubts and share the results, including the bad ones. Working with data is first of all a change of mindset and company culture: the goal is for the team to see AI as support for their own decisions, not as a control tool.

How do you train managers and teams on AI?

With ongoing training, split by role and tied to real company cases. It is also a legal duty: Article 4 of the EU AI Act, applicable since 2 February 2025, requires providers and deployers of AI systems to take measures to support AI literacy among their staff and other persons using those systems on their behalf, such as contractors and service providers. According to the European Commission, after the Digital Omnibus amendments that entered into force in mid-July 2026 no minimum level is mandated, but the obligation remains, and supervision and enforcement by national authorities apply from 2 August 2026. Italy adds its own AI law (Law 132/2025), in force since 10 October 2025.

The Commission's FAQ lists what to consider at a minimum: a general understanding of what AI is, which systems the company uses and their opportunities and dangers; the company's role, that is whether it develops AI systems or uses systems built by others; the risks of the systems in use; and actions built on people's knowledge and experience and on the context of use. Different levels of depth are allowed, no certificate is needed and an internal record of the training is enough. Simply asking staff to read the instructions for use, on the other hand, may be ineffective in many cases.

A practical, role-based programme:

• Senior management: AI limits, legal and reputational risks, reading project KPIs; refreshed whenever strategy changes.
• Department managers: use cases in their own process, checking outputs, data and privacy rules, working on a real company case.
• Day-to-day users: using approved tools, typical errors such as hallucinations, when to ask for a check; short, repeated sessions with a go-to colleague.
• Contractors and suppliers using AI on the company's behalf: usage rules and, where relevant, contract clauses, in proportion to the type of system and the risk.

An internal hub with guides, approved instructions and up-to-date examples helps keep training continuous and tailored, because tools change every few months. For the impact on HR and compliance, see the article on AI in HR and compliance processes.

How to apply it in your company: first steps

1. Pick one recurring decision. Just one, with a financial impact: restocking, lead priority, discounts or returns.
2. Measure the current state. For four weeks collect times, errors and results: without a baseline you cannot tell whether AI improved anything.
3. Get your data in order. Connect ERP, CRM and online store before adding AI: business process automation often brings gains before any AI is added.
4. Run a 60 to 90 day pilot. One named owner, three to five KPIs, short training and a rule on who may override the AI's suggestion.
5. Decide on the numbers. At the end of the pilot, scale, fix or stop. What you learn becomes the template for the next project.

To see how these steps become working systems, take a look at the solutions and at the other articles in the AI for business section.

Is your leadership ready for AI? An 8-question checklist

Answer yes or no. Every "no" is something to fix before rolling projects out to other departments.

1. Can you name the three recurring decisions with the biggest impact on margin or revenue?
2. Do you have a measured baseline of time, errors and results for each of them?
3. Are your online store, CRM and ERP data connected, and is someone accountable for them?
4. Is there a list of the AI tools used in the company, including free ones picked by individuals?
5. Are there written rules on which data may be entered into AI tools?
6. Does every AI project have a named owner and three to five KPIs?
7. Have you started, and recorded, AI training for the people who use it?
8. Is it clear who may override an AI suggestion and how that choice is recorded?

Mistakes to avoid

• Starting from the tool. Buying licences before choosing the decision to improve leads to the purpose-free usage Istat finds in a third of Italian companies using AI.
• Leaving it all to IT. The business question and the success criteria belong to management.
• Trusting the most convincing answer. Well-written output is not necessarily correct: spot checks are needed.
• Letting everyone use whatever tool they like. Without a list of tools and data rules, customer and employee information ends up in services nobody controls.
• Automating decisions about people without human review. Job candidates, employees and individual customers are protected by Article 22 of the GDPR: always build in a human check and a way to contest the decision.
• Training once. Tools change every few months; training should be short, repeated and tied to real company cases.

Frequently asked questions (FAQ)

What does leadership mean in the age of AI?

It means deciding where AI enters your processes, which data it uses and which controls apply, while staying accountable for the outcome. A leader does not need to build models, but must frame the business question, read the numbers AI produces and know when not to trust them.

Will AI replace managers?

It changes their work more than it replaces them. AI speeds up how decisions are prepared, with reports, summaries and scenarios, but the choice and the accountability stay with people. For decisions based solely on automated processing that significantly affect individuals, Article 22 of the GDPR sets limits and, in several cases, a right to human intervention. That is why it pays to learn how to use AI and how to check it.

Which skills do managers need to lead AI projects?

Five practical skills: AI literacy and knowing where AI fails, critical reading of data, breaking processes down into tasks, basic governance and privacy, and leading change in teams. They are built faster on a real company use case than through generic courses.

Where should a company start when bringing AI into decision-making?

Start with a recurring decision that takes time today and produces errors, such as stock replenishment or lead prioritisation. Measure the current state, choose a tool and run a 60 to 90 day pilot with one owner and three to five KPIs. Scale it to other departments only if the pilot delivers.

Which industries and departments benefit most from AI in decision-making?

Any industry with recurring decisions on prices, stock, customers and quotes, from retail to manufacturing. In the EU, AI is mostly used for marketing and sales (34.70% of companies using it) and for business administration or management (31.05%); in retail trade the share for marketing and sales rises to 48.18%. For an online store or a B2B company the first candidates are restocking, pricing and lead prioritisation.

Does the EU AI Act require companies to train staff on AI?

Yes. Article 4 of the AI Act requires providers and deployers of AI systems to take measures to ensure AI literacy among their staff, and it has applied since 2 February 2025. After the Digital Omnibus amendments, in force since mid-July 2026, no specific level is mandated, but the obligation remains. No certificate is needed: an internal record of the training is enough. National market surveillance authorities supervise it from 2 August 2026.

How do you measure a leader's success with AI adoption?

With the same numbers used to measure the project, not with the count of tools switched on. What matters is decision time, errors avoided, margin or conversion against the baseline, and the share of people still using the tool after 90 days. Adaptability counts too: fixing or stopping a project when the numbers do not add up, and getting the team to work in a new way.

Sources

• Istat, Enterprises and ICT, 2025 (press release, 15 December 2025, in Italian)
• Eurostat, Use of artificial intelligence in enterprises
• European Commission, AI Literacy: questions & answers (Article 4 of the AI Act)
• Regulation (EU) 2016/679 (GDPR), Article 22 on automated individual decision-making
• MIT Sloan, How generative AI can boost highly skilled workers' productivity (2023)
• Brynjolfsson, Li, Raymond, Generative AI at Work, NBER Working Paper 31161
• NIST, AI Risk Management Framework
• NIST, AI RMF Core: the Govern, Map, Measure and Manage functions
• Italian Official Gazette, Law no. 132 of 23 September 2025 on artificial intelligence (in Italian)

If you want to work out which decision to start from and how to prepare your team, book a free assessment.