- Is it really worth going all-in on AI builders in your team?
- What is really happening: why the AI roles problem blew up right now?
- What are the four AI roles every marketing team should have?
- What actually happens when maintainers and monitors are missing?
- Three concrete scenarios: how do roles change in an e-commerce, a professional firm, and a manufacturing company?
- How do the possible paths for structuring AI roles compare?
- What rules and risks do you need to know before assigning AI roles?
- Why are we talking about it now: is the Italian AI market really ready for this leap?
- How to actually get started: steps to define AI roles in your team?
- What are the most common mistakes people make when trying to set up AI roles?
- How to tell if the AI role structure is working in your team?
- So, is it worth structuring AI roles in your team right now?
- How we see it at SHM
- Frequently Asked Questions
- Rewarding only those who build AI tools creates a risky ‘monoculture’: you also need maintainers (those who maintain), monitors (those who check), and users (those who use).
- When an automation breaks or goes haywire, the fallout hits the person who was never officially named responsible: the invisible maintainer.
- Starting in February 2025, the EU AI Act requires all companies — even SMEs — to ensure an adequate level of AI literacy for their staff.
- Defining the four roles before launching any automation is cheap and pays off big: it prevents abandoned prototypes and silent crises.
84% of CIOs surveyed by Dataiku state that employees create AI agents faster than IT can govern them. The problem isn't the speed: it's that no one has yet appointed who is responsible for maintaining what is being built.
Is it really worth going all-in on AI builders in your team?
No. A team made up only of people who build AI tools produces many prototypes and few stable results. Roles that maintain, control, and use what is created are also needed.
Many companies today reward those who build AI automations — agents, automatic flows, system integrations — as if it were the only skill that matters. But who keeps those automations running when something goes wrong? Who notices that the brand's tone of voice has changed after a model update? Who uses the tool every day without having to build it?
The honest answer is: often no one. And that's the real problem.
| Your situation | Is it worth structuring the roles? | why |
|---|---|---|
| You already have 2+ AI automations active in production | Yes | Without a maintainer and monitors, the risk of silent drift is high |
| You are evaluating your first AI agent | Yes | Defining roles before launch costs less than doing it after an incident |
| You only use third-party AI tools (e.g. ChatGPT) | Yes | Even usage requires monitoring, and the AI Act mandates literacy for everyone |
| You have a small team (1-3 people) | A middle ground | One person can cover multiple roles, but they must be made explicit |
| You haven't adopted any AI tools yet | Yes | The right time to define roles is before starting, not after |
What is really happening: why the AI roles problem blew up right now?
The pressure to build AI tools has grown rapidly in 2025, driven by company policies and career expectations. The result is an excess of prototypes and a lack of governance.
In 2025, Shopify's CEO stated in an internal memo that using AI has become a basic requirement for all employees, directly impacting performance reviews. Many companies have followed the same direction: the Shopify memo marked a cultural turning point.
The result? People are building. But building isn't enough. According to Dataiku , which interviewed 685 CIOs, 72% admit they cannot verify if the AI agents created by employees are actually producing the results they were designed for.
It's the classic Goodhart's Law effect: when a measure becomes a target, people optimize for that measure — not for the actual outcome.
What are the four AI roles every marketing team should have?
Builder, maintainer, monitor, and user: four figures with different temperaments, all necessary. Ignoring three to reward only the first is the fastest way to accumulate operational debt.
The original source published on MarTech Describes four distinct roles, with precise characteristics and responsibilities. They are not seniority levels: they are different temperaments, equally valuable.
| Role | What it does | Typical characteristics | Risk if missing |
|---|---|---|---|
| Builder | Designs and builds automations, agents, integrations | Creative, comfortable with risk, loves the blank page | No problem getting started, but everything stays in the prototype phase |
| Maintainer | Keeps what has been built running, fixes issues | Methodical, loves troubleshooting, doesn't chase novelty | Automations break down and nobody fixes them |
| Monitor | Checks that the system behaves as expected, detects drift | Detail-oriented, notices when something goes off the rails | Silent drift: the brand sounds weird, nobody notices |
| User | Use AI tools in your daily work without building them | Pragmatic, results-oriented, doesn't need to understand code | Unused or misused tools, wasted investment |
The largest group is that of users. As the HTML analogy reminds us: thirty years ago it was said that everyone would learn to write HTML code. Today, most people use the web without ever having written a line. The same will happen with AI.
This is not a problem: it's normal. The problem is not recognizing it and continuing to evaluate everyone as if they were — or should be — builders.
What actually happens when maintainers and monitors are missing?
Automations break silently, the brand loses consistency, and no one knows who owns the problem. Damage always emerges at the worst possible moment.
Here are three concrete situations that happen every day in marketing teams that have built without appointing maintainers and monitors.
- 1 · None appointed The promotion ends, the agent doesn't An AI agent manages the communications for a promotional campaign. The promotion expires, but the agent continues to send messages with the offer. The builder has moved on to another project.
- 2 · None appointed The model updates, the tone changes The vendor updates the underlying AI model. The tone of voice in automated emails becomes more formal or more informal than usual. The monitor doesn't exist: no one notices for weeks.
- 3 · None appointed A CRM field gets renamed, the automation breaks Someone renames a field in the CRM (the contact management system). Lead routing automation — which assigns contacts to sales reps — stops working. No one is notified.
- 4 · Maintainer appointed The issue is caught within 24 hours With a maintainer assigned at launch, every anomaly has a point person. Resolution time drops from weeks to hours.
AI governance — that is, the set of rules and responsibilities that govern how AI tools are used and controlled — doesn't arrive on its own. As a platform manager interviewed by the source said: 'Governance arrived uninvited.' It's better to invite it beforehand.
Three concrete scenarios: how do roles change in an e-commerce, a professional firm, and a manufacturing company?
The structure of roles changes based on the size and complexity of active automations. Even a small team can cover all four roles, as long as they make them explicit.
The numbers and situations that follow are illustrative scenarios: they're meant to show how to apply the model, not real customer data.
| Company type | Typical automations active | Who acts as builder | Who acts as maintainer | Who acts as monitor | Who acts as user |
|---|---|---|---|---|---|
| E-commerce (8 people) | Post-purchase emails, FAQ chatbots, seasonal promotions agent | Marketing manager or external agency | E-commerce manager (already manages the site) | Customer care (notices anomalies in responses) | The whole team in daily activities |
| Professional firm (4 people) | Automatic email reply, document draft generation | A curious collaborator, with external support | The same employee (overlapping roles) | The owner (reads the final output) | All professionals in the firm |
| Manufacturing company (50 people) | CRM lead routing, automated reports, quote agent | Internal IT or specialized agency | IT or operations manager | Sales manager (checks the output data) | Sales, marketing, purchasing department |
The data from AI Observatory of the Politecnico di Milano it's clear: Italian SMEs are still lagging behind in adoption. But those who start now have an advantage: they can structure roles from the beginning, without having to fix broken governance.
How do the possible paths for structuring AI roles compare?
There are three main approaches: let roles emerge on their own, formally assign them internally, or entrust some functions to an external partner. Each has different costs and risks.
| Approach | Initial cost | Main risk | Suitable for |
|---|---|---|---|
| Implicit roles (no formal assignment) | Zero | High: nobody knows who is responsible when something breaks | No one — this is the situation to avoid |
| Explicit internal assignment | Low (mapping time) | Medium: requires training and a shift in evaluation culture | Team with at least 3-4 people and automations already active |
| External builder + internal maintainer/monitor | Medium | Low: the builder brings expertise, the internal team maintains the knowledge | SMEs that outsource development but want control |
| Full outsourcing (builder + maintainer + monitor) | High | Medium: vendor lock-in, risk of losing internal know-how | Companies without dedicated internal resources, with an adequate budget |
What rules and risks do you need to know before assigning AI roles?
The European AI Act already imposes AI literacy obligations on all employees today, regardless of company size. It's not a future issue: it's been in effect since February 2025.
Article 4 of the AI Act — the European regulation on artificial intelligence — obliges every company using AI systems to guarantee a sufficient level of AI literacy (AI literacy) to its staff. The obligation came into effect on February 2, 2025, and also applies to those who only use tools like ChatGPT. Details are available at Il Sole 24 Ore .
This means that defining AI roles isn't just an organizational choice: it's also a response to a regulatory obligation. A monitor who doesn't know what they're monitoring doesn't meet the literacy requirement. A user who uses a tool without understanding its limitations doesn't either.
| Risk | Whoever generates it | Whoever intercepts it | How to mitigate |
|---|---|---|---|
| Brand voice drift (out-of-control tone of voice) | Automatic update of the AI model | Monitor | Periodic output review, alerts on variations |
| Orphaned automation (no owner) | Builder changing team or project | Maintainer | Mandatory builder+maintainer pairing at launch |
| AI Act violation (lack of training) | Lack of a training program | HR / Management | Structured, documented training for all roles |
| Agent that keeps running after the expiration date | Lack of an end-of-campaign logic | Monitor + Maintainer | Shutdown checklist planned at launch |
| Silent break due to CRM update | System change without notification | Maintainer | System dependency registry, automated notifications |
Why are we talking about it now: is the Italian AI market really ready for this leap?
The AI market in Italy is growing fast, but training and governance remain the weak point. Those who structure roles now are better positioned than those who chase technology without organization.
Last year, the AI market in Italy reached 1.8 billion euros, with 50% growth compared to 2024, according to the AI Observatory of the Politecnico di Milano . Growth is real. But 76% of Italian SMEs have not invested or plan to invest in AI, and only 7% have started structured training programs, according to the research by the Digital Innovation in SMEs Observatory .
The gap isn't between those who use AI and those who don't: it's between those who use it with structure and those who use it chaotically. Structuring roles is the first step to moving from the second group to the first.
How to actually get started: steps to define AI roles in your team?
All it takes is four questions and a responsibility map to turn a team of builders into a functioning AI team. You can do it in one meeting.
Before launching the next automation — or before auditing the ones already active — follow these steps. They don't require special tools: they require clarity.
- 1 · Management / HR Map what you already have in production List all active AI automations: agents, email flows, CRM integrations, chatbots. For each one, write who built it and who manages it today.
- 2 · Management Identify your natural maintainers and monitors Who do you call when something breaks? Who is the first to notice when an output is strange? These people already exist on your team. Give them a name and formal recognition.
- 3 · Management / HR Review the evaluation system If performance reviews only reward those who launch new automations, you are paying for prototypes. Add metrics for those who maintain and those who monitor.
- 4 · Builder + Maintainer Launch every new automation with two names Starting today, every automation that goes live has a designated builder and maintainer. It costs almost nothing at launch. It prevents costly crises later.
- 5 · HR / Management Start a documented training program The AI Act requires AI literacy for everyone. A formal program — even a short one — fulfills the obligation and improves the quality of tool usage.
What are the most common mistakes people make when trying to set up AI roles?
The first mistake is thinking that roles emerge on their own. The second is training only the builders and leaving others without tools. The third is not updating performance evaluations.
Anyone working in the area AI and process automation recognizes these errors frequently. They are not errors of incompetence: they are system errors, which arise when the organizational culture is not aligned with the technical structure.
| Error | Why it happens | How to avoid it |
|---|---|---|
| Train only the builders | External pressure (e.g., company policies) pushes to build, not to maintain | Include maintainers and monitors in training programs from the very beginning |
| Not appointing a maintainer at launch | It's assumed that 'someone will take care of it' | Mandatory pairing: every automation needs two names before going live |
| Evaluate everyone as a builder | Performance metrics reward novelty, not stability | Add maintenance and control KPIs to reviews |
| Ignore the users | Users seem 'passive', thus less valued | Explicitly recognizing that using a tool well is a skill |
| Waiting for governance to show up on its own | People think it is a future problem | Define the rules before the first incident, not after |
A similar theme also emerges when talking about AI quality in marketing : the risk of ‘good enough’ — accepting mediocre outputs because no one is tasked with checking them — is directly linked to the absence of monitors on the team.
How to tell if the AI role structure is working in your team?
Three clear signals: problems are caught before they become visible to customers, automations last over time without degradation, and the team knows who to turn to when something goes wrong.
| Indicator | Positive signal | Red flag |
|---|---|---|
| Average time to fix an AI glitch | Less than 24 hours | More than a week, or nobody notices |
| Number of automations with a named owner | 100% of automations active | Less than 50% |
| Frequency of AI output reviews | At least monthly for every critical automation | Only when something breaks |
| Presence of documented AI training | Formal program for all roles | Only for builders, or none |
| Performance metrics for maintainer and monitor | Included in official reviews | Absent or informal |
| Automations properly decommissioned | Planned shutdown procedure | Orphaned automations still active |
If you are introducing autonomous AI agents into your workflow, it's also worth reading how to structure budget and team on the model of autonomous AI agents : governance roles overlap directly.
So, is it worth structuring AI roles in your team right now?
It is worth it if you already have at least one AI automation active in production, or you are about to launch one. Defining the builder, maintainer, monitor, and user before launch costs little and prevents costly crises.
A middle ground your team is small (2-3 people) and roles overlap. You can start with a simple map: who builds, who checks, who uses. Even an informal one is better than nothing.
Not yet if you haven't adopted any AI tools yet and don't have a plan in the next six months. But keep this structure ready: when you start, you'll know where to begin.
How we see it at SHM
For us, AI is the means — not the end. We believe that integrating artificial intelligence tools only makes sense when it frees people from routine and gives them back time for the work that truly matters. This also applies to how a team is structured: not everyone has to build, and not everyone has to become a tech expert. They need to know what their role is and have the tools to do it well.
In the projects we follow, we often see teams that have built brilliant automations without assigning anyone responsible for maintenance. The problem isn't the technology: it's the organization around the technology. Consider a manufacturing company that integrated an AI agent for routing sales leads: without a designated maintainer, the first change to the CRM would render the automation useless within days. Defining roles before launch was the difference between a working investment and one that silently deteriorates.
If you are figuring out how to structure your AI team or how to start a path of AI training for your company , we can help you map roles and build governance that lasts. The same applies if you are looking for support for the Strategic AI integration in your marketing processes, or to understand how the data processing and business intelligence can support the work of your monitors.
You can also dive deeper into AI agents by reading what changes with OpenAI AI agents for SMB marketing : it is a good starting point to understand what you are building — and who will have to keep it running.
Frequently Asked Questions
No. In a small team, the same person can cover multiple roles. The important thing is that the roles are explicit: every automation must have a name for the person who maintains it and one for the person who checks it, even if they are the same person.
The main risk is silent drift: automation keeps running, but it produces wrong outputs — off-brand tone of voice, expired offers still active, incorrectly assigned leads. The damage always emerges at the worst moment, often in front of a customer.
Yes. Article 4 of the AI Act, in force since February 2, 2025, obliges every company that uses AI systems to ensure a sufficient level of AI literacy for its staff. This also applies to those who only use tools like ChatGPT, and there are no exemptions for SMEs.
It's the person you call when something breaks. They are usually methodical, love solving concrete problems, and don't need to build something new to feel useful. They are often already on the team but aren't formally recognized.
Yes. Even using pre-packaged tools requires a monitor — someone to check the outputs — and a user aware of the tool's limitations. The AI Act applies regardless of whether you build or use ready-made tools.
Start with costs: a broken automation that no one manages has a direct cost in lost leads, poorly managed customers, or incorrect communications. Showing the economic value of maintenance is more effective than any cultural argument. Then add specific metrics: 'active automations without anomalies in the last 30 days' is a measurable metric, just like 'new automations launched'.
The problem isn't how many AI automations you've built: it's how many you can keep running.
Related articles
Discover more articles exploring similar topics, selected to offer you a more complete and stimulating perspective. Each piece of content is carefully chosen to enrich your experience.
