- Is it really worth upgrading your company from traditional automation to AI automation?
- What is really happening in the market: the end of fixed-rule automation
- What changes in practice: who does what in the new AI + human model?
- Three concrete scenarios: how much does an Italian company that migrates earn?
- Which platforms and paths can you choose for AI automation?
- What risks and rules do you need to keep under control before migrating?
- Why everyone is talking about it now: is the Italian market really ready?
- How to actually get started: the steps to migrate from traditional automation to AI automation?
- What are the most common mistakes made by those trying to take this leap?
- How do you know if the new system is actually working?
- So, is it worth it for your company?
- How we see it at SHM
- Frequently Asked Questions
- Rule-based automation is no longer enough: new AI systems read the context and act independently, without waiting for you to write every single rule.
- In Italy, only 30% of large enterprises have adopted smart automation; SMEs are even further behind, with a real risk of losing ground to competitors.
- Migrating is worth it if you have medium-high contact volumes and repetitive workflows: the real win is time given back to the team, not just faster campaigns.
- The winning model is not 'AI instead of people', but AI that proposes and humans who approve — the so-called human-in-the-loop.
62% of large Italian companies already use automation tools, but only 30% have integrated AI features. The rest are still running on 2010s autopilot — and the gap widens every month.
Is it really worth upgrading your company from traditional automation to AI automation?
It depends on how many contacts you manage and how much time your team wastes writing and updating rules. If the flows are many and change often, the migration pays off in a few months.
Traditional automation works like this: you write a rule ( «if the contact opens the email, send the follow-up after 3 days» ), the platform runs it. It used to work well when there were few channels and behaviors were predictable. Today, touchpoints — the points of contact with the customer — are multiplying and the rules run into the hundreds. Maintaining them is a full-time job.
New AI systems don't wait for the rule: they read customer data, figure out the right moment, and decide the action. The human team oversees, corrects, and approves. This is the model Human-in-the-loop — literally «human in the loop» — mentioned by MarTech in their breakdown of the future of automation.
| Your situation | Is it worth migrating? | why |
|---|---|---|
| You have more than 5,000 active contacts and complex flows | Yes | AI quickly pays off the migration costs |
| You run an e-commerce with abandoned carts and upsells | Yes | Contextual personalization boosts conversion rates |
| You have less than 1,000 contacts and simple flows | No, for now | Classic automation is still enough and cheaper |
| You are using a legacy platform for more than 4 years | Yes | You risk losing integrations with new channels |
| Your team doesn't have time to manage rules | Yes | AI measurably cuts down on operational busywork |
What is really happening in the market: the end of fixed-rule automation
Static rule-based systems are giving way to platforms that use AI language models to read customer context and act in real-time, without waiting for manual instructions.
According to MarTech , the main problem with classic automation is the «context rot» — literally "context rot". Data gets old, rules don't get updated, and the result is that the customer gets the wrong messages at the wrong time. A welcome email sent three weeks after signing up, for example.
New AI systems tackle this problem differently: instead of running rules, they constantly analyze signals — opens, clicks, purchases, site visits — and adapt their behavior. However, just adding a "thin layer of AI" on top of an old platform isn't enough: you need a redesign of the data flow.
The topic of data quality is central. Large language models (LLMs, meaning AI engines like the ones behind ChatGPT) cannot be pointed at billions of raw data points without prior cleaning and structuring. Garbage in, garbage out.
The data is clear: 85% of European companies already use AI tools in marketing, according to IAB Europe . Italian companies are lagging behind, with only 30% having adopted intelligent automation, according to the Digital Innovation Observers from Politecnico di Milano . This lag is a risk, but also an opportunity: those who move now start from an already accumulated database and can fast-track their progress.
What changes in practice: who does what in the new AI + human model?
In the human-in-the-loop model, AI proposes and optimizes, while the human team validates strategic decisions and handles exceptions. Roles don't disappear: they change.
- 1 · AI platform Collects and cleans data Aggregates signals from CRM, website, email, and social channels. Identifies outdated data and automatically excludes it.
- 2 · AI Model Generate the recommended action It suggests the message, channel, and best time for each touchpoint, based on recent behavior.
- 3 · Marketing team Approve or correct Validates high-impact proposals, sets ethical limits and brand guidelines. No longer writes every single rule.
- 4 · AI platform Execute and monitor It launches the campaign, measures results in real time, and updates the model for the next cycle.
- 5 · Marketing team Analyze strategic results It focuses on high-level decisions: segments, offers, positioning. The day-to-day operations are delegated.
This framework solves one of the most cited problems by marketers: AI generating content nonstop, creating more mess than value. With human-in-the-loop supervision, every output is filtered before reaching the customer. As we explored in the article on MoEngage AI agents and the new face of marketing automation , true personalization always requires human oversight regarding data and message quality.
Three concrete scenarios: how much does an Italian company that migrates earn?
The benefits vary greatly depending on the industry and contact volume, but the main gain is always the same: hours of work given back to the team and more relevant messages for the customer.
The numbers below are illustrative examples : they are meant to show the scale of the benefit, they are not real customer data.
| Company type | Current situation | With AI automation | Estimated benefit |
|---|---|---|---|
| Fashion e-commerce (50k contacts) | 20 manual flows, 18% open rate | Adaptive flows, 26-28% open rate | ~+8 pp open rates, -15h/week of management |
| B2B professional studio (2k contacts) | Monthly newsletter + 3 fixed sequences | Contextual sequences for funnel stage | Fewer unsubscribes, more qualified bookings |
| Store network (8k contacts) | Promotions equal for all customers | Personalized offers based on shopping behavior | Fewer wasted coupons, higher average order value |
Which platforms and paths can you choose for AI automation?
The market offers three main paths: upgrade the platform you already use, switch to a native AI solution, or build a custom AI layer on top of your data.
| Approach | Suitable for | Pros | Cons |
|---|---|---|---|
| Update existing platform | Who has already invested in HubSpot, Salesforce, ActiveCampaign | Less discontinuity, data already present | AI features often superficial or extra cost |
| Migration to native AI platform | Those starting from scratch or with super messy data | Architecture built for AI right from the start | Migration and team training costs |
| Custom AI layer | Companies with tricky workflows and proprietary data | Maximum flexibility and control | Requires in-house tech skills or a partner |
| Agentic AI | Early adopters with budget and technical team | Maximum autonomy, continuous learning | Still in the experimental phase for most SMEs |
The Agentic Automation — systems based on language models capable of using external tools and learning autonomously — is still in its early stages in Italy: only 8% of large companies have started experiments, while 25% plan to do so in the coming months, according to the Politecnico di Milano Observatories . To learn more about how these autonomous agents work, check out the article on Autonomous AI agents and the Monks model offers a practical overview.
What risks and rules do you need to keep under control before migrating?
The main risks are three: data quality, privacy compliance, and vendor lock-in. All three can be managed with a plan before turning on any system.
The first risk is dirty data . An AI system fed with duplicate contacts, outdated emails, or expired consents produces worse results than a traditional automation. Before migrating, a database audit is mandatory, not optional.
The second risk is GDPR compliance (the European data protection regulation). AI systems that profile user behaviors must comply with the legal bases for processing. If you use behavioral data to personalize, verify that the consent collected also covers this use.
The third risk is vendor lock-in — vendor lock-in. Choose platforms with open APIs (interfaces that allow you to connect other systems) and always verify that your data is exportable. On the topic of AI agent governance, it is also worth reading the analysis on OpenAI and out-of-control AI agents .
| Risk | Warning sign | How to manage it |
|---|---|---|
| Outdated or duplicate data | Email bounce rate > 5% | Audit and database cleanup before migrating |
| Insufficient GDPR consent | Consents collected before 2018 without updates | Re-opt-in campaign before turning on AI profiling |
| vendor lock-in | No data export API | Choose platforms with open standards |
| AI Hallucinations | Messages out of context or mismatched with the brand | Mandatory human review for high-impact content |
| Over-automation | The team no longer knows what's running | Centralized registry of all active workflows |
Why everyone is talking about it now: is the Italian market really ready?
The AI market in Italy is worth 1.8 billion euros and grows by 50% in a year, but real adoption in SMEs is still low. Those who move now have a concrete competitive advantage.
The market is growing, but actual adoption is concentrated in large enterprises. Among SMEs, only 8% have started at least one AI project, according to the Digital Innovation Observers from Politecnico di Milano . Italy ranks last in Europe for the adoption of smart automation (23%), far behind the United Kingdom (32%), Spain, and Germany (26%), according to data from InnovationCity . The delay is real, but those who start now still find little competition on the advanced personalization front.
Italian B2C e-commerce exceeded 62 billion euros in 2025, with 35.2 million digital consumers, according to the Politecnico di Milano and Netcomm Observers . At this scale, personalized automation isn't a luxury: it's infrastructure.
How to actually get started: the steps to migrate from traditional automation to AI automation?
Migration is done in phases, not all at once. The first step is always data cleanup, not platform selection.
The most common mistake is choosing the tool before understanding your data. A poorly fed AI system produces worse results than a well-configured classic automation. Start from the ground up.
- 1 · Internal team / consultant Contact database audit Clean duplicates, update GDPR consents, segment by source and behavior. Goal: a reliable database.
- 2 · Marketing team Map existing flows List all active automation flows. Identify the high-volume ones and those that change often: they are prime candidates for priority migration.
- 3 · IT team / partner Evaluate necessary integrations Make sure your CRM, e-commerce platform, and analytics tools talk to each other. Without unified data, AI has no context.
- 4 · Marketing team + management Choose the platform or the approach Based on your priority flows and budget, decide whether to upgrade your existing platform or migrate. Test on a small segment before scaling up.
- 5 · Marketing team Set up the human-in-the-loop model Define which decisions the AI can make independently and which ones need human approval. Write down these rules before turning on the system.
- 6 · Marketing / analytics team Measure and iterate Monitor KPIs week by week for the first three months. Adjust model parameters based on real results.
What are the most common mistakes made by those trying to take this leap?
90% of mistakes happen even before choosing the platform: dirty data, wrong expectations, and team not involved.
According to Omnichannel Customer Experience Observatory of the Polytechnic University of Milan , about one in three companies still hasn't launched any personalization initiatives in 2025. Among those that do, most use basic approaches. The problem isn't the technology: it's the method.
| Error | Why it happens | How to avoid it |
|---|---|---|
| Start with the platform, not the data | You want to see something working immediately | Database audit before buying anything |
| Thinking that AI will replace the team | Exaggerated expectations from vendors | Define human roles in the process immediately |
| Activating too many flows at once | Initial enthusiasm | Pilot on one segment, then scale |
| Not updating GDPR consents | Assuming old consent is still enough | Re-opt-in campaign before migration |
| Not tracking results for the first 90 days | Lack of predefined KPIs | Choose 3 key metrics before starting |
| Ignoring content quality | People think AI writes by itself | Strategic copywriting remains human |
How do you know if the new system is actually working?
Positive signs show up within 60-90 days: rising open rates, fewer unsubscribes, and fewer hours spent managing flows manually.
| Indicator | What it measures | Alert threshold |
|---|---|---|
| Email open rate | Message relevance for the recipient | Under 20% after 60 days: review your segmentation |
| Click-through rate (CTR) | Quality of the offer and sending time | Under 2%: the content is not contextual enough |
| Unsubscribe rate | Annoyance perceived by contact | Over 0.5%: you're sending too much or out of context |
| Weekly hours of flow management | Team operational efficiency | If it doesn't drop within 90 days, the system is not configured properly |
| Conversion rate per segment | Effectiveness of personalization | Compare with the pre-migration period, same segment |
| Errors or off-brand messages | Quality of human control in the loop | Even a single case requires immediate process review |
So, is it worth it for your company?
It is worth it if you have more than 5,000 active contacts, frequently changing flows, and a team wasting hours managing rules. The return in terms of time and message relevance is measurable within three months.
A middle ground you are an SME with average volumes and an already working platform. You can start by adding AI features to the platform you use, without migrating everything. Test on one flow, measure, then decide.
Not yet if you have fewer than 1,000 contacts, simple and stable flows, or you don't have a clean database yet. Consolidate the basics first: a well-configured traditional automation beats poorly fed AI.
How we see it at SHM
For us, AI in automation only makes sense if it gives real time back to the team — not if it adds another tool to manage. We believe technology should free people from routine, not create new layers of complexity to monitor.
In our experience with companies managing complex catalogs and multi-channel lead flows, the critical point is never the platform: it's data quality and clarity on decision-making. An AI system configured without these foundations produces noise, not results.
If you're wondering whether and how to evolve your automation, our starting point is always an audit: figuring out where you are right now, before deciding where to go. You can explore the topic in our hub at AI and process automation or dive straight into our services for strategic AI integration and of Data processing and business intelligence . For those who want to train their internal team, we also have a training path for business process automation training .
The article on OpenAI AI agents for SMEs offers another perspective on how these tools are changing operational marketing.
Frequently Asked Questions
Not necessarily. Many platforms already in use (HubSpot, ActiveCampaign, Salesforce) are adding AI features. The first step is to check if the functions available in your current plan are enough for your priority flows. A full migration only makes sense if your current platform doesn't support the integrations you need.
The first signs appear within 60-90 days: open rate, clicks, and unsubscribes are the fastest to measure. The benefits for team operational efficiency (fewer hours spent managing rules) are already visible in the first few weeks. Revenue results require at least one complete purchase cycle.
It depends on how it is configured. Systems that profile user behavior must have a valid legal basis (consent or legitimate interest, depending on the case). Before activating any AI personalization system, check that the consents collected also cover this type of processing and update your privacy policy.
This is one of the real risks, especially in the initial stages. The human-in-the-loop model serves precisely for this: high-impact messages (offers, sensitive communications, tone of voice) are approved by a person before sending. Immediately define which flows require human approval and which can run autonomously.
With under 1,000 contacts and simple flows, traditional automation is still enough and cheaper. AI makes sense when there are many flows, they change often, or they require personalization that would be impossible to handle manually. The advice is to consolidate your database and basic processes first.
This is the issue where the data used by automation gets old without being updated: a customer who bought six months ago still gets new buyer messages, for example. Modern AI systems try to fix this by constantly updating the contact profile based on recent signals. But it only works if the starting database is clean.
Automation has never been a tool problem: it has always been about data, processes, and who decides what — AI changes the speed, not the substance.
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.
