- Is investing in AI really worth it, or am I throwing money away?
- What does the research really say: does AI not work, or is it being used badly?
- How are Italian companies doing with AI?
- What are the three concrete scenarios: what changes for an e-commerce store, a professional firm and a manufacturing company?
- Which tools and approaches can I compare to avoid mistakes?
- What risks do I need to manage before investing in AI?
- Why it's in the spotlight now: the AI market is growing, but profits aren't?
- How to get started: the steps to turn AI spending into results
- What mistakes are made most often when introducing AI in a company?
- How do I know if AI is really delivering results for my company?
- So, is investing in AI worth it for your company?
- How we see it at SHM
- Frequently Asked Questions
- AI delivers profits almost only to large tech companies: outside that sector, margins stay flat or get worse in the initial adoption phase.
- The problem isn't the technology, but the lack of an operational strategy: vague goals, no KPIs, no dedicated owner.
- Companies that integrate AI deeply improve their margins by +5%, but only 11% of large enterprises have gotten there.
- Before spending more, you need to measure: define a use case, an indicator and an owner.
80% of companies have invested in AI without recording any measurable productivity gain. If your company has bought licenses without seeing results, you're not alone, but there's a specific reason.
Is investing in AI really worth it, or am I throwing money away?
It depends on how you're investing. Buying licenses without an operational plan almost always produces zero returns; integrating AI into specific processes with measurable goals can improve margins significantly.
The question every entrepreneur should ask isn't "Am I using AI?" but "Do I know what it's giving me?". According to an analysis published on TechCrunch, the chief economist of Apollo Global Management examined the 493 S&P 500 companies that are not pure tech, leaving out giants like Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla, and found profit margins and revenue per employee essentially flat. Zero measurable financial impact.
This isn't a temporary delay. The issue is structural: AI is an excellent calculator, but in business most decisions require judgment, trust, culture and accountability, elements that no language model handles on its own.
| Your situation | Is it worth investing? | why |
|---|---|---|
| You've bought licenses but haven't defined KPIs | No, for now | You're paying for a tool without measuring its use |
| You have a repetitive process with structured data | Yes | AI automates and frees up measurable time |
| You want to replace human judgment on critical decisions | No, for now | AI doesn't handle accountability and reputational risk |
| You have an AI owner and clear goals for each function | Yes | The minimum conditions for a real ROI are in place |
| You're an SME with no in-house AI training | No, for now | Without in-house skills, adoption stays superficial |
What does the research really say: does AI not work, or is it being used badly?
The data converges on one point: AI works, but only when it's deeply integrated. Most companies are still at the stage of licenses purchased and left unused.
A study by the MIT Initiative on the Digital Economy on S&P 500 companies found a J-curve in AI adoption: non-tech companies in the early stages of adoption record profit margins 2-3 percentage points lower than those that haven't adopted anything yet. Only those that reach deep integration see a margin improvement of +5%. The catch? Only 11% of large enterprises have reached that level, according to data published by MIT IDE.
A survey of almost 6,000 managers in the US, UK, Germany and Australia, published by NBER, confirms it: more than 80% of companies record no measurable productivity gain from AI.
The J-curve is the key. Those who buy an AI tool and use it superficially, without redesigning processes around it, bear costs without reaping benefits. Those who rebuild the workflow around AI, train their people and measure results get to a +5% margin. The leap isn't technological: it's organizational.
| AI adoption stage | Impact on margins | Share of companies |
|---|---|---|
| No adoption | Baseline (0%) | Non-tech majority |
| Initial adoption (licenses, experimentation) | -2/-3 percentage points | Most adopters |
| Deep integration (redesigned processes) | +5 percentage points | Only 11% of the S&P 500 |
How are Italian companies doing with AI?
The Italian market is growing fast, but the gap between those who buy licenses and those who get results is huge: 84% of large enterprises have at least one generative AI tool, but only one in five actually uses it across multiple functions.
The Italian AI market reached €1.8 billion in 2025, up +50% from the previous year, according to the Artificial Intelligence Observatory of Politecnico di Milano. That growth sounds good, but it hides a paradox: 84% of large enterprises have purchased licenses for at least one generative AI tool, yet only 20% report pervasive use across multiple business functions.
Among SMEs the picture is even starker: 76% have neither invested nor plan to invest in AI, and only 7% have launched structured training programs, according to the 2025-2026 Research of the Digital Innovation in SMEs Observatory.
Three real-world scenarios: what changes for an e-commerce store, a professional firm and a manufacturing company?
The impact of AI depends on the kind of process you apply it to. Repetitive processes built on structured data deliver quick returns. Processes that call for judgment and relationships stay in human hands.
Three illustrative scenarios to show where AI creates value and where it doesn't. The numbers are built to show the reasoning, not to reflect real client data.
| Company type | Process applied | Expected result | Main risk |
|---|---|---|---|
| E-commerce (50 orders/day) | Automated replies to return and tracking requests | 60-70% less operational time spent on that workflow | Wrong tone on complex cases → reputational damage |
| Professional firm (consulting) | Document summaries and draft reports | 2-3 hours saved per week for each consultant | Factual errors go undetected without human review |
| Manufacturing company (100 employees) | Production planning based on historical data | Less waste and better shift optimization | Integration with existing management systems is costly and slow |
The pattern is always the same: AI works well where the process is repetitive, the data is available and mistakes can be fixed. Where judgment comes into play, like a negotiation, a risk assessment or a relationship with a strategic client, AI can support but not replace.
If you're deciding where to start, the article on the ROI of AI in business offers a practical method to measure returns before you spend.
Which tools and approaches can I compare to avoid mistakes?
There is no single right tool: the choice depends on the process, the volume of data and the company's digital maturity. The deciding factor is always the operating strategy, not the platform.
| Approach | Suitable for | Indicative cost | Risk |
|---|---|---|---|
| AI SaaS licenses (e.g., Copilot, ChatGPT Team) | Companies that want to experiment on individual tasks | Low (monthly subscription) | Superficial use with no impact on processes |
| AI integration into management software | Companies with structured processes and historical data | Medium-high (customization) | Long timelines, vendor dependency |
| Custom AI agent development | Companies with complex workflows and high volumes | High (custom development) | Requires in-house expertise or a reliable partner |
| In-house training + standard tools | SMEs starting from scratch | Low-medium | Slow results without a dedicated lead |
One often overlooked point: fewer than half of Italian companies have a dedicated AI lead (45%), and only 35% define management KPIs to measure its impact, according to data from Oxford Economics / SAP (2026). Without these two elements, any tool is a waste.
What risks do I need to manage before investing in AI?
The main risks are not technical but organizational: wrong expectations, lack of governance and no training are the most frequent causes of AI investments that deliver no results.
The Apollo economist hit on a precise point: AI doesn't understand accountability, trust, company culture, unwritten rules or ethical trade-offs. These elements drive revenue growth and reputation for a company. Handing them to an automated system without supervision is the fastest way to create problems.
On the security front, it's also worth reading the analysis of the risks of autonomous AI models for corporate compliance: the risks concern not only profits, but also data protection and legal liability.
| Risk | Warning sign | How to manage it |
|---|---|---|
| Unrealistic expectations | "AI will solve everything" | Define a specific, measurable use case before you buy |
| No AI lead | Licenses bought by IT without involving the business | Appoint an internal owner with clear goals |
| Low-quality data | AI produces inconsistent or wrong output | Audit your data before integration |
| No training | Employees don't use the tool, or use it badly | A structured onboarding plan, not just a tutorial |
| No KPIs | No way to know if the tool works | Define 1-2 measurable indicators for each use case |
Why it's in the spotlight now: the AI market is growing, but profits aren't?
The growth in AI investment is real and accelerating. But the gap between spending and financial return is widening, not shrinking, which makes a change of approach urgent.
Italian companies expect to invest an average of $18.4 million in AI in 2026, with 45% growth over the following two years, according to Oxford Economics / SAP. Expected ROI rises from 20% in 2026 to 38% by 2028. These are optimistic forecasts, and the gap between expectations and real results is exactly the problem highlighted by the Apollo, MIT and NBER data.
Productivity attributed to AI shows positive but modest signs: about 0.6% in 2025, with growth forecast at 1.8% in 2026, according to a CEPR/NBER study of US companies published on VoxEU. The numbers are real, but far from what those who invested expecting rapid transformation had in mind.
How to get started: the steps to turn AI spending into results
The starting point isn't choosing a tool, but identifying a specific process with a measurable problem. Everything else comes after.
- 1 · ManagementPick a process, not a toolIdentify a repetitive workflow with available data and a measurable cost (time, errors, resources). That's your first use case.
- 2 · AI Lead / Internal OwnerAppoint a leadWithout an owner with clear goals, adoption stays an experiment. They don't have to be a techie: they need to understand the process and have the authority to change it.
- 3 · Operations TeamDefine 1-2 KPIs before you startExamples: average customer response time, number of errors in a report, hours of manual work on a task. Measure first, compare later.
- 4 · IT / External PartnerChoose the simplest tool that worksDon't chase the most advanced solution. Look for one that fits your current systems and that your colleagues will actually use.
- 5 · HR / AI LeadTrain people before go-liveA 2-4 hour onboarding plan for direct users dramatically cuts the risk that the tool gets abandoned.
- 6 · ManagementMeasure after 60-90 days and decideIf the KPIs improve, scale up. If they stay flat, change the process or the tool. Don't wait a year to evaluate.
To see how AI fits into daily workflows without upending the organization, the analysis of AI in Jira and Confluence shows a concrete case of integration into tools teams already use.
What mistakes do companies make most often when introducing AI?
The most common mistakes aren't about technology but about change management: companies buy before they understand, measure too little and train too late.
- Buying without a defined use case. "Let's try it and see what it does" is the surest way to waste budget. Every license should answer a specific problem.
- Expecting immediate results. MIT's J-curve is real: the first months of adoption often make the numbers worse, because people are learning and processes are being redesigned.
- Ignoring training. Only 7% of Italian SMEs have a structured AI training program. Without training, the tool is used badly or not at all.
- Not naming an owner. If AI belongs to everyone, it belongs to no one. You need an owner with clear goals and real authority.
- Measuring satisfaction instead of results. "The team likes the tool" is not a KPI. You need data on time, costs or quality.
How do I know if AI is really delivering results for my company?
You need indicators tied to the specific process where you applied AI, not generic metrics. If you didn't define KPIs before starting, now is the time.
| Indicator | How it is measured | Positive signal |
|---|---|---|
| Average time per automated task | Hours per week before and after | Over 20% reduction in 60 days |
| Error rate on AI output | Revisions needed / total outputs | Drops below 10% after onboarding |
| Actual tool adoption | Active users / licensed users | Exceeds 70% after 30 days |
| Impact on process margins | Process cost before and after | Measurable reduction in 90 days |
| Team satisfaction | Internal survey (scale 1-5) | Average above 3.5 after 60 days |
So, is investing in AI worth it for your company?
It's worth it if you have a repetitive process with available data, a dedicated owner and at least one defined KPI. In that case, AI can cut operating costs and free up time in a measurable way within the first 90 days.
Somewhere in between: you're still exploring and haven't found the right use case yet. Start with a limited pilot on a single workflow, measure the results, then decide whether to scale.
Not yet if you're looking for a tool to replace human judgment on strategic decisions, key client relationships or processes with a high reputational risk. In that context, AI supports but doesn't decide.
How we see it at SHM
For us, AI is the means, and people remain the goal. We believe bringing AI into a company only makes sense when it helps give time back to people, not when it just uses up an innovation budget. Every project we take on starts with a simple question: what specific problem are you trying to solve, and how will you measure it?
We've worked with manufacturing and consulting companies that had already bought AI tools without an operating strategy. The first step wasn't changing the tool, but redefining the process and naming an internal owner. Only then did we move on to technical integration.
If you're evaluating where and how to invest in AI without wasting budget, we can help you build a measurable strategy: from strategic AI integration in your processes, to data processing for business intelligence, to AI training for company teams. You'll find all our in-depth content on the topic in our hub dedicated to AI and process automation.
It's also worth keeping an eye on how AI is changing online visibility: the article on OpenAI Zero Data Retention explains how AI APIs are evolving on the privacy front in campaigns as well.
Frequently Asked Questions
Buying licenses is only the first step. Without a defined use case, an internal owner and measurable KPIs, the tool gets used superficially or not at all. MIT data shows that companies in the early stages of adoption often see their margins get worse before they improve.
On repetitive, well-defined processes, the first signals show up in 60-90 days. For an impact on a company's overall margins, MIT data indicates you need deep integration, which most companies reach in 12-24 months if they start with a clear strategy.
It works for SMEs too, but it takes a different approach. SMEs have fewer resources for training and integration, so it makes sense to start with a single simple process using standard tools. 76% of Italian SMEs haven't invested in AI yet, which means that those who start now with a solid method gain a real competitive advantage.
You don't necessarily need an in-house technical expert. You need a process owner: someone who knows the workflow and has the authority to change it. The technical side can be handled by an external partner, but responsibility for the goals must stay in-house.
Look for a process with three traits: it's repetitive, data is available, and the cost of a mistake is recoverable. Answering customer support requests, summarizing documents and planning based on historical data are often the most effective starting points for non-tech companies.
No, and the data backs it up. Apollo's economist pointed out that AI doesn't handle accountability, trust, company culture or ethical trade-offs, all of which drive reputation and revenue. AI can support decisions with data and analysis, but the final judgment stays human.
The problem isn't AI. It's that most companies are buying a tool without yet having a clear question to answer.
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.
