# Automating business processes: the risks nobody explains to SMEs

The **business process automation risks** small companies face are rarely technical. Gartner has published an uncomfortable forecast: **by 2027, more than 40% of projects based on agentic artificial intelligence will be cancelled**. The reason is organisational rather than technological: pre-existing problems that AI has made more visible — and more expensive. The figure deserves attention, especially in Italy, where according to the Digital Innovation Observatories of the Politecnico di Milano, 84% of small and medium businesses have not yet integrated artificial intelligence tools into their core processes. Those starting now face a real window of opportunity. They also face an equally real risk: investing in automation without first understanding what they are trying to automate.

## The underlying misunderstanding: AI does not replace a process, it reveals it

The wrong starting point is almost always the same. A company decides to use AI to speed something up — order handling, customer replies, document production — and picks a tool. It implements it. A few weeks later it discovers that the problem was not speed. It was fragmentation: data scattered across different systems, undocumented procedures, overlapping roles nobody had ever formalised. Automation solved nothing. It made the chaos faster.

This mechanism has a precise technical name, well documented in change management literature: automation amplifies the characteristics of the system it is grafted onto. If the system is solid, results improve. If it is dysfunctional, the dysfunctions multiply. An emblematic case cited by Gartner concerns a company that tried to automate accounting entries with an AI system: the model began generating creative output instead of coherent transactions. The error was not in the algorithm. It was that the company had never defined precisely what a correct transaction was. The process was implicit, entrusted to people's memory. The AI did not know where to start.

## Three concrete areas of risk for small businesses

Automation risks in small organisations cluster in three distinct, often interconnected areas.

The first concerns the **lack of operational documentation**. In many Italian SMEs — especially in manufacturing and professional services — operational knowledge lives in people, not in systems. Procedures, exceptions and unwritten practices cannot be automated until they are first made explicit. An AI system trained on implicit processes produces incoherent output, because it works on partial or contradictory data.

The second risk is the **fragmentation of data across unintegrated systems**. Management software, CRM, spreadsheets, email, corporate chat: they often coexist without talking to each other. Automating one step of this unintegrated system moves the bottleneck; it does not remove it. The flow breaks one step later than before.

The third — and often the most underestimated — is the **misalignment between the stated goal and the real one**. "We want to automate quote handling" can hide very different problems: slow response times, configuration errors, poor quality of the information collected during the sales phase. Automating quote production without addressing the upstream problems produces wrong quotes faster.

## The measurement problem: without KPIs you do not know if it works

One critical aspect, rarely addressed with the necessary concreteness, is defining success metrics before implementation. **The absence of clear KPIs is one of the main reasons automation projects are abandoned** in the intermediate phases. The reason is rarely that the project fails to work; it is that nobody knows how to establish that it does. Hours recovered per role, errors avoided, cost per process before and after, average customer response time: these measures must be set in advance, not searched for afterwards when the project is already under discussion.

The CFO survey conducted by Battery Ventures in 2026 on 129 finance leaders found that 21% cite unclear ROI as the main barrier to AI adoption — second only to model inaccuracy, cited by 71%. Two problems that look technical but are in fact organisational: ROI is unclear because it was never defined at the outset, and models are inaccurate because they are fed unstructured or incomplete data.

## What distinguishes projects that work from those that are abandoned

Companies that get real results from automation share a common methodological approach. Before choosing any tool, they **map the existing operational flows** in enough detail to identify where time is lost, where errors originate and where information deteriorates as it passes between people or systems. Only then do they ask which part of that flow is actually automatable — and with which technology.

Ramp, the US corporate finance management platform, reached 99% internal AI adoption by building internal infrastructure that connected all existing tools through unified access, removing the manual configurations that hindered employee adoption. The point was not the technology. It was removing the organisational friction that stopped people from using it. A replicable approach, at a different scale, in smaller organisations too.

The same principle appears in the experience of Block — the US company behind Cash App and Square — in its project to redesign the organisation around AI agents: the competitive advantage came from the quality of the data and processes the tool operated on, rather than from the tool itself. Structured data, documented flows, defined roles: these were the necessary conditions, not the consequences of automation.

## The role of preliminary analysis: a condition, not a cost

In the daily practice of Italian SMEs, the preliminary analysis phase is often seen as an extra cost to compress. The reasoning is understandable: everyone wants results as soon as possible. But that compression is exactly what generates the failures Gartner measured. Process analysis is not the bureaucratic premise to implementation: it is the condition that determines whether implementation will make sense.

A document automation system introduced in a company where documents are produced in heterogeneous formats, saved in non-standard paths and shared by email without defined naming will not produce efficiency. It will produce an automatic archive of disordered material. The problem is not automation. It was already there, invisible until someone tried to automate it.

This explains why the most solid implementations always follow a precise sequence: first understanding the real flow, then identifying the intervention points, then choosing the adequate tool. Reversing the sequence — choosing the tool and adapting the process to it — produces partial results at best, and adopts inadequate solutions at worst.

## A perspective for Italian SMEs

The Italian artificial intelligence market reached 1.2 billion euros in 2025, growing 58% year on year according to the Politecnico di Milano Observatories. It is real growth, driven by concrete investment. But the figures on actual adoption tell a more nuanced story: most small companies are still in the evaluation phase, often influenced by expectations built around the success stories of large organisations with very different structures, resources and digital maturity.

The right question is not which AI tool to adopt. It is: which processes, in the specific context of this company, with these data, these systems and these people, are ready to be automated? And which must first be redesigned? Answering that question requires an analysis that cannot be improvised and cannot be delegated entirely to the tools themselves, however sophisticated.

SMEs that approach this phase methodically — before choosing any platform — will find themselves in a much stronger position than those chasing the technology of the moment without first understanding what they want to transform. The difference, over the next two years, will be measurable.

## From process analysis to sustainable automation

Automating well means first understanding the process: where the data originates, who uses it, which steps generate errors, which activities need human control and which can be delegated to a system. EBM Solution designs custom software and automation starting from this preliminary reading. The goal is to build digital systems that reduce manual work and dispersion, while keeping control over critical activities and consistency with the way the company actually operates.

The same process work also opens a new offer. A connected product can support [a recurring service built on the connected product](/en/ecorner/2026/product-to-service-recurring-revenue.html), once the data and the responsibilities behind it are defined.

Automation moves the decision, not just the task, and [the governance of AI agents in business processes](/en/ecorner/2026/ai-agents-business-governance.html) is where responsibility has to be assigned.

## Frequently asked questions

**What are the main risks of business process automation for SMEs?**
The main risks are three: undocumented processes that AI cannot interpret correctly, data fragmented across unintegrated systems, and automation goals defined imprecisely against the real problem to solve.

**Why do many artificial intelligence projects in companies fail?**
According to Gartner, more than 40% of agentic AI projects will be cancelled by 2027, for organisational rather than technological reasons: missing KPIs, undocumented implicit processes and poor-quality data are the most frequent causes.

**Where should an SME that wants to automate its processes start?**
The correct starting point is mapping the real operational flows: where time is lost, where errors originate, where information deteriorates between one step and the next. Only after that analysis does it make sense to choose adequate automation tools.

## Sources

Gartner — forecast on agentic AI projects and cancellation by 2027 — https://www.gartner.com/en/newsroom
Osservatori Digital Innovation, Politecnico di Milano — artificial intelligence in Italian companies, 2025 — https://www.osservatori.net
Battery Ventures — CFO survey 2026 on AI adoption, ROI and model accuracy — https://www.battery.com
Ramp — internal AI adoption and unified access infrastructure — https://ramp.com
Block — organisational redesign around AI agents — https://block.xyz