Why most AI pilots stall before they add value

A survey of 2,485 small and medium-sized businesses found that 59.6% were already using AI. That sounds like progress. But when you look one level deeper, a more instructive picture appears. Of those AI-using businesses, 80.9% were using AI only as a chat tool: asking questions, drafting text, summarising documents. Just 19.1% had connected AI to business systems or automated any part of a workflow. The gap between experimenting with AI and embedding it into how work actually gets done is wider than most managing directors realise.

The real shape of the AI adoption gap

The survey data tells a clear story. More than four in five AI-using SMEs have not moved beyond conversational use. They have accounts, they have licences, and team members are opening chat windows every day. But the underlying processes, the purchase approvals, the reporting cycles, the invoice matching, the member renewals, the engineer job sheets, remain unchanged.

This is not a technology problem. The tools capable of automating those processes already exist. It is an organisational problem: the gap between a useful experiment and a repeatable operational process requires deliberate effort that most businesses have not yet made time for.

For operations managers and managing directors, this distinction matters. AI used only in chat does not appear in any process map. It cannot be audited, measured, or handed off. Its value stays personal and invisible.

Why cost anxiety is the wrong diagnosis

The survey also asked about barriers. Among businesses still at the chat stage, 74.5% cited concerns about cost or return on investment. That figure is often quoted as the reason SMEs are not progressing.

But look at the businesses that had progressed. Among those that had connected AI to systems and automated at least one workflow, 54.0% still cited the same cost concerns. The anxiety did not disappear once they moved forward. It was present throughout.

This is important evidence. Cost concern is not what separates the businesses that progress from those that do not. Something else is keeping most SMEs at the chat stage, and the data points clearly at operational readiness: the capacity to identify a process, connect AI to it, and measure what happens.

For finance managers evaluating whether to expand AI spending, this reframes the question. The issue is rarely whether AI is affordable. It is whether the organisation has built the internal conditions to turn AI use into a measurable business process.

Why chat-only usage is 80% a people problem

At gecco, our view is that AI adoption is 80% people and culture, 20% technology. The survey data supports this directly.

The businesses stuck at chat are not stuck because the automation tools are too complex or too expensive. They are stuck because no one has been given the responsibility, the time, or the method to move AI from personal use to shared process. That is a management and culture question, not a software question.

Operations managers often recognise this instinctively. They can see which processes are high-volume and repeatable. They know which team members are already using AI productively on their own. What is missing is a structured approach to converting that individual behaviour into something the whole team can rely on.

This is also why the 19.1% who have moved beyond chat tend to report a shift in how they think about AI. Once one process is automated and the results are visible, the conversation inside the business changes. The question stops being "should we use AI?" and starts being "what should we automate next?"

What businesses that progress actually do differently

The 19.1% of surveyed businesses that had connected AI to their systems had made a specific choice: to treat one process as a pilot with a measurable output, rather than letting AI use remain diffuse and personal.

The pattern, drawn from what the evidence shows about businesses that move forward, looks like this.

First, they identify one high-volume, repeatable process. Not a complex strategic task: something that happens frequently, involves clear inputs and outputs, and currently consumes disproportionate staff time.

Second, they connect AI to the systems already in use for that process. This is where most businesses stall without support. Connecting an AI model to a CRM, an inbox, a finance platform, or a document store requires workflow configuration that goes beyond a chat interface.

Third, they measure the result before expanding. Demonstrating ROI from AI remains a genuine challenge for many organisations, but businesses that automate a single defined process find it easier to point to a before-and-after comparison. That evidence then funds the case for the next stage.

This is not a sophisticated programme. It is a disciplined choice to do one thing properly before doing more.

Implementation safeguards every operations manager should know

Connecting AI to business systems introduces data-protection considerations that chat-only use does not. Before any workflow automation goes live, SMEs should work through several practical questions.

Data-protection risk assessment comes first. Which data will the workflow access, process, or store? Does it include personal data covered by UK GDPR? If so, a short data-protection impact assessment is advisable before building anything.

Access permissions need to be scoped carefully. AI workflows should access only the data they need for the specific task. Broad access credentials create unnecessary risk and complicate audit trails.

Retention rules matter. Automated workflows can generate significant data volumes quickly. Retention periods should be defined at the design stage, not added later.

Human oversight must be built in. For any workflow that takes action on behalf of a team member, including sending emails, updating records, or processing approvals, there should be a defined review point. Fully autonomous action on consequential tasks is rarely appropriate for a first deployment.

Finally, avoid using confidential business or client information in consumer-grade AI tools without appropriate controls. Enterprise-grade platforms with contractual data protections exist for this reason.

For finance managers and operations managers, these are not reasons to delay automation. They are the checklist that makes automation defensible when questions are asked internally or externally.

Making this work for your business

The gap the survey describes is closeable. It does not require a large programme or a long delay. It requires identifying the right process, connecting it properly, and measuring what changes.

gecco's Automations service builds no-code, event-triggered workflows that connect AI to the platforms a business already uses, drawing on 7,000+ platform connections. Rather than leaving AI at the chat stage, it identifies one high-volume process, connects it to existing systems, and builds in a full audit trail so results can be measured.

For an operations manager who already knows which process is the right candidate, the practical next step is straightforward: define the inputs, define the outputs, and establish how you will measure success before anything is built. That clarity is what separates a useful automation from another experiment that fades.

For a managing director looking at the survey numbers and recognising the business in them, the question is not whether to invest more in AI. It is whether the organisation has the internal conditions to move AI from chat to process. Businesses that measure AI against a specific process outcome report clearer ROI than those measuring general productivity. That shift in framing is available to any SME that chooses to make it.

Start with what the data already tells you

If your business sits in the 80.9% using AI only in chat, the survey data offers a practical prompt rather than a cause for concern. The businesses that have progressed did not have better technology or larger budgets. They made a deliberate choice to connect AI to one process and measure what happened.

Take the AI Readiness survey. You will receive 65+ free resources and a custom AI Readiness report based on your answers. From there you can book a free 30-minute showcase to walk through the results with a gecco adviser.

---

Website · LinkedIn · Case Studies · Newsletter