That makes the headline less a measure of mature AI adoption than a measure of easy access. A staff member using ChatGPT to draft an email, Copilot to summarise a document, or Google Translate to prepare customer correspondence is using AI. It does not mean the firm has integrated approved data sources, defined human review, set permissions, established retention rules, or connected an AI workflow to its CRM, accounting platform or line-of-business software.
The distinction is becoming important for Windows administrators and Microsoft 365 tenants. “We use AI” is no longer enough information for a business owner or IT provider to assess exposure, cost, security controls or potential productivity gains.
The 92% figure measures access, not integration
According to the Irish Independent, market-research firm Amárach surveyed more than 400 Irish firms during April and May, and the Small Firms Association found generic assistants in use at 92% of respondents. Sales, marketing and basic administrative work such as emails and file storage were identified as leading uses.
Those are precisely the areas where public or broadly available AI services have the lowest entry cost. A user can paste text into a chatbot or invoke a feature inside an office application without a project plan, an integration budget or an internal AI specialist. The result may save time on a first draft, a summary or translation, but it may also sit outside the company’s normal records, data-governance and approval processes.
The survey’s 30% figure for custom-made AI is the more useful indicator of operational change. It suggests fewer than one in three responding firms has gone beyond off-the-shelf assistance toward a tool adapted to its own processes. Even that category requires care: the reporting does not specify whether “custom-made” means a bespoke internal application, a configured chatbot, a vendor product tailored to a firm’s data, or a workflow assembled with low-code automation.
That missing definition matters. A Copilot Studio agent connected to a curated SharePoint library has very different security, maintenance and business-value implications from a custom system that accesses customer records or automatically makes recommendations. Both could plausibly be called customised AI.
The Small Firms Association’s earlier, official 2025 survey release pointed to the same underlying pattern. It said nine in 10 small firms used AI for at least one process, with simple-task automation and analytics reporting leading the uses, and said deeper adoption was constrained by technical expertise, time pressure and the lack of a clear strategy. Its director, David Broderick, described adoption at the time as shallow and concentrated in content generation and simple analysis.
The new 92% estimate should therefore not be read as a sudden, clean increase in enterprise AI deployment. It is broadly consistent with the SFA’s prior finding that AI is already common at the edge of small-business work. What remains scarce is governed, repeatable use that changes how a firm serves customers or completes core work.
The year-over-year comparison does not hold up cleanly
The Irish Independent says generic-assistant use declined to 92% from 96% last year, while use of custom-made AI rose. But the SFA’s public release from August 2025 reported “nine out of 10” businesses using AI for one or more processes, not 96% using generic assistants.
That is not necessarily an error. The surveys may have asked different questions, used different samples, changed the definition of AI use, or separated generic tools from bespoke systems only in the latest research. But neither the Independent report nor the SFA’s currently available 2025 release provides the 2026 questionnaire, response breakdown, weighting method or margin of error needed to establish a like-for-like trend.
The practical conclusion is straightforward: do not treat the apparent four-point decline as evidence that Irish small businesses are stepping back from AI. The more defensible reading is that generic AI usage remains near saturation among the survey respondents, while deeper deployment is still limited.
This is a familiar measurement problem with AI surveys. A broad question such as “does your business use AI?” can group together an employee who has tried a free chatbot once, a firm paying for Microsoft 365 Copilot licences, and a company running a controlled retrieval system over internal documents. Those three organisations have profoundly different requirements for identity management, data classification, licensing, support, audit trails and risk ownership.
For managed service providers, the survey is a warning against accepting a customer’s broad assurance that “we already have AI.” The useful follow-up is narrower: which tool, which tenant or account, which data is entered, who can approve output, and whether the organisation has a record of the use case.
Copilot use changes the governance question
Microsoft Copilot is named among the generic services used by firms in the report, but “Copilot” is not a single deployment model. Consumer Copilot, the version included with certain Microsoft 365 plans, Microsoft 365 Copilot, Copilot Chat and a purpose-built Copilot Studio agent differ in licensing, data grounding and administrative controls.
For a small company, this distinction can determine whether business material is being handled within the Microsoft 365 tenant’s established identity and compliance boundaries or copied manually into a separate web service. The issue is not that a public AI tool is automatically unsafe; it is that unmanaged use removes the organisation’s ability to answer basic questions after an incident, an erroneous customer communication or a subject-access request.
A modest but meaningful AI programme should begin with an inventory rather than a procurement exercise:
- The business should identify every AI service staff use, including browser-based chatbots, translation services, meeting transcription and AI features inside Microsoft 365.
- The company should define which data types may be entered into each service, with particular attention to customer information, payroll details, contracts, pricing and confidential product material.
- The organisation should decide where human review is mandatory, especially for customer-facing content, employment decisions, financial information and instructions that could create legal or safety consequences.
- Administrators should establish whether staff are using personal accounts, unmanaged browser sessions or approved organisation-managed identities.
These controls sound basic because they are basic. They are also the boundary between experimentation and a tool that can be safely expanded.
The SFA’s call for training money addresses a genuine constraint, but training alone will not create usable systems. Small firms also need templates: an approved-use policy, a data-handling rule, standard language for AI-assisted customer content, a process for checking accuracy, and a named person who owns each use case. Without those, an investment in skills may simply produce faster, more widespread use of tools that remain disconnected from the company’s actual work.
Regulation now reaches the customer-facing edge
The timing of the Irish survey coincides with a change in the European Union’s AI rules. The European Commission says the AI Act’s Article 50 transparency obligations began applying on August 2, 2026, eight days before this report was published.
The legal shift is narrower than the suggestion that every piece of AI-assisted work must carry a label. The Commission says providers of relevant AI systems must make users aware when they are interacting directly with AI and add machine-readable markers for AI-generated or manipulated content. Deployers have disclosure duties in specified cases, including deepfakes, certain AI-generated content published to inform the public without human review or editorial control, and uses involving emotion recognition or biometric categorisation.
For the small business producing a draft email with Copilot and having an employee edit and approve it, this does not create a universal requirement to stamp every message “AI-generated.” For a business publishing realistic synthetic images, audio or video, operating a customer-facing chatbot, or distributing public-interest content without meaningful editorial review, the compliance question is more immediate.
The Commission’s guidance gives small firms a reason to separate internal drafting from public deployment. Marketing teams are likely to be early candidates for AI-generated images, ad copy and social posts because the SFA identifies sales and marketing as a leading use. That is also where a casual approach to disclosure can quickly become a customer-trust problem, even before any regulator becomes involved.
The funding request is really a request for implementation capacity
Broderick has urged the Irish government to direct money from the €1.5 billion National Training Fund toward IT and AI skills, arguing that business owners lack the time, technical staff and money to deploy technology strategically. The size of the fund is not a new discovery: the SFA and other industry bodies have called for its surplus to be used for skills for several years, and the government had already set out plans to release funding over a multi-year period.
What is new is the sharper evidence that access to generic AI is no longer the central barrier. Small businesses already have access; 92% of the SFA’s respondents say they are using it. The harder task is converting scattered use into a few reliable workflows with measured outcomes.
That means the best public support would not be limited to introductory prompt-writing courses. A worthwhile programme would help a small firm identify one high-volume process, map the data involved, decide where an existing Microsoft 365 capability is sufficient, configure access safely, train staff, and measure whether the change reduced turnaround time or error rates. If it cannot clear that bar, the firm should not call it a productivity programme.
The survey points to a market in which AI has become routine at the desktop but is still uncommon as managed business infrastructure. For small Irish companies, the next spend should not be on another chatbot subscription. It should be on the policy, identity controls, data boundaries and narrowly defined workflow that make the tools they already use accountable.
Update: Survey adds figures on in-house software and implementation barriers (August 10, 2026)
Business Plus reports that 20% of surveyed firms are building or coding standalone in-house AI software, alongside the 30% using “customised AI models.” It also says 31% have not started or investigated AI adoption, despite 92% reporting use of generic tools—further evidence that respondents may distinguish casual tool use from operational deployment.
The outlet also reports that 31% cite insufficient technical expertise, 28% lack time to investigate AI and 17% lack finance. Its breakdown lists simple-task automation at 61%, dashboards and analytics at 52%, and “holding digital business files” at 65%, though the survey reporting does not define what AI-enabled file holding entails.
Update: Tailored AI users report efficiency and customer-service gains (August 10, 2026)
RTÉ reports that customised AI use rose to 30% of surveyed firms, from 22% last year. The comparison is more specific than the generic-AI figures cited previously, suggesting a measurable increase in firms moving toward systems tailored to their own needs.
Among businesses using specially designed models, nearly two-thirds said the tools improved efficiency, while almost half reported faster customer-service delivery. Those findings point to benefits where AI is connected to a defined operational task rather than used only for drafting or summarising.
The survey covered 404 companies, according to RTÉ. For IT providers and administrators, the results reinforce the case for starting with a contained workflow—such as customer-query triage or document processing—where access controls, review requirements and outcomes can be clearly measured.
References
- Primary source: independent.ie
Published: August 10, 2026 at 4:30 AM UTC
Majority of small firms now use AI tools like ChatGPT, survey finds | Irish Independent
The vast majority of small Irish companies are using generic AI assistants such as ChatGPT, Microsoft Copilot, or Google Translate, and lack the skills to upgrade to more specialised technology.
www.independent.ie
- Related coverage: uschamberfoundation.org
Half of Small Business Workers Use AI — Most to Boost Productivity, Not Automate Jobs | U.S. Chamber of Commerce Foundation
Half of all workers at small businesses already use AI at work — and the vast majority are using it to get more done, not to automate themselves out of a job. That is the central finding of the inaugural Main Street AI Monitor, released today by the U.S. Chamber of Commerce Foundation in...www.uschamberfoundation.org
- Related coverage: ai2.work
71% of Small Businesses Now Use AI: The 2026 Tipping Point | AI2Work
New 2026 survey data shows 71% of small businesses actively use AI. Explore adoption stats, top tools, ROI data, and a step-by-step implementation guide.ai2.work - Related coverage: ibec.ie
9 out of 10 Small Businesses Use AI tools, mostly for automation and analytical functions - SFA
Research from the Small Firms Association (SFA), shows that 9 out of 10 small firms surveyed are using artificial intelligence (AI) for one or more business processes. The nationwide survey was recently conducted by Amárach. The most popular uses for AI were reported to be automation of simple...www.ibec.ie
- Related coverage: sbecouncil.org
- Related coverage: ibec.ie
Small Firms Association Welcomes Government Measures for Small Businesses - SFA
The Small Firms Association (SFA), the representative body for businesses employing fewer than 50 people, has welcomed the Government’s new support package for small businesses. - Primary source: businessplus.ie
SFA sounds alarm over 'shallow' AI adoption
Research from the Small Firms Association (SFA) shows there remains a significant gap between basic AI usage and strategic adoption among small businesses.businessplus.ie - Primary source: rte.ie
Published: August 10, 2026 at 7:35 AM UTC