From knowing to doing: the real challenges of AI adoption in accounting firms

The questions being asked about carbon numbers now look increasingly like the questions asked about the P&L and balance sheet: who owns the figure, what evidence supports it, and how it reconciles back to spend, suppliers, assets and operational activity. That shift is exposing a gap between the scrutiny applied to sustainability claims and the way many organisations still capture the underlying data, often across spreadsheets, emails and supplier documents where version control, approvals and evidence trails are harder to demonstrate consistently. As a result, more organisations are pulling carbon data closer to core financial controls, using ERP for accounting as the anchor point for ownership, approvals and traceability, not to “do sustainability” in a new system, but to make emissions reporting stand up to an audit-style challenge. 

But these results represent outcomes, not the underlying explanation. The real driver is what our firms are doing within their own markets: strengthening capability, developing new service offerings, and responding directly to the needs of clients navigating increasingly complex environments.

Across PrimeGlobal’s recent series of regional technology webinars, one pattern came up in every conversation: firms know AI matters, but most are still in the experimentation phase.  

In Latin America, participants were straightforward about where they stand. “We are still at an early, incipient stage,” one said. “Navigating, experimenting, testing.” In Africa there is an additional problem: certain tools that practitioners in London or New York now take for granted, arrive in sub-Saharan markets later and at greater relative cost.  

In North America and some parts of Europe and Asia Pacific, member firms have moved on to managing tools in use, governing staff behaviour and running training. One firm launched a “bounty programme” to find out how their people were using AI because individual enthusiasm had outpaced company-wide rollouts. The challenge there is not whether to adopt, but how to manage adoption that started without anyone formally deciding to start it. 

One concern is raised globally: over-reliance. AI tools produce work that looks authoritative, arrives quickly, and is hard to interrogate unless you already know the subject. A participant in PrimeGlobal’s African sessions pointed to cases already visible in the legal profession, where courts in South Africa and the UK have reprimanded practitioners for submitting AI-generated work that had not been properly reviewed. Accounting carries the same exposure, particularly in areas like tax, where jurisdiction-specific nuance is everything. 

The most useful framing came from a practitioner in one of the Latin American sessions: treat AI output the way you would treat work submitted by a junior staff member that a partner has not yet reviewed. You would not send it to a client unseen. It works as a principle because it maps onto a quality control discipline that accountants already use, without requiring anyone to understand the technology. Some firms have already formalised this in internal AI usage policies. 

Many examples of AI use are small and easily replicable: a firm consolidating questionnaire responses from five clients in 35 seconds rather than most of a day; a practice auto-drafting client communications in a consistent house style; a firm cross-referencing financial statements against international accounting standards and flagging gaps for review. None required specialist technical staff or significant investment. 

What firms do need is someone inside the firm with ownership of the process. Without that, firms tend to default to scattered experimentation with no shared view of what is working or what their data governance obligations require. 

In many regions, smaller firms are not necessarily at a disadvantage. The big firms have resources but also inertia. For example, they are not curating the relationships or building the capabilities that AI-enabled advisory services will require in sub-Saharan markets. That is a gap small firms can fill before the large players get there. 

The knowledge that matters most to an independent firm in Guadalajara at this point, is what a similar firm in Mexico City found out about data privacy when they put client documents into a paid AI tool, or what a practice in Missouri worked out about managing AI across a multi-partner firm with no dedicated IT function. PrimeGlobal’s community plays a critical role in circulating this kind of hands‑on knowledge through open conversations and shared practitioner experiences. 

Across recent webinars, members proposed several ways that PrimeGlobal can support consistent, safe, and practical AI adoption. Many of these ideas are already taking shape: 

1. A curated library of recommended AI tools
   Members asked PrimeGlobal to maintain a vetted list of safe, reliable AI solutions suited for tax, audit, advisory, and internal operations.  
2. Shared templates for AI governance and client communication
   Firms expressed a need for clear guidance on data handling, consent language, model review requirements, and retention controls—all areas where network‑level consistency would be helpful.  
3. Regional working groups and knowledge communities
   These groups, already active across Latin America, Africa, and North America, are proving to be one of the most practical routes for firms to share real experiences, compare methods, and learn from each other.  
4. Training across all seniority levels
   Members asked for partner‑level training on strategy and ethics, manager‑level training on workflow redesign, and staff‑level training on prompts and tools.  
5. A PrimeGlobal AI readiness framework
   Members welcomed the idea of a structured readiness model that helps firms assess their current position, plan their next steps, and identify realistic early wins. 

As AI continues to shape the future of the profession, PrimeGlobal remains committed to ensuring that every member, regardless of size, geography, or current digital maturity, has access to the support, knowledge, and community needed to adopt AI responsibly and effectively.