If you opened any accounting software vendor’s website today and searched for the word “AI,” you would find it everywhere. AI-powered insights. AI-driven reconciliation. AI-assisted coding. AI-enhanced workflows. Smart features. Intelligent automation. Machine learning at the core.
The language is consistent, the claims are compelling, and the marketing is, it has to be said, very good indeed. The technology underneath it, in many cases, is considerably less impressive.
This is not a minor distinction. When accounting firms make investment decisions based on AI capability — when they choose platforms, justify costs to partners, and tell clients they are operating at the frontier of the profession — the gap between what is claimed and what is delivered matters enormously. It matters commercially. It matters professionally. And it matters strategically, because a firm that believes it has implemented AI when it has implemented something substantially less sophisticated is a firm that has stopped looking for the real thing.
This article is about how to see the difference — not as a technology specialist, but as a professional who needs to make informed decisions about the tools that will define their practice for the next decade.
A different vantage point
Most commentary on AI in accounting is written by people who came from accounting. They understand the profession deeply, and they understand AI as the profession is experiencing it: as a new, rapidly evolving feature set layered onto familiar platforms. I came from the other direction.
I have spent over a decade designing and deploying AI and automation systems in some of the most demanding, regulated, and operationally complex environments that exist — financial services, healthcare, critical infrastructure. Environments where AI is not a product feature. It is load-bearing. Where the consequences of a poorly designed system are not a frustrated user or a miscategorised transaction; they are regulatory sanctions, failed audits, and serious operational risk.
That origin — AI-first, entering accounting — is fundamentally different from accounting-first, adding AI. The ceiling of what you can build is shaped, more than anything else, by the depth of your understanding of the technology you are building with. This matters when evaluating any vendor’s AI claims, including ours. The question is not just what the product does, but how deeply the team behind it understands what they have built, and what they could build next.
Start with the goal, not the tool
The single most important thing any accounting firm needs to develop is not an understanding of AI. It is an understanding of what they are trying to achieve. What outcome matters to your firm? What does success look like in two years — specifically, measurably, in terms your partners would agree on? What is genuinely preventing you from getting there today?
These are not technology questions. They are business questions, and they need to be answered before any technology conversation begins. AI is one of many possible enablers. For automating high-volume, repetitive, well-defined workflows, it is an exceptionally powerful one. In other cases it is not the right tool at all — sometimes the problem is a process that needs redesigning, sometimes a data-quality issue no amount of AI can compensate for, sometimes a people, pricing or client-communication issue that AI does not solve.
The firms getting the most from AI are not the ones that started by asking “how do we use AI?” They are the ones that started by asking “what are we trying to do?” — and then evaluated AI honestly against other approaches. Start with the purpose. Let the technology follow from that.
What AI-washing actually means
AI-washing is the practice of applying the language of artificial intelligence to products and features that do not, in any meaningful sense, involve it. It is not unique to accounting; it has happened in financial services, healthcare and retail — every sector where AI became commercially attractive before the market developed the sophistication to interrogate the claims.
The tools being marketed as AI in accounting exist on a very wide spectrum. At one end is genuine artificial intelligence: systems that learn, adapt, reason, and act across complex multi-step workflows with meaningful autonomy. At the other end is a predefined rule that fires when a transaction matches a keyword and puts it in a category. Both are currently being sold to accounting firms as “AI.”
The maturity spectrum — where things actually sit
Level 1 — Rule-based automation. Predefined logic. If a transaction contains “HMRC,” put it in the tax category. No learning, no intelligence — just rules applied consistently at speed. Useful. Not AI.
Level 2 — Pattern recognition. The system learns from historical data and suggests the categorisation you used before. More sophisticated than Level 1, but still pattern matching trained on your own past behaviour.
Level 3 — Predictive analytics. Historical patterns used to forecast future states: cash-flow predictions, revenue forecasts, anomaly detection. Genuinely useful, but still extrapolating from the past rather than reasoning about the present.
Level 4 — Natural language interaction. The system understands plain-English queries, retrieves relevant data and answers conversationally. This is where most accounting platforms’ large-language-model features currently sit.
Drowning in manual, repetitive work? Tell us the task and we’ll show you what to automate.
Level 5 — Autonomous decision-making. The system makes decisions independently, within defined parameters, and acts on them without prompting. This is where meaningful workflow transformation begins. Very few accounting tools genuinely operate here.
Level 6 — Agentic AI. The system plans across multiple steps, uses external tools, handles situations it has not seen before, and pursues a goal with genuine self-direction. It is coming to accounting; it is not yet widely present in production form, whatever the marketing says.
Most embedded “AI” features sit firmly at Levels 1 and 2. Some advanced tools reach Levels 3 and 4. Levels 5 and 6 remain the territory of specialist, purpose-built platforms. Accounting is not behind by months. It is behind by years.
How to spot it in practice
“Can you show me a workflow that runs end-to-end without a human initiating each step?” If the demo shows smart suggestions that a human then acts on, that is Level 2 automation, not a workflow.
“How does the system handle something it hasn’t seen before?” Rule-based systems fail badly on unfamiliar inputs. Ask for a live demonstration with an unusual input and watch what happens.
“What is the system actually learning, and how do we know?” Many systems store a correction and reapply it next time the same pattern appears. That is memory, not learning. Vague answers are not answers.
“Is this feature built on your own model or a third-party API?” A significant proportion of “AI” features are third-party LLM APIs with a prompt and interface on top. Worth knowing where the competitive advantage really sits.
“If I correct the same error next month, will the system have learned from it?” If the same correction must be made repeatedly, that is automation with a marketing label, not a learning system.
Bolt-on vs. built-for
There is a structural dimension beyond individual features. Traditional platforms built for accounting have added AI as the market demanded it; other platforms were built for AI-native workflows from the ground up. This distinction determines the ceiling. Adding AI to a legacy architecture is impressive engineering — but engineering in service of an existing paradigm, not a new one. A platform built AI-first, where the architecture assumes systems will reason, decide and act autonomously by default, is a categorically different thing.
AI-washing has a short shelf life — but what replaces it matters more
AI-washing, as a problem, will largely resolve itself within a year or two — not because vendors become more precise, but because AI will become so commoditised that the label ceases to carry any signal. When every product includes AI as standard, claiming “AI” will be about as differentiating as claiming “cloud” is today.
What matters then is not whether the tools have AI, but whether the AI has been shaped to work for your firm specifically: personalisation, localisation, the ability to configure AI to reflect your way of working, your client relationship model, your terminology, your risk appetite. The firms that lead in five years will not simply be the earliest adopters. They will be the ones that made it their own.
A word on fairness
The platforms operating at Levels 1 and 2 — Xero, QuickBooks, Sage and others — are not being dishonest when they describe their features as AI. In many cases the features genuinely involve machine learning, even if basic. What they are doing, intentionally or not, is allowing a broad definition to obscure meaningful differences in capability. The point is not that established platforms have nothing to offer. It is that choosing a platform because it says “AI” on the box, without understanding what level that represents, is a decision made without adequate information.
What to do with this
The takeaway is not “distrust AI claims.” It is “interrogate them” — and before you interrogate the vendor, interrogate your own assumptions. What are you actually trying to achieve? Is AI genuinely the right tool for that goal? If it is, what level of capability does the problem require, and is the platform genuinely operating at that level, or at Level 2 while telling you it is Level 5? These are not difficult questions. They are just questions the market does not currently encourage you to ask. Start asking them.
Daniel Lawrence is the Founder of Bots For That and creator of the Beanieverse platform, a suite of AI-native tools for the accounting and bookkeeping sector. With over a decade of experience deploying enterprise automation and AI in financial services and other highly regulated industries, he writes about AI transformation in accounting from the outside in. Part of the Making Accounting AI thought-leadership series.
