Accounting Software and its (No) Right to Exist in the GenAI Era

A recent UBS research report (global accounting software trends[1]) indicated that SMEs are ready to pay +8.5% more for AI features, down from 11% last year. My initial reaction was “this can’t be right”. My second reaction was “of course, the software vendors shipped too early, not understanding themselves the actual value of AI in accounting systems”. My third reaction was “what right to exist does accounting software even have in the GenAI era? They don’t. I’ll write a blogpost about it”.

The simple answer to the question about accounting systems’ right to exist in the GenAI era should go through the shoebox test – how much time (and cost) would it take a skilled accountant to turn a shoebox full of invoices, receipts and bank statements into a fully reconciled, audit-ready set of financial statements or tax return? Now let’s do the same using Large Language Models (LLMs). The ratio would probably be at least 1:100? And by LLM, for this example I used one of the frontier models with access to a local folder (e.g. Claude CoWork). What would be the ratio with a proper vertical agentic system? 1:500?

Let’s put real numbers on this. A UK SME currently pays £1,200–3,000 per year for basic bookkeeping and accounts preparation, plus £250–500 for tax filing. A US small business pays $3,600–10,000 annually for equivalent services. That’s before any management reporting, cash flow analysis, or strategic advice – services that require a fractional CFO at $175–450 per hour, or $24,000–120,000 per year for meaningful engagement. Most SMEs simply go without.

An LLM-native system collapses this entirely. The marginal cost of processing transactions approaches zero. The bookkeeping layer becomes a feature, not a product. And the strategic insights that were previously reserved for businesses that could afford a finance director? Now available to every SME, continuously, at negligible incremental cost.

If we agree with this example, and that a small capable team could build it in a few months (and another few months to add the bells and whistles), it’s clear that any SME-facing service provider (banks, accountants, CRAs) can offer a free version, that will do at a fraction of the time-cost, what accounting systems do today.

Does it mean that the accounting software market is over? In its current form, pretty much.

In the rest of this blogpost I’ll try to answer the question – what will the next generation look like?

The Three Layers of GenAI Impact

GenAI doesn’t just bolt onto accounting systems. It replaces them at three distinct layers. Understanding these layers explains why “AI-enhanced Xage” isn’t the answer – it’s a category error.

Layer 1: Voice and Image as the new UI

Think about what accounting software actually does at the front end. It provides structured forms for humans to type data into. That’s it. The entire UX paradigm assumes a human sitting at a keyboard, manually entering invoices, coding transactions, filling in fields. Either directly to the General Ledger, (GL) or through deterministic financial processes (e.g. order-to-cash) that create the transactions in the GL and sub-ledgers.

Voice and image make this obsolete. Not incrementally better – obsolete.

As a simple example – photograph a receipt. Say “this was dinner with the Henderson client, charge to the M&A project.” Done. The system captures the image, extracts the data, applies the correct GL coding, tags it with client, project, cost type, business line. No typing. No dropdown menus. No chart of accounts lookup.

For permanent data feeds – bank transactions, supplier invoices via API – voice and image aren’t needed (while can make the APIs redundant as well). But for everything else? The data is captured and streamlined at source by voice and image.

Layer 2: Agentic AI Replaces the Business Logic

Here’s where it gets interesting. The middle layer of accounting software – the bit that identifies, classifies, records, enriches, reconciles, adjusts – that’s just business logic. Rules. Pattern matching. Lookup tables.

An LLM doesn’t need your chart of accounts. It doesn’t need your predefined categories. It can tag every transaction with unlimited dimensions at the point of capture and let you query however you want later.

The process chain that traditional systems hardcode becomes fluid:

• Identification – what is this?
• Classification – where does it go?
• Recording – into the database (do we even need a “General Ledger” as a concept?)
• Enrichment – tag with every dimension that might matter
• Reconciliation – match, detect anomalies, flag exceptions
• Analytics – slice, analyse and present however you want. Whenever you want, using your own natural language.

Every step is LLM-native. The question isn’t “how do we automate this process” – it’s “why does this process exist as a separate step at all?”. And where needed, the old investor, pricing and other modules, can be replaced by a two pages Word document that describes the policy.

A small business running on a shoebox-to-LLM pipeline can now do analysis that previously required an Enterprise Resource Planning (ERP) system. Profitability by client? Cost allocation by business line? Multi-dimensional reporting? etc.

Layer 3: The Digital CFO Runs Continuously

This is the part that moves from “automation” to “intelligence.” When AI is embedded throughout the system with real-time access to all financial data, you get something that looks like a very good, very proactive Financial Director – but one that never sleeps.

What does a good FD do? They don’t wait for month-end. They’re continuously monitoring:

• Cash flow – where are we heading, what needs attention?
• Profitability – which clients/ products/ lines are actually making money?
• Collections – who’s late, who needs a call, what’s the pattern?
• Payments – what’s due, what can we defer, what earns us early payment discount?
• Inventory – what’s moving, what’s stuck, what needs reordering?
• Vendors – whose contract is up for renewal, whose prices are out of line?
• Tax – what’s due, what’s the exposure, what needs filing?
• Compliance – what’s missing, what’s at risk, what needs attention?

A human FD does this periodically, from memory, based on reports they request. A digital FD does it continuously, comprehensively, proactively. It doesn’t just answer questions – it surfaces the questions you should be asking.

This isn’t “AI copilot for your accounting software.” This is AI replacing the function that accounting software was supposed to enable.

The business value here isn’t just cost reduction – it’s capability expansion. A £2m turnover professional services firm has never been able to afford a finance director. Even a fractional CFO at $5,000–10,000 per month is a stretch. So they fly blind between quarterly accountant meetings, discovering cash flow problems after they’ve happened, missing early payment discounts, carrying bad debt too long, never really understanding which clients or projects make money. A digital FD changes this equation entirely. Real-time visibility into cash position. Continuous monitoring of receivables ageing. Proactive alerts on margin erosion. Pattern detection on late payers. This isn’t automation – it’s intelligence that was previously reserved for companies ten times their size.

The following table maps how GenAI transforms key dimensions of an accounting system across the three layers:

What Does This Mean for Specific Processes?

Let me be concrete about where value lands.

Lead-to-Cash

Traditional: Manual order entry → credit check delays → invoice generation → payment matching → collections chase.

GenAI-native: Voice/ image capture validated against CRM → real-time credit scoring → auto-generated invoices with full dimensional coding → intelligent reconciliation with exception handling → proactive, personalised collection outreach triggered by pattern detection.

The human touches exceptions. Everything else flows.

Purchase-to-Pay

Traditional: PO creation → three-way match → approval workflows → payment processing → vendor management.

GenAI-native: Voice instruction generates policy-compliant PO → image capture of invoice auto-matched → exception-only human involvement → cash-flow-optimised payment timing → continuous vendor price benchmarking and contract renewal alerts.

Again – humans handle exceptions and decisions. The system handles everything else.

Payroll and Cost Allocation

Traditional systems require explicit time reporting, manual cost centre coding, predefined allocation rules.

What if the system inferred time allocation from calendar data, location, activity patterns? Auto-allocated costs to projects and business lines based on learned patterns? Flagged anomalies for the user’s review (and clarification by voice), rather than requiring upfront coding?

The expense report is dead. Photograph receipt, add context if needed, done. Categorisation, policy compliance, approval routing, reimbursement – all handled.

Bank Reconciliation

Traditional: Download statement or auto-feed, match transactions, investigate exceptions, post adjustments, close the period.

GenAI-native: Continuous reconciliation. Real-time anomaly detection. Exceptions flagged as they occur, not discovered at month-end. The concept of “reconciling the bank” as a discrete activity starts to feel quaint.

Migration from other system

AI on bad data produces faster wrong answers. The shoebox test works because the shoebox contains source documents. If your data is already corrupted – duplicates, missing relationships, inconsistent formats; this is where GenAI can make the migration easier and much faster and operate as a data-cleansing and enrichment engine.

So Who Wins?

The accounting software vendors? Probably not in their current form. They’re selling a keyboard-era product with AI features bolted on.

The winners are:

1. Service providers who can bundle free accounting – who use accounting data for their actual business and can offer the bookkeeping layer at zero cost
2. Vertical-specific agentic systems – purpose-built for construction, or professional services, or retail, with deep domain understanding
3. The “picks and shovels” plays – whoever provides the underlying LLM infrastructure, the integrations, the compliance layer

The losers are anyone trying to sell “accounting software” as a standalone product to SMEs. The value has moved.

The Bottom Line

SMEs aren’t wrong to pay only 8.5% more for AI features in their current accounting software. They’re correctly intuiting that “AI features in legacy software” isn’t the transformation. The transformation is replacing the software entirely.

Think about what that UBS number really means. SMEs are being offered a slightly better version of a £30 per month bookkeeping tool and asked to pay £32.55. Meanwhile, the GenAI alternative offers to eliminate the £1,500–3,000 they spend annually on bookkeeping and accounts preparation, plus give them capabilities they’d otherwise pay £50,000+ per year to access. The 8.5% premium for AI features isn’t the relevant comparison. The relevant comparison is: accounting software subscription versus a digital FD that costs a fraction and delivers multiples more value.

The next generation of accounting doesn’t look like Xage with ChatGPT. It looks like an agentic system that makes the concept of “accounting software” unnecessary.

The shoebox test proves it. A frontier LLM with folder access already beats traditional workflows by 100x. A purpose-built vertical system? 500x.

At those ratios, “accounting software” as a product category is living on borrowed time.

[1] UBS Evidence Lab, 9^th SME Survey – pricing stable, SMEs willing to pay for AI