Practical AI

This narrative documents the third stage of the AI editorial progression developed at Sage — the implementation phase, roughly 2024. It follows Agentic AI (the vision argument) and precedes Trusted AI. The sequence: Generative AI → Agentic AI → Practical AI → Trusted AI.

The problem

Once the generative AI conversation had established that the technology was real, capable and faster than expected, the market question shifted.

Professionals were already experimenting — personally, informally, often ahead of their firm’s official position. The hype-versus-reality debate was largely settled. What remained was a harder, more operational question: how do businesses actually use this today?

Not in theory. Not in a pilot. In actual workflows, with actual data, producing actual outputs that clients and auditors would rely on.

For accounting and finance, this question had a specific shape. The stakes were high. Hallucination in a tax return was not a minor inconvenience. Incorrect reconciliation was not something you could A/B test your way out of. Domain specificity was not optional.

The editorial challenge in this phase was to move the conversation from possibility to implementation — with enough rigour that practitioners could trust what they were being told.

The narrative thesis

Practical AI required a specificity argument. Not AI in general — AI for this workflow, this profession, this risk profile.

The core argument had three parts:

Domain-specific models are a different product. A finance-trained model built on accounting standards, compliance requirements and professional practice is not the same as a general-purpose large language model applied to finance. The training data, the accuracy requirements and the tolerance for error are fundamentally different. “A product providing no response is better than one providing an incorrect response” — the standard that emerged from Sage’s AI research team — would not occur to a general-purpose tool.

Integration matters more than capability. The organisations that made AI work in this phase were not those with the most sophisticated tools. They were those that embedded AI into existing workflows rather than creating parallel systems. The crawl-walk-run methodology from the generative phase became practical advice: start with one process, get it right, then expand.

The client relationship is the measure. For accounting practices, the meaningful test of practical AI was not hours saved on a task — it was what those hours enabled. The pattern that emerged: time recovered from compliance and data processing was reinvested in client advisory. AI was not replacing the accountant. It was returning the accountant to the work that required human expertise.

The content that came out of it

The practical AI phase produced more targeted, evidence-led content — less explainer, more implementation guidance and proof.

The Sage Copilot launch needed editorial work that made a clear case for a domain-specific tool in a market where generative AI had become associated with general-purpose chatbots. The argument was differentiation by specificity: built for accounting, trained on financial data, designed for professional use.

Accountex became a critical signal capture moment. With 25,000 UK accountants and bookkeepers using Sage Copilot, and Making Tax Digital creating structural pressure to move from annual to continuous workflows, the practical AI story had real evidence. “This might be the last year you charge annually for a tax return” — the observation made at Accountex — was not a technology claim. It was an editorial diagnosis of a profession in structural transition.

The capacity gap argument developed here: AI was creating time savings, but organisations were not reliably converting those savings into strategic value. The question shifted from does AI save time? to what do you do with the time it saves? That question set up the maturity curve that followed.

Signals from this phase

The 25,000 number meant something Adoption at scale among UK accounting and bookkeeping professionals was not theoretical. It was evidence that domain-specific AI was working in practice — and that the profession’s concern about hallucination and trust could be addressed with the right tool.

MTD was the structural catalyst Making Tax Digital for Income Tax — and the shift from annual to continuous compliance it required — was not primarily an AI story. But it created the conditions in which AI was no longer optional. Practices that had been managing annual tax cycles were suddenly managing continuous data flows. AI was not a nice-to-have in that context. It was an operational requirement.

The capacity gap was the real finding The pattern that kept emerging: AI created time savings, but those savings were absorbed by existing workload rather than reinvested in higher-value work. The gap between time saved and value captured became the central strategic problem — and the foundation of the AI Maturity Curve.

The client relationship remained the differentiator Firms that used AI to accelerate compliance work and then invested the recovered time in client advisory reported stronger relationships and clearer differentiation. Firms that used AI only to do more of the same work at lower cost found the gains were temporary. The editorial argument that emerged: AI amplifies your strategy, not just your efficiency.

What this phase set up

The practical AI narrative did its job: it moved the conversation from capability to implementation, gave professionals something actionable, and produced evidence that adoption was working in the right conditions.

But as adoption deepened, a harder question emerged. Not how do we use AI? but should we trust it with this decision? Not can AI help with compliance? but how do we govern AI in a regulated, high-accountability environment?

That question — about trust, governance and what AI should and should not be permitted to do — became the Trusted AI phase.

See Trusted AI for how the narrative evolved from here — governance and confidence becoming the central editorial argument.

Why this narrative matters

The practical AI narrative matters because it documented what AI adoption actually looked like at scale — not in theory, not in pilots, but in 25,000 accounting and bookkeeping professionals changing how they work.

The finding that mattered most was not about efficiency. It was about what happened to the efficiency gains. Firms that reinvested recovered time in client advisory reported stronger relationships and clearer differentiation. Firms that used AI only to process more of the same work at lower cost found the gains were temporary. AI amplifies strategy. It does not create one.

That distinction — between AI as an efficiency tool and AI as a strategic enabler — is the most important thing the practical AI phase established. It explains why the capacity gap emerged as the central problem: organisations were generating time savings without having a clear answer to the question of what those savings were for.

For content professionals and editorial strategists, the parallel is direct. AI accelerates production. The organisations that benefit most from that acceleration will be those that already know what they are trying to build. The argument underneath the content matters more as the volume of content increases. That is the editorial intelligence argument — and practical AI is where it became undeniable.

Articles from this phase

The following pieces are by Chris Downing, Director of Product Management for Accountants and Bookkeepers at Sage — evidence that the practical AI narrative transferred into product leadership messaging.