One client, four systems: Why a single source of truth decides whether AI works

Mark McNee By Mark McNee | November 11, 2026
One client, four systems: Why a single source of truth decides whether AI works

AI in accounting firms depends on consistent, traceable client data rather than the technology alone. When client information is fragmented across practice management, accounts production, tax and billing systems, AI can generate plausible but unreliable answers. Establishing a single authoritative client record helps improve data accuracy, accountability and confidence in AI-generated outputs.

What you'll learn:

  • Most accounting firms hold multiple versions of client records across different systems, often with inconsistent names, details and identifiers.
  • Years of system migrations and disconnected workflows have increased data fragmentation, making accurate cross-system reporting difficult.
  • AI tools can produce misleading answers when they access conflicting client records, even when the underlying information comes from legitimate sources.
  • A single source of truth establishes one authoritative client record that other systems reference, improving consistency and traceability.
  • Outdated client records and excessive data retention create additional risks when AI tools can access historical information.
  • Data quality requires ongoing management rather than a one-off clean-up project, with clear ownership and consistent processes.
  • APIs and data lakes can help firms establish a single source of truth without replacing every existing system.

A single source of truth is one place that answers who a client is, with every other system referencing it. In most UK accounting firms the client record lives in at least four places at once, across practice management, accounts production, tax and billing, and none of those versions agrees with the others. An AI tool answers from whichever version it reaches first. That is why the record, more than the tool, decides whether AI works.

Where does the client record actually live in an accounting firm?

Ask a partner where the client record lives and you get a list rather than an answer. Four systems usually hold a version of it, each built for its own job, and each storing the client under a different name. Ask two of them the same question about a client and you get two different answers. That is the two-answer problem, and it is older than any AI tool in the firm.

System What it holds Whose name the client is under
Practice management The working relationship: contacts, jobs, deadlines The name the firm calls them
Accounts production The statutory numbers The name the accounts are filed under
Tax Whatever a return needed Whatever the return needed
Billing Whatever was required to raise an invoice The entity that pays

Every entry was made by somebody doing their job properly. Two decades of sensible decisions produced this, one system at a time. John Toon, Head of Technology at HLB International, was asked whether any of the 350 or so member firms he sees holds a single authoritative client record.

Every system has its own version. If there is commonality, it's probably in a client reference number ... people are trying to keep that as the link between systems.
John Toon, Head of Technology, HLB International

It shows up mundanely. A client sits in practice management under an abbreviation somebody typed years ago, and in accounts production under the legal name the accounts are filed with. Both are correct. Nothing in either system knows they are the same company.

Why did changing systems make the client record worse?

Each system migration added to the estate instead of replacing it. Firms that changed systems four or five years ago often still hold the old records on the old platform. The new ones are structured differently, so the number of places a client can live went up with every upgrade meant to simplify things.

Evan Jones, who is specifying the data architecture for a private equity backed group of accountancy businesses, arrived at one business that could not say how many clients it had. One client might exist in four places under four descriptions. His explanation is the most honest account of how it happens.

People will only keep in those systems the information that they need to keep in those specific systems. And that leads you to a completely fragmented dataset.
Evan Jones

Nobody chose fragmentation. It is the sum of reasonable local choices, which is why it rarely appears on anyone's risk register. Each team can find what it needs. The problem only becomes visible when someone asks a question that crosses systems, such as how profitable a client group is, or how many clients the firm actually has. Answering that cleanly needs reporting built on a shared data model, and most estates were never built that way.

That cross-system question is precisely what AI tools are sold to answer.

What happens when AI answers from whichever version it reached?

An AI tool pointed at a fragmented estate answers from whichever version of the client record it reached. The answer is not fabricated. It is assembled from real figures in a real system, which is what makes it plausible, and afterwards nobody can say which version of the record it came from.

Firms lived with fragmentation because people absorbed it. A partner who knows the client base reads two lines on a report and knows one is not comparable with the other. Nobody ever wrote that check down, and the hours it absorbs are part of the real cost of manual processes that rarely gets measured.

The AI is not clever enough to see when something's not right. If you've been doing this for 30 years, you look at stuff and you know whether it's right or not.
Evan Jones

What has changed is that nobody experienced is standing between the data and the client. An output you cannot trace is one you cannot defend to a client or an insurer.

The sector knows this. Cambridge's Centre for Alternative Finance surveyed 628 organisations across financial services worldwide in April 2026 and found data availability and quality to be the leading obstacle to AI adoption. It was named by 66% of AI vendors, 46% of regulators and 40% of industry respondents. The people selling the tools name it most often.

Fixing it starts with one place that answers who the client is. Acting Office is built on a single client record across the whole firm, so the engagement, the accounts, the tax return and the bill all read the same entry, and an AI answer can be traced back to it.

How does data retention turn into AI exposure?

Retention puts every file a firm has kept in scope the day an AI tool is pointed at the estate. HMRC's six-year rule and the money laundering requirements push firms towards keeping everything, while the Information Commissioner's storage limitation principle says data held longer than needed is unlikely to have a lawful basis.

Most firms settle that squeeze by keeping everything. Toon finds client data held for ten and twenty years with no live reason for it. Under the ICO's storage limitation principle, that is already a question. Add an AI tool and the twenty-year-old file sits in scope alongside this month's.

So a retention decision nobody has revisited since the last money laundering review is now an AI exposure decision. It deserves a named owner and a date in the diary, alongside the other steps in getting accounting data ready for AI. The tool will not distinguish between the client you lost in 2009 and the one you onboarded last week.

Why does a single source of truth hold when a clean-up won't?

A clean-up project cannot fix fragmentation, while a single source of truth can, because data quality is a permanent condition rather than a task with an end date. The change that holds is structural: fewer systems holding a version of the client, and one of them answerable for it, so the record has fewer places to disagree with itself.

No organisation's data is perfect. And it probably never will be ... A data culture needs to be embedded in every organisation.
Alistair Brisbourne, Head of Technology Research, ACCA

Brisbourne's view means remediation has no finishing line. A single source of truth is also narrower than it sounds. Toon calls a shared data model across every system a bit of a pipe dream, and he is right, because that means every system agreeing on structure and definitions. One authoritative client record, with everything else referencing it, is achievable. Most firms get there through APIs or a data lake, which is the single source of truth doing its job without a decade of migrations.

The number of places your client record is allowed to disagree with itself sets what AI can safely do in your firm, more than any tool choice. Before you deploy AI, the guide Acting Office published with contributors from ACCA, HLB International, Apteus Group and a private equity backed accountancy group, takes the single source of truth question further, through to what an AI-ready estate looks like. Download the full guide to see the six attributes of an architected firm and the fourteen questions that tell you which ones you already have.

FAQ

What is a single source of truth in an accounting firm?

A single source of truth is one authoritative client record that answers who a client is, with every other system referencing it instead of keeping its own version. It is narrower than a shared data model, which would need every system to agree on structure and definitions. Most firms reach it through APIs or a data lake.

Why is client data quality a problem for AI in accounting?

AI tools assume a complete, consistent client record. Most firms hold four or more versions under different names, so the tool answers from whichever one it reaches. The result is plausible because it uses real figures, but nobody can say which version it came from, which makes it hard to defend.

How long should an accounting firm keep client data?

HMRC's six-year rule and money laundering requirements set minimums, while the ICO's storage limitation principle says data kept longer than needed is unlikely to have a lawful basis. Many firms keep everything for ten or twenty years. Any AI tool connected to that estate can reach all of it. 

Can an accounting firm create a single source of truth without replacing its existing systems?

Yes. A single source of truth does not require every system to use the same data structure or an entirely new software estate. Firms can connect existing platforms through APIs or a shared data layer, with one authoritative client record referenced across them. The important change is establishing which record each system should trust, rather than allowing every application to maintain a separate version of the client.

What should an accounting firm check before connecting AI to its client data?

Start by checking whether client records can be matched reliably across practice management, accounts production, tax and billing. Look for duplicate entries, inconsistent client names and records that cannot be traced to an authoritative source. Firms should also review what historical data an AI tool could access. If the underlying records cannot be reconciled, the AI's answers may be difficult to verify even when the figures come from genuine sources.

Who should be responsible for maintaining a single source of truth?

Responsibility needs to be assigned rather than left to individual teams maintaining their own records. A named owner should oversee how client information is updated, how inconsistencies are resolved and which system holds the authoritative record. Data retention also needs regular review. Without ongoing ownership, even a successful clean-up can gradually return to the fragmentation it was intended to fix.

Note: Please note that the content of the above blog and the aforementioned information are solely for the purpose of awareness and are informative in nature. The content is designed with intent to ease the understanding while preserving the essence and importance of the compliance rules and shall not be considered as an ultimate replication of the rules. ActingOffice does not own any responsibility whatsoever for any unpleasant event that may arise due to the misinterpretation of a specific part or whole of the information.
View more latest Blog Post