We have a standing practice with prospective Spendaq customers: before we run a categorization demo on their live transaction data, we ask to see their chart of accounts. The CoA tells us more about what the onboarding experience is going to look like than any other single document. And a meaningful fraction of the time, what we see in that CoA is the real reason their close is taking two weeks — not the transaction volume, not the ERP, not the staffing ratio.
A chart of accounts that has accumulated six years of organic growth without deliberate pruning is a categorization liability. Every account code that overlaps in meaning with another creates ambiguity that humans and automated systems both struggle to resolve consistently. Every inactive account that still accepts postings creates a cleanup problem at year-end. Every account created for a one-time transaction that never got retired adds noise to every subsequent close cycle.
Automating categorization on top of a CoA in this state doesn't fix the problem. It accelerates the accumulation of miscoded transactions that reflect the structural ambiguity, and it adds the additional challenge of retraining the model every time someone reclassifies something that was ambiguously coded in the first place.
Signs Your Chart of Accounts Needs Work
Several patterns in the CoA reliably predict categorization problems downstream. The first is functional overlap between account codes. If you have both 6210 — Software Subscriptions and 6215 — SaaS Tools as separate accounts, and there's no documented rule for which one a given transaction goes to, you have an overlap problem. In practice, both accounts will get used inconsistently, often depending on who processed the transaction and what they called the product in question.
The second sign is account proliferation in the 6000s expense range. A company with 180 expense account codes in a chart of accounts that was originally designed for 80 has almost certainly created situational accounts over time — 6347 — COVID-related Travel Refunds, 6389 — Offsite Venue Q3 2022 — that were never retired after the situation ended. These dead accounts don't post transactions anymore, but they create length and clutter that slows down manual coding and confuses automated systems looking for pattern matches.
Third: inconsistent account naming conventions. When some accounts follow a noun-first convention ("Software, Cloud Infrastructure"), others use a verb-noun pattern ("Purchase — Computer Hardware"), and others are free-form descriptions ("Various Marketing Services"), the taxonomy lacks a coherent grammar. Human coders approximate, and the approximations diverge.
The Standardization Process: What We Actually Do
A CoA standardization project doesn't have to be a six-month accounting overhaul. For a mid-size company — 50 to 250 employees, operating in a single reporting entity — a focused cleanup typically takes two to four weeks of dedicated effort. The key steps follow a clear sequence.
Step 1: Export and audit. Pull a full account list from your ERP with transaction counts and total posting volume for the trailing twelve months. Any account with zero postings in the last twelve months is a candidate for inactivation. Any account with very low transaction counts (under 5 postings in twelve months) is a candidate for either merger with a parent account or formal retirement. This audit alone often identifies 15% to 30% of active accounts as candidates for consolidation or removal.
Step 2: Map overlapping accounts. For the remaining active accounts, identify any pairs or clusters where the account names or descriptions could plausibly describe the same type of transaction. These are your overlap candidates. Don't rely on names alone — pull the actual transactions from each account and compare what's in them. Often, what looks like a logical distinction in the account name (6200 — Computer Equipment vs. 6205 — IT Hardware) is functionally identical in practice, with transactions distributed arbitrarily between the two.
Step 3: Establish and document the coding rules. For every account that survives the overlap review, write a one-sentence coding rule. This rule should be specific enough that two different AP staff members would arrive at the same coding decision for any given transaction. "SaaS subscriptions billed on a recurring monthly or annual basis, where the primary deliverable is software access rather than professional services" is a workable rule. "Software costs" is not.
Step 4: Inactivate, don't delete. In most ERPs, you can mark accounts as inactive to prevent new postings while preserving historical transaction data. For year-end and audit purposes, you want the history to remain. Inactivate aggressively — you can always reactivate an account if needed. Deletion is permanent and eliminates history.
Step 5: Validate with your auditor before going live. Before the cleaned CoA takes effect, a quick review with your external auditor or CPA is worth the hour. They may have opinions on account granularity driven by audit evidence requirements. They may flag accounts that seem redundant to you but serve a specific financial statement line item disclosure purpose. Better to know before you merge two accounts than to have to reopen them during fieldwork.
What This Does for Automated Categorization
When Spendaq trains on a company's chart of accounts, the quality of that training is a direct function of CoA clarity. Each account code needs enough historical transaction examples with consistent meaning for the model to learn what belongs there. An account that has been used for three different types of transactions — because the coding rules were ambiguous — trains the model on an ambiguous target. The output reflects the input.
A clean CoA with 80 well-defined accounts will almost always produce higher initial categorization accuracy than a 200-account CoA with overlapping scopes and inconsistent historical usage. Not because of any technical sophistication in the training process, but simply because the target is clearer. We typically see 8 to 12 percentage point accuracy differences in the first four weeks on accounts with clean CoAs versus messy ones.
This matters for close cycle time. An 88% initial accuracy rate means 12% of transactions need human review. A 96% initial rate means 4% need review. On a thousand transactions per month, that's the difference between 120 review items and 40. The human review queue is the variable that determines how long the close actually takes — and it's directly controlled by how well the automated categorization performs at the front end.
One Counterintuitive Point About Granularity
Finance teams sometimes assume that more account codes equals better visibility — that a finer-grained CoA gives you more precise reporting. This is true up to a point, and then it becomes false. Past a certain level of granularity, every additional account code you add requires a corresponding investment in coding accuracy. If you can't consistently code to that level of detail, the additional account codes generate noise rather than signal.
We're not saying a flat, high-level CoA is better than a detailed one. For companies that need granular P&L reporting — where the difference between 6400 — Marketing: Digital Advertising and 6405 — Marketing: Content Production is material to how the CMO manages her budget — that granularity is worth maintaining with the accompanying coding discipline. The point is that granularity without coding discipline produces worse reporting than less granularity with consistent rules.
The Consolidation vs. Standardization Question
For companies that have grown through acquisition or merger, the CoA challenge is more complicated. Multiple legacy entities often come with multiple legacy charts of accounts, and the post-merger question is whether to standardize to a single CoA or maintain separate CoAs per entity with a mapping layer for consolidated reporting.
The mapping layer approach sounds attractive because it preserves each entity's historical account structure and avoids a disruptive migration. In practice, it adds permanent maintenance overhead and makes cross-entity spend analysis harder. Unless there are strong regulatory or contractual reasons to maintain separate CoAs, standardizing to a single CoA — even if it requires mapping and migrating legacy history — pays back in reduced complexity over the following three to five years.
The time to make that decision is before any new automation is layered on top. A categorization system trained on two different CoAs that have been loosely mapped together will perform worse than one trained on a clean, unified structure. If consolidation is on the roadmap, it should precede the automation investment.
The Version Control Problem
Once your CoA is clean, keeping it clean requires treating it with the same discipline as any other controlled artifact in your financial systems. That means version control: a record of every account added, inactivated, merged, or renamed, with the date and the reason for the change.
Most ERPs maintain an audit trail of account changes, but it's often buried in a system log that nobody looks at. The more useful discipline is maintaining a human-readable CoA changelog — a simple spreadsheet that documents what changed, when, and why. When a Controller joins six months later and tries to understand why two similar accounts exist, the changelog answers the question immediately rather than requiring a forensic investigation through historical transactions.
That changelog also serves as a communication artifact during audits. External auditors frequently ask about account reclassifications and CoA changes. Being able to hand them a timestamped record of intentional decisions is significantly better than trying to reconstruct the reasoning after the fact.
The close is only as accurate as the accounting structure underneath it. A CoA cleanup is, in accounting terms, unglamorous work. But it's the kind of unglamorous work that eliminates entire categories of close-week friction and makes everything downstream — budgeting, reporting, automated categorization, audit prep — materially faster and more reliable. The two weeks you spend cleaning it up front come back to you every month for the next several years.