Benchmarked across 2,300 organisations, the median monthly close runs 6.4 calendar days, the top quartile 4.8, and the bottom quartile ten or more. The gap between those groups is rarely effort. It is how much of reconciliation, intercompany elimination and bill posting happens inside the system versus in a spreadsheet beside it.
Ask where the days go and the answer is usually "reconciliation", which is true and not specific enough to act on. Four distinct jobs sit in that week, and they fail for different reasons.
Getting transactions in — supplier bills posted, expenses coded, revenue recognised. Volume work, and the most automatable, because the rules are the same every month.
Agreeing balances — bank reconciliation, subledger-to-GL, intercompany balances between entities. This is where matching logic lives, and matching logic is specific to your entities, your tolerances and your conventions.
The group-level join — consolidation and elimination, turning several correct entity ledgers into one trustworthy set of group numbers.
Explaining the result — variance analysis, flux commentary, the thing the CFO actually reads.
The first and fourth are effort problems: more people, better checklists, earlier cut-offs all move them. The second and third are not. They are decided by what your ERP will let you express, and no amount of process discipline moves a ceiling you do not own.
Because the slowest steps are not throughput-limited. Posting bills and writing commentary scale with effort — two people finish them faster than one. Reconciliation and intercompany elimination do not, because the work is not volume, it is judgement encoded as matching rules: which transactions correspond, at what tolerance, under which entity's conventions. A second person cannot apply a rule the system has no way to hold. Where that logic cannot live in the ERP it lives in a spreadsheet, and a spreadsheet has to be rebuilt, re-checked and re-explained every period regardless of headcount. That is why close duration tends to plateau at a company's structural floor and then stay there through several rounds of process improvement. The APQC benchmark puts the median at 6.4 calendar days and the bottom quartile at ten or more; moving between those bands is usually a question of what the system can express, not how hard the team works.
Source: Perry D. Wiggins, "Metric of the Month: Cycle Time for Monthly Close," CFO.com, 2018, reporting APQC Open Standards Benchmarking across 2,300 organisations. ↗
On a closed-schema ERP you get the record types the vendor exposes, under the vendor's permissions and governance limits, with no way to extend the data model or read the code. That is a legitimate design choice and it buys real things — upgrade safety, a supported surface, predictable behaviour.
What it costs is this: matching logic specific to your entities and tolerances has to be bought as an add-on or written as vendor script by a licensed developer and maintained against the platform's governance limits. Neither is unavailable. Both are a purchase and a dependency rather than a change you make.
That is the whole mechanism behind the plateau, and it is why reconciliation never gets faster no matter what you buy is a structural article rather than a productivity one. This is not a people problem, and it is not a discipline problem.
| Step | What decides how long it takes | Moves with effort? |
|---|---|---|
| Bill and expense posting | Whether posting rules can be expressed and enforced in-system | Partly — until the rules run themselves |
| Bank and subledger reconciliation | Whether matching logic can live in the ERP or must live beside it | No |
| Intercompany balances | Whether both sides of a transaction are visible in one place | No |
| Consolidation and elimination | Whether the system has a group layer at all | No |
| Variance and commentary | People and calendar | Yes |
Two rows on that table are worth sitting with. Consolidation is the only step that can be entirely absent from a system rather than merely awkward in it — on the edition del.ai deploys it is not a partial capability, it is absent, and that has to be said before anything else about it. And bill posting is the one step where the rules genuinely are the same every month, which is why it is the first place automation earns its keep.
Three things, and they are the three that do not respond to effort. First, whether reconciliation matching rules can be expressed inside the system: if the data model cannot hold your tolerances and entity conventions, the matching happens in a spreadsheet and is rebuilt every period. Second, whether both sides of an intercompany transaction are visible in one place, which determines whether agreeing balances between entities is a query or a negotiation between two people with two exports. Third, whether a group-level consolidation layer exists at all — some systems have one, some sell it as a separate edition, and some have nothing matching it, which leaves elimination as a spreadsheet exercise no matter how cleanly each entity closes. Everything else in the close — cut-off discipline, checklist quality, when accruals get posted — is genuinely yours to improve, and improving it is worth doing. It just runs out of room at the floor these three set.
Source: del.ai analysis of ERP close mechanics, modelled from vendor documentation, 2026 ; IFRS Foundation, "IFRS 10 Consolidated Financial Statements," 2026. ↗
An agent is useful here for the same reason a rule is: the work repeats. Bill posting is the clearest case — same fields, same coding logic, same period rules, every month. The behaviour that matters is not the posting, it is the refusal: an agent that will not post into a locked period, and that stops and asks rather than guessing, is the only kind that belongs anywhere near a ledger during close.
What an agent does not do is decide your matching tolerances or your elimination policy. Those are accounting judgements, they belong to the Controller, and a system that lets an agent invent them has a governance problem rather than a productivity feature.
The mechanism behind the plateau: where matching logic has to live, what it costs to put it there, and why the answer does not change with headcount. → Why NetSuite Reconciliation Never Gets Faster
The benchmark bands, which steps sit above the automation ceiling on a closed schema, and what changes when the schema is open. → Cutting Month-End Close from 10 Days to 2
The group-level join explained from first principles — what elimination does, why it survives outside the system, and where it is simply absent. → Intercompany Elimination, Explained
One job, end to end, including the part that matters: the guard that refuses to post into a closed period rather than posting and apologising. → Automating AP Bill Posting in Odoo
The APQC benchmark puts the median at 6.4 calendar days, the top quartile at 4.8 and the bottom quartile at ten or more. Those are cycle-time bands across 2,300 organisations rather than a target for your company; the useful question is which band your system's structural floor allows.
Getting transactions in, agreeing balances, joining the entities at group level, and explaining the result. Most published checklists expand these into twenty or thirty line items, which is useful for running the close and unhelpful for diagnosing it — the diagnosis needs the four, because two of them respond to effort and two do not.
Parts of it, and the parts differ by system rather than by ambition. Rule-shaped, repeating work — posting, coding, period guards — automates on any platform that lets the rules be expressed. Reconciliation matching and elimination automate only where the data model can hold the logic, which is a property of the ERP rather than of the finance team.
It is worth what the delay costs you, which is not a number we will invent for you. The two arguments we would stand behind are that a number arriving on day three supports a decision a number arriving on day eleven cannot, and that the hours are recurring rather than one-off.
Sources
1. Perry D. Wiggins, "Metric of the Month: Cycle Time for Monthly Close," CFO.com, 2018, reporting APQC Open Standards Benchmarking across 2,300 organisations. ↗
2. IFRS Foundation, "IFRS 10 Consolidated Financial Statements," 2026. ↗
3. Odoo S.A., "Year-end closing," Odoo 19.0 Accounting documentation, 2026. ↗
4. del.ai analysis of ERP close mechanics, modelled from vendor documentation, 2026