Perspectives · Record to report

The close, reimagined.

How the accounting close runs today at most companies, where it breaks, and how we redesign it end to end with AI.

ValX Intelligence · September 2026 Written for CFOs, controllers and transformation leaders

The monthly close, across more than 10,000 organizations

≤5 days

Top-quartile companies complete the monthly close in five days or fewer.

6 days

The median company. For the annual close the figure is eighteen days.

≥10 days

The bottom quartile takes ten days or more. Progress on this spread has stalled since 2019.

A two-to-one gap between the top and the bottom of the same process, in the same industries, on largely the same software, is not a technology gap. It is a design gap.

Source: APQC Open Standards Benchmarking, cycle time to complete the monthly close. Top performers redesigned the work; most others automated pieces of it — ValX view.

In brief

Five things a transformation leader should know about the close.

  1. It is the most standardized process in the back office, and one of the slowest to change.

    Every company runs the same nine sub-processes. The median company closes the month in six days and the year in eighteen, and the share of companies closing within six business days has not moved since 2019.

  2. Most of the effort is not accounting judgment. It is gathering, matching and checking.

    Finance functions spend about half their time collecting data rather than analyzing it. Reconciliations, accruals and data hygiene top every ranking of close bottlenecks, and a third of accountants report making several errors a week under capacity pressure.

  3. The cost shows up as control risk and people, not just days.

    The financial close process and accounting personnel shortages each account for roughly a fifth of material-weakness root causes. 73% of controllership staff say new regulation has added to their workload, while the pipeline of new accountants shrank.

  4. AI changes the shape of the close, not just its speed.

    The redesign moves work into the period, reviews exceptions instead of populations, orchestrates approvals, and drafts the explanations people now write from scratch. The step count barely moves. What falls is the number of steps a person performs, and the hand-offs between them.

  5. The failure mode is automating the current process.

    Adoption is broad — 59% of finance functions use AI. Value is narrow: 21% of finance leaders report clear, measurable value, and Gartner expects more than 40% of agentic AI projects to be canceled by 2027. Decision rights, controls and roles have to be redesigned first.

What actually compresses

Days hide the mechanism. Count the hands instead.

When the steps cannot be removed, the honest measure is how many steps a person performs and how many hand-offs sit between them. Across the whole close, for a synthetic reference company anchored to the median benchmark:

Steps in the process 9192 Unchanged. The work does not disappear.
Steps a person performs 8438  −55% Most of it stops being somebody's queue.
Hand-offs between steps 3512  −66% The waiting, not the doing, is where the days sit.
Touch hours in close week, all roles ~620 h~230 h Reported as capacity, not as headcount.

The close does not get shorter because things stop happening. It gets shorter because most of them stop being a person’s queue.

The days come in roughly equal thirds: parallelism in intercompany and consolidation, work moving out of close week into the month, and judgment moving earlier. Automating single tasks is a means to those three, not the source of the days.

How to read the evidence

Every number in this work carries its provenance.

Testimony proves a statement was made, not that the statement is true. So we mark where each claim comes from, and we say which ones are ours.

Benchmark

Independent benchmark, regulator, Big 4 or analyst research.

Academic

Peer-reviewed or working paper.

Vendor

Survey or case published by a software vendor. Directionally useful, not independently verified.

ValX view

Our judgment and insight, based on our experience. Not evidence, and marked so you can discount it.

Reference

A figure for a synthetic reference company. Not a claim about any client, and replaced by measurement in the deep dive.

A worked example — one row of the value case, with what would prove it wrong
Lever

Close-week touch hours

Baseline → target

~620 h per close → ~230 h

Mechanism

Steps a person performs fall 84 → 38; pools in accruals and reconciliations.

What would falsify it

The external auditor rejects sampled review of below-threshold items as a control. The target becomes ~330 h — still positive, materially smaller.

We test that assumption in week two, before the pilot is designed around it, rather than after.

The full document

The full report contains additional info including

  1. 08–11The redesign, stage by stage: cutoff, accruals, reconciliations, intercompany, consolidation, flux, external reporting
  2. 10Decision rights and autonomy by step, with the gauges that show they hold
  3. 12–13Why most close automation disappoints, and what to do differently
  4. 14Eleven capabilities the redesigned close needs, and seven market categories set against them
  5. 15–16Candidate pilot boundaries, a seven-week plan, and the value case lever by lever
  6. 17–18The reference company, the capability catalog, and every source with its date

Get the full report

Email us and we will send it over. We will not add you to a mailing list.

Email us for the report