Ask an accountant for their least-favorite part of the month-end close, and the answer is often accruals. The reason is structural: an accrual records an expense or revenue in the period it actually occurred, even when the invoice or documentation has not arrived yet, which means the team has to figure out what happened without the paperwork that would normally tell them, piecing together data from emails, procurement systems, and conversations with budget owners, estimating amounts in spreadsheets outside the ERP, and manually booking and then reversing each entry, all under deadline. Automation can take most of this off the team's plate, but not uniformly: some accruals automate cleanly, others still need human judgment. Here is how to automate accruals and close faster, honestly, by accrual type.
TL;DR
Accrual accounting records expenses and revenues in the period they occur, not when cash moves, which requires booking accruals for things that have happened but are not yet in the system (an expense incurred but not yet invoiced, revenue earned but not yet billed). Accruals are widely the most painful part of the close because the team has to determine what to accrue and how much without the supporting documentation, gathering data from multiple systems and people, estimating amounts in spreadsheets, and manually booking and reversing entries under deadline.
Automation helps substantially, but accruals fall into types that automate very differently. Recurring, rule-based accruals (rent, utilities, subscriptions, predictable monthly costs) automate almost completely, because the amount is known or follows a rule. Received-but-not-invoiced accruals (goods or services received but not yet billed) automate well, because the amount can be calculated from purchase-order and receipt data. Calculable estimate accruals (project or percentage-of-completion accruals) automate well where the data exists to compute them. Judgmental accruals (genuine estimates requiring professional judgment) are where automation assists but human review stays essential. The honest picture is that AI automates the mechanical and calculable majority while judgment calls remain human, which is where the value and the appropriate boundary both sit.
A specific and often-missed point: true accrual automation handles the full lifecycle, both the accrual entry and its automatic reversal in the next period, because automating the entry but manually posting the reversal reintroduces error and defeats much of the benefit. Effective automation also assembles the underlying data, generates the entries with the right accounts, syncs them to the ERP, and keeps a full audit trail, so accruals are accurate, consistent period to period, and auditable.
Automating accruals is one of the higher-leverage close accelerators (teams report close cycles dropping from 8-10 days to 3-5 with accrual and reconciliation automation), and it fits naturally into the continuous-close model, where recurring and calculable accruals are processed continuously rather than in a period-end scramble. For the operating model, see Continuous Close: How AI Is Ending the Month-End Scramble.
What accrual accounting is, and why accruals are painful
Accrual accounting records revenues and expenses in the period in which they are earned or incurred, regardless of when cash changes hands, which is what distinguishes it from cash accounting and what GAAP requires. The mechanism that makes this work is the accrual: an entry that records something that has economically happened but is not yet reflected in the books because the invoice, bill, or documentation has not arrived. If a company received consulting services in March but the invoice will not arrive until April, an accrual records the expense in March (when it was incurred) so the March financials are accurate, and the entry is reversed in April when the actual invoice is booked, to avoid double-counting.
This is essential for accurate financial statements, without accruals, the period's results would be distorted by the timing of paperwork rather than reflecting what actually happened, but it is also why accruals are the most painful part of the close. The difficulty is inherent in the definition: an accrual is needed precisely when the normal documentation is not yet there, so the team has to determine what happened and how much it was without the invoice that would normally tell them. In practice this means:
Piecing together data from multiple sources: because there is no invoice, the team assembles the information from purchase orders, receipts, contracts, emails, procurement systems, and conversations with the budget owners and departments who know what was received or delivered. Chasing estimates and confirmations: the team follows up with other departments and vendors to confirm what should be accrued and for how much, which is slow email-and-spreadsheet back-and-forth under deadline. Calculating in spreadsheets outside the ERP: accrual amounts are typically worked out in spreadsheets separate from the accounting system, creating versioning errors, limited visibility, and reconciliation headaches. And manually booking and reversing entries: each accrual has to be posted and then remembered and reversed the next period, and forgotten or inconsistent reversals are a common error source.
The result is that accruals consume a disproportionate share of the close, days of tedious assembly, estimation, follow-up, and manual entry, right in the middle of the deadline crunch, which is why they are so often named the worst part of the close and why automating them is high-leverage.
The key insight: accruals are not all the same
The most useful thing to understand about automating accruals is that they fall into types that automate very differently, and treating "accruals" as one undifferentiated task obscures both what automates easily and where human judgment genuinely remains necessary. Sorting accruals by how determinable the amount is clarifies the whole automation question.
Recurring, rule-based accruals. Many accruals are for predictable, recurring items, rent, utilities, subscriptions, software, card and network fees, regular service contracts, where the amount is known or follows a simple rule and recurs every period. These are the easiest to automate almost completely: the accrual can be generated automatically each period from the rule or the known amount, with no estimation and little judgment. This category is pure automation upside.
Received-but-not-invoiced accruals. A large and important category is expenses for goods or services received but not yet invoiced (the classic accrual). The amount here is often calculable from data the company already has, the purchase order and the receipt confirm what was ordered and received, so the accrual can be computed from PO and receipt data rather than estimated from scratch. This automates well, and because it draws on purchasing and receiving data, it is closely tied to the accounts payable process. This is one of the highest-value accruals to automate because it is common, high-volume, and data-determinable.
Calculable estimate accruals. Some accruals require an estimate, but one that can be calculated from available data, project accruals based on percentage-of-completion, usage-based costs, accruals derived from contract terms and milestone status. Where the underlying data exists (project completion, usage, contract terms), these can be calculated systematically rather than guessed, and automation handles the calculation, though the inputs and result usually warrant review.
Judgmental accruals. Finally, some accruals are genuine estimates requiring professional judgment, where the amount is not cleanly determinable from data, legal or warranty provisions, certain complex or uncertain liabilities, accruals depending on judgment about future events. Here automation assists (assembling relevant data, applying prior patterns, drafting a proposed entry) but human judgment and review remain essential, and appropriately so, because these are the accruals where getting it right requires professional expertise, not just data.
The honest through-line is that AI automates the mechanical and calculable accruals, the recurring, the received-but-not-invoiced, the data-calculable, which are the majority of accrual volume, while the genuinely judgmental estimates remain human-reviewed. This is not a limitation to apologize for; it is the correct boundary, and it is also where most of the time savings are, because the mechanical accruals are the bulk of the tedious volume.
How AI automates accruals
Within the accrual types that automate, effective automation does several things, and doing all of them (not just generating the entry) is what delivers the full benefit:
Assemble the underlying data. Automation gathers the data that determines the accrual, purchase orders and receipts for received-but-not-invoiced accruals, contracts and usage for calculable accruals, prior-period patterns for recurring ones, from the source systems, rather than the team assembling it manually from emails and spreadsheets. A common concrete capability is generating the received-but-not-invoiced report automatically from PO and receipt data, which is often the most laborious accrual input. This data assembly is where much of the manual accrual effort actually goes, so automating it is central.
Calculate the accrual amount. For recurring accruals, apply the rule or known amount; for received-but-not-invoiced, compute from PO and receipt data; for calculable estimates, apply the completion or usage calculation. Automation produces the amount from the data, replacing manual spreadsheet calculation, and can use historical patterns to improve or sanity-check the estimate.
Generate the entry with correct coding. Automation creates the journal entry with the right GL accounts, departments, locations, and cost centers, so the entry lands correctly in the ERP rather than being manually keyed, reducing both effort and coding errors. This connects to the broader invoice coding automation pattern in AP.
Handle the full lifecycle, entry and reversal. This is the point most often missed, and it matters. Many teams automate the accrual entry but still manually post the reversal the next period, which reintroduces error (forgotten or inconsistent reversals) and undermines the benefit. True accrual automation handles the full lifecycle: it generates both the accrual and its reversing entry, and the reversal posts automatically at the start of the next period, so there is no double-counting and no forgotten entries. Automating entry-and-reversal as a pair is what makes accrual automation actually reliable period over period.
Sync to the ERP and keep an audit trail. The entries sync to the ERP (the system of record) in real time rather than via delayed spreadsheet exports, and every accrual, its data basis, calculation, and reversal, is documented with an audit trail, so the accruals are auditable and the basis for each is reconstructable. This matters because accruals are estimates that auditors scrutinize, and the audit trail requirements for AI-assisted finance make this a non-negotiable capability.
Route the exceptions and judgmental cases. For accruals that need review (judgmental estimates, unusual items, calculable accruals above a threshold), automation routes them to the right person with the assembled data and a proposed entry, so human attention focuses on the accruals that need judgment rather than on the mechanical majority. This is the exception-based pattern that keeps the human effort on the cases that warrant it.
Done across all these, accrual automation converts the accrual process from days of manual assembly, calculation, entry, and reversal into a largely automated flow with human review focused on the genuine judgment calls, which is a large share of the close-acceleration that AI delivers. Teams report close cycles dropping from 8-10 days to 3-5 days with accrual and reconciliation automation.
Accruals and the continuous close
Accrual automation fits naturally into, and is an important part of, the continuous-close model. In a traditional close, accruals are done in the period-end scramble, all the assembly, estimation, entry, and reversal compressed into the close window. In a continuous close, the recurring and calculable accruals are processed continuously through the period, recurring accruals generated on schedule, received-but-not-invoiced accruals computed as goods and services are received, so that by period-end most accruals are already booked and only the genuinely judgmental or late-arriving ones need attention.
This makes accruals one of the natural early targets for continuous close, because the recurring and received-but-not-invoiced accruals are high-volume, data-determinable, and repetitive, exactly the work that benefits from moving out of the period-end crunch into continuous processing. Automating accruals and making them continuous are complementary: the automation makes each accrual fast and reliable, and the continuous model spreads them across the period so they do not pile up at close. Together they remove one of the largest single contributors to the month-end scramble.
Where agentic AI fits accrual automation
Within accrual automation, an agentic, deterministic platform like Kognitos fits on the data-assembly, calculation, and exception-reasoning work for the determinable accruals, and the honest scope matters. Kognitos is not a close-management or consolidation platform and not the ERP that holds the ledger; it is not a replacement for the accounting system of record.
Where it fits is the reasoning-and-execution work that produces the accruals: assembling the underlying data across systems (notably the received-but-not-invoiced data from purchase orders and receipts, which is an accounts-payable-adjacent data problem and a natural fit for non-PO invoice and AP-adjacent automation), reasoning about what should be accrued and computing the amount from that data, generating the entry and its reversal, and routing the judgmental cases for review, all deterministically and with a full audit trail.
Two things make the deterministic, auditable approach fit accruals well. First, accruals feed the financial statements and are scrutinized by auditors, so the basis for each accrual, the data, the calculation, the reversal, must be reconstructable and explainable; because Kognitos executes deterministically and logs every step in plain language, each accrual's basis is auditable and consistent period to period, and the entry-and-reversal pair is handled reliably rather than depending on someone remembering the reversal. Second, the honest boundary is built in: Kognitos automates the determinable accruals (recurring, received-but-not-invoiced, data-calculable) and routes the genuinely judgmental estimates to human reviewers with the data assembled, rather than purporting to replace professional judgment on the accruals that require it. This is consistent with the broader deterministic AI vs generative AI for finance controls distinction. Kognitos works on top of the ERP and alongside the close tools, producing accurate, auditable, determinable accruals and their reversals continuously, so the accrual portion of the close is largely handled and the team's judgment is focused where it belongs.
Book a working session with a Kognitos solutions engineer → Try Kognitos free →
How to approach accrual automation
For a finance team automating accruals, a practical sequence:
Start with the recurring accruals. The recurring, rule-based accruals (rent, utilities, subscriptions, regular fees) are the easiest to automate and deliver immediate, low-risk benefit, so they are the natural starting point. Automate their generation and reversal first.
Then automate received-but-not-invoiced. The received-but-not-invoiced accruals are usually the highest-volume and most laborious, and they are data-determinable from PO and receipt data, so automating the received-but-not-invoiced report and the resulting accruals is typically the highest-value step, and it ties into AP data you likely already have. This is explored further in Accounts Payable Automation: The 2026 Guide.
Insist on full-lifecycle automation. Ensure the automation handles both the accrual entry and its automatic reversal, not just the entry, because manual reversals reintroduce the error and effort the automation was meant to remove. Full lifecycle is the difference between reliable and half-automated accruals.
Keep judgment human, and route to it cleanly. Do not try to fully automate the genuinely judgmental accruals; instead, have automation assemble the data and propose an entry, and route it to the right reviewer. This keeps the boundary correct and focuses human effort where it adds value.
Sync to the ERP and preserve the audit trail. Ensure accruals sync to the ERP in real time and that each accrual's data basis, calculation, and reversal are documented, since accruals are estimates that must be auditable and reconstructable.
Make it continuous where you can. Process the recurring and calculable accruals continuously through the period rather than at period-end, so they do not contribute to the month-end scramble. The automated financial reporting that follows the close also benefits from accruals being pre-booked rather than last-minute.
The through-line: automate the determinable accruals (recurring, received-but-not-invoiced, calculable) fully, including their reversals, keep the genuinely judgmental accruals human-reviewed with data assembled for them, and sync everything to the ERP with an audit trail, which removes the bulk of the accrual burden from the close while keeping the accruals accurate, consistent, and auditable.
Putting it together
Accruals record expenses and revenues in the period they occur even before the documentation arrives, which is what makes accurate financial statements possible and also what makes accruals the most painful part of the close: the team must determine what happened and how much without the invoice, assembling data from many sources, estimating in spreadsheets, and manually booking and reversing entries under deadline. Automation helps most when it recognizes that accruals are not uniform: recurring rule-based accruals and received-but-not-invoiced accruals (calculable from PO and receipt data) automate cleanly, calculable estimate accruals automate where the data exists, and genuinely judgmental accruals stay human-reviewed. Effective automation assembles the data, calculates the amount, generates the entry with correct coding, handles the full lifecycle including automatic reversal, syncs to the ERP, keeps an audit trail, and routes the judgmental cases for review. Automating accruals is one of the higher-leverage close accelerators and a natural part of the continuous-close model, where recurring and calculable accruals are processed continuously rather than in the period-end crunch. Done with the determinable accruals fully automated (entry and reversal) and judgment kept human, accrual automation removes one of the largest contributors to the month-end scramble while keeping the numbers accurate and auditable.