Key Takeaways

• Legacy financial close solutions track work that people still prepare in Excel, while an agentic close platform prepares it and leaves exceptions for review.

• Established suites now ship agents of their own, so treat them as a third category and test what those agents actually prepare.

• Repeatability and audit evidence beat autonomy: in Stacks' own reliability test, general-purpose agents answered identical runs differently in 22 of 24 cases.

• Pilot on your messiest recent month, run it in parallel, and judge each vendor on its exceptions and its exported audit trail.

Every close vendor now calls its product agentic, so the label tells you nothing when you compare agentic close platforms with legacy financial close solutions. The useful dividing line is who prepares the work, and what evidence that work leaves for your reviewer and your auditor.

Most controllers face a three-way decision. You can replace the tool, add agents to it, or wait for your enterprise resource planning (ERP) vendor to embed artificial intelligence (AI). The standard advice says pick the most autonomous agent, or the incumbent your auditor knows, and both miss what an auditor will actually test.

This article covers how the two categories compare, where suites adding agents fit, whether agents can be trusted, and how long switching takes.

How Do Agentic Close Platforms Compare to Legacy Financial Close Solutions?

Legacy financial close solutions organise and track close work, which your team still prepares in Excel and the ERP. Agentic close platforms prepare that work themselves as data arrives, covering matching, journals, intercompany and flux commentary (the explanations of balance movements). Your team reviews the exceptions and signs off.

Everything else follows from that split, including when the work happens and what the auditor receives. A continuous close, where accounting work is completed as transactions land during the month, is only possible when software does the preparation. For the full category definition, read our controller's definition of agentic financial close software.

Table 1: how agentic close platforms and legacy financial close solutions differ

Source: Stacks implementation practice, 2026.

What Counts as a Legacy Financial Close Solution

Legacy close software comes in three types, each built to fix one part of the close. Most groups run at least one of them today.

• Checklist and task tools that hold the close calendar, assign preparers and reviewers, and record sign-offs.

• Rules-based reconciliation and matching modules, where someone writes and maintains a rule for each account.

• Robotic process automation (RPA) bolted onto the ERP, which repeats fixed keystrokes and file moves on a schedule.

These tools earned their place. They replaced close-by-email with a structured calendar everyone could see, and they gave auditors a record of who signed what and when.

What they leave untouched is the preparation itself. A ticked reconciliation records that someone signed it, and the checklist has no way of knowing whether the balance was right. We cover the wider pattern in the limits of checklist close tools.

What Makes a Close Platform Agentic

KPMG Global AI in Finance 2026 defines agentic AI as systems that "can plan, reason, act and learn autonomously or semi-autonomously to achieve goals".

In a close, the practical test is what the agent produces. A matched reconciliation or a staged journal you can open counts as agentic work. A status flag that turns green is task tracking with a new label.

Gartner calls that relabelling "agent washing", meaning existing products such as AI assistants and RPA rebranded without real agentic capability. In a June 2025 forecast, Gartner estimates only about 130 of the thousands of agentic AI vendors are real. Run the produce-something test in every demo, ours included.

The Third Category: Established Suites Adding Agents

The market is no longer a choice between old and new. The established suites have announced AI agents for their finance platforms, and they deserve a category of their own.

Table 2: what the established suites have added, and what to ask

Source: vendor announcements, 2025–2026.

The case for this route is fair. If you already run one of these suites, agents inside it mean no new vendor review and an auditor who knows the control environment. Use the questions in Table 2 to see how much of the preparation those agents actually do.

Cloud ERP vendors are moving too. In a February 2026 prediction, Gartner forecasts a 30% faster close by 2028 for cloud ERP users with embedded AI assistants. That is a forecast, and a competing position, because it points buyers towards AI built into the ERP itself.

The same release names inconsistent multi-entity support as one of the hindrances to adoption. If your group runs one ERP across a handful of entities, waiting for the embedded assistant is a defensible choice. If it runs several ledgers, ask how that assistant would reach the entities on the other systems.

Where the Difference Shows Up in the Close

In the American Productivity & Quality Center (APQC) close cycle benchmark, the fastest organisations take 5 calendar days, the median 6 and the slowest 10. APQC counts from the first entity trial balance (TB) to agreed consolidated statements.

Four days separate the median from the slowest, and in our experience process explains that gap more often than headcount. APQC itself ties faster cycles to a centralised chart of accounts and integrated cloud tools. The days go on hand-offs, which is exactly where an agentic close platform and a legacy tool behave differently.

Bank and Subledger Reconciliation

Rules-based matching breaks on small changes. A rule keyed to a bank description fails when the bank changes one character, and someone rewrites the rule only after the exceptions pile up.

European groups add format variety on top. Statements arrive as CAMT.053 files under the ISO 20022 standard alongside the older MT940 format. Subledgers sit outside the ERP entirely, in Stripe, Adyen, Coupa and the payroll provider.

Stacks' published reconciliation data shows learned matching reaching 95%+ out of the box. [VERIFY: source and link for the 95%+ out-of-the-box figure] In our experience, hand-maintained rules stall at around 70–75%, and our reconciliation guide shows where rules-based matching stalls. During evaluation, Stacks matched Nivoda's own bank and general ledger (GL) data with 95% accuracy.

For audit, the number that matters is the unmatched remainder. At a 70–75% match rate, a quarter or more of transactions land on a preparer's desk with no explanation attached. Ask every vendor to show who owns each unmatched item and what evidence sits behind it.

Intercompany and Multi-Entity Close

Intercompany pairs grow quadratically with entity count. Five entities give 10 possible pairs, while 13 entities give 78, and each pair is another tie-out to prepare and document.

Foreign exchange (FX) makes it harder. Under International Accounting Standard (IAS) 21, the entity holding the balance in a foreign currency retranslates it at the closing rate, while its counterparty may have booked at a different date or rate, so the two sides rarely tie exactly. In our experience, most residuals still come from timing or a one-sided booking, with FX layered on top.

The legacy approach exports both ledgers and rebuilds the intercompany matrix in Excel each month. An agentic approach matches both sides in the platform and suggests matches with an explanation. It also flags differences that may be caused by exchange rates, as our page on intercompany matching across entities shows.

Journals, Accruals and Flux Commentary

In a legacy close, journals are prepared in Excel, uploaded as comma-separated values (CSV) files and reviewed in rounds. Each round re-checks the same spreadsheet, and each upload is a fresh chance for a keying error.

On an agentic close platform, agents stage journals and accruals against your policy. A posting gate, a control that holds each journal until the required approvals are recorded, stops anything reaching the GL early. Flux commentary arrives as a draft that ties to the TB, for the reviewer to edit and own.

Nivoda moved journal posting from several hours to around 15 minutes. For a controller, the bigger gain is the control, because the gate is testable and every approval sits on the record.

Can You Trust an Agent More Than a Rule?

Only when the matching underneath is deterministic and every action is reviewable. Deterministic means the same data produces the same result on every run, which is what lets an auditor rely on a sample.

We tested the alternative directly. In the Stacks reliability test report, we ran 3 frontier models on 8 reconciliation datasets, 10 times each.

Table 3: results from Stacks' own reliability test

Source: Stacks, Can You Trust an AI to Match Your Transactions?, 2026 [VERIFY: year].

The variance is the worrying part for a controller. A false match can balance to the cent, so the reconciliation looks clean while the pairing underneath is wrong. No checklist catches that, because the difference column shows zero.

The design that holds up uses a deterministic engine for matching, with agents orchestrating the work and handling exceptions on top. In a demo, ask which parts of the product are deterministic and which are generated. Then ask the vendor to rerun the same match twice while you watch.

Audit Evidence Is Now the Deciding Criterion

Provision 29 of the Financial Reporting Council's (FRC) UK Corporate Governance Code applies to financial years beginning on or after 1 January 2026. It asks boards of companies in the commercial companies or closed-ended investment funds listing categories to declare, on a comply-or-explain basis, whether their material controls were effective. That gives controls evidence board-level weight.

Whether Provision 29 applies to your group depends on its listing status. Auditors' expectations are moving in the same direction either way.

On 30 March 2026, the FRC published its FRC guidance on agentic AI, stating that "the human auditor is always accountable". The guidance is written for audit firms using AI on their engagements. Expect auditors to apply the same logic to your close: a named person behind every agent output.

The evidence pays off beyond the audit. In KPMG's survey, organisations that can produce AI audit evidence efficiently were far more likely to report significant improvement in error reduction (33% against 6%).

On a legacy close, evidence is assembled after the close, when the auditor's requests arrive. On an agentic close platform, each action is logged as it runs, with the preparer, reviewer, timestamp and change made. Ask whether segregation of duties (SoD) is enforced by the system or only by convention.

Write that standard into your requirements before you see a demo. For US-listed groups, the same log supports Sarbanes-Oxley (SOX) testing of internal control over financial reporting (ICFR).

Does Moving to an Agentic Platform Mean Replacing Your ERP?

No. An agentic close platform connects to the ERP you already run and leaves its configuration untouched, so there is no ledger migration.

The requirement is transaction-level live sync, plus two-way posting so approved journals land back in the ERP. A nightly batch is fine for low-volume accounts. On a busy cash account, a journal posted after the export means you have reconciled to a ledger that no longer exists.

European groups rarely run a single ledger. A typical group runs NetSuite, SAP or Microsoft Dynamics at the parent, with Xero, Sage 200 or Sage X3, Exact, Visma, Unit4 or Fortnox in local subsidiaries. Test the connection on your two most different ERPs plus one subledger, and use our guide to planning your ERP integration.

How Long Does It Take to Switch From a Legacy Close Tool?

Expect weeks for standard work and several closes for anything built on your own policies. With Stacks, standard agents reach production in 4–8 weeks and custom agents take 2–4 close cycles, because they encode your policies and need live closes to tune against. [VERIFY: 4–8 weeks and 2–4 close cycles]

Epidemic Sound signed on 29 August 2025 and closed on Stacks on 6 October 2025, five weeks from contract to first close. That go-live covered close management, reconciliations and the NetSuite integration, with agentic matching and flux commentary planned as the next phase [VERIFY: current status of Epidemic Sound's agent rollout]. That close covered 102 bank accounts across 13 entities and ran 713 tasks at 100% completion, as the Epidemic Sound case study sets out.

The team already ran a four-day close, so its aim was to hold that pace as the business grew. It now saves about 5 hours per close on approvals.

Run at least one parallel close before you switch off the old tool. Calendar time stretches for business reasons, usually peak trading or audit fieldwork, so plan the go-live around them.

Why the Standard Advice on Replacing Legacy Close Tools Fails

The usual playbook for replacing a close tool sounds prudent at every step, and each step hides the failure you are trying to avoid. Gartner predicts over 40% of agentic AI projects will be cancelled by the end of 2027, citing cost, unclear value and weak risk controls.

The first piece of advice is to pick the most autonomous agent. Autonomy without repeatability is the failure our reliability test exposed. An agent that acts alone and reaches a different answer each run gives your reviewer more to check and your auditor nothing to rely on.

Next comes the demo on clean sample data. Sample data has no short pays, duplicate bank lines, one-sided bookings or intercompany balances booked at different rates. You never see the exception queue, which is where your team will spend its close.

"Fix your data first" sounds sensible and turns a close project into a data project with no end date. In our experience, the platform is what surfaces bad data, because it matches every line and shows you what failed.

Finally, there is the case for the incumbent your auditor knows. It has substance: established suites have scale, and auditors who already know how to test them. That familiarity measures the auditor's comfort with the tool and says little about the manual work left inside it.

The fix is to change what you test. Bring your messiest recent month, and look at the exceptions before the dashboards. A pilot built this way takes one close:

1. Baseline days to close, days to first TB, open reconciliations at day 3, exception volume and match rate.

2. Connect read-only to the ERP and bank feeds, so nothing the platform does can touch the live ledger.

3. Run the close in parallel on the same period and data your team is closing.

4. Force exceptions, such as a short pay and an intercompany balance outside FX tolerance.

5. Export the audit trail and check that a reviewer could follow every step.

6. Compare each measure with your baseline, and decide on that evidence.

When Is It Time to Move Off a Legacy Close Solution?

Move when the cost of staying shows up in the calendar or the audit file. These are the signals we see most often in groups that switch:

• You close at the slow end of the APQC range, nearer 10 calendar days than 5.

• Your entity count is growing faster than your accounting team.

• Reconciliations are rebuilt in Excel each month before they reach the close tool.

• Your auditors need extensions because evidence is assembled after the close.

• You pay for two platforms, one to track the close and another to match transactions.

In our experience, buying AI is no longer the hard part. Trusting its output enough to sign off on it is.

That is why the tests in this article matter more than any feature list. If two or more of the signals above apply, schedule the parallel close before your next audit cycle starts.

Why Choose Stacks for the Agentic Close

Stacks is built around the argument this article makes. Deterministic matching sits underneath, and agents prepare reconciliations, journals, intercompany and flux commentary while your team reviews the exceptions and signs off.

Stacks close management software runs the close on a live calendar with enforced SoD and GL posting gates. At Juni, Ruben A., Head of Finance, says Stacks "reduced the time to financial close by three and a half days".

Stacks connects directly to your ERP, so there is nothing to migrate. Stacks holds System and Organization Controls (SOC) 2 Type II and ISO 27001:2022 certification, complies with the General Data Protection Regulation (GDPR) and provides SOX-ready controls.

Nivoda ran NetSuite across 13 entities and more than 130 bank accounts, and its close took more than 20 days. Within months of starting implementation in September 2025, it closed in 12 days [VERIFY: month of the 12-day close]. Read how Nivoda cut eight days for the detail.

FAQs for Agentic Close Platforms vs Legacy Close Solutions

Short answers to the questions controllers ask most when they compare the two.

Is Agentic AI the Same as RPA?

No. RPA repeats fixed steps and stops when an input changes format. An agentic platform completes the task and hands over an output a reviewer can open.

Do AI Agents Post Journals Without Approval?

On a well-controlled platform, they don't. Agents stage journals, and posting gates hold each one until the approvals your policy requires are recorded.

Are BlackLine and FloQast Agentic Close Platforms?

Both have announced AI agents for their close products, which puts them in the third category described above. Test which agents are live in your edition and how much matching still runs on rules.

Can an Agentic Platform Run Alongside an Existing Close Tool?

Yes. The agentic platform can take the matching and journals while the existing tool keeps the calendar, and you consolidate once parallel closes prove the result.

Is an Agentic Close Platform Suitable for Mid-Market Groups?

Yes, once a group runs several entities or currencies, because that is where intercompany pairs and exception volume grow fastest. A single-entity business on one bank account will see less benefit.

How Much Faster Is an Agentic Close Than a Legacy Close?

It depends on where your days go today, so baseline before you pilot. Nivoda went from more than 20 days to 12 within months of moving to Stacks, and Juni cut three and a half days.

See How Stacks Maps to Your Close Process

Book a 30-minute walkthrough. We'll discuss your close process, show you which tasks run automatically, and outline the next step from there.

GET STARTED

"If you want to achieve an agile, transparent, and significantly faster close without adding headcount, Stacks is the way to go."

Jessica Lindholm, VP of Accounting, Epidemic Sound

Trusted by fast-growing companies including: Epidemic Sound, Nivoda, Cleo, Volt and Motorway