Blog

Close Automation Vendors: Roadmap, Lock-In, Support and Exceptions

Close Automation Vendors: Roadmap, Lock-In, Support and Exceptions

Written by

Naman Mathur

Published on

By the second-round call, every vendor left on your shortlist can match transactions and run a close checklist. The scorecard has done its job. What it cannot tell you is what happens after you sign.

That is what the final round is for. In our experience, deals rarely go wrong on features; they go wrong after signature. This article covers four questions for your final vendor call: the artificial intelligence (AI) roadmap, lock-in, implementation support and exception handling.

Key Takeaways

  • Price only what has run in a customer's close. Sort roadmap items into shipped, in beta or promised, and contract for anything promised.

  • Lock-in depends on where your accounting record lives. If journals and sign-offs sit only in the vendor's database, leaving means migrating your books.

  • Support belongs in the order form, with named accountants, your team's hours, a sandbox on your data and a timeline counted in close cycles.

  • An unsure AI should hand over, with its reasoning attached. A named reviewer gets the item and can reject or undo the suggestion.

Why the Feature Scorecard Stops Helping in the Final Round

The standard advice is to keep scoring vendors against the same feature matrix until one pulls ahead. By the second round, that matrix has stopped separating them: everyone left can match and run a checklist, so the scores converge and the decision drifts to demo polish and price. The questions below test what a scorecard can't, which is life after signature with the vendor you might take to the chief financial officer (CFO).

Controls belong earlier. If you have not yet tested segregation of duties and evidence retention, run the approvals and audit trail checks first. Send the question sets below to each vendor a week before the call, so the answers arrive in writing.

How Should Finance Teams Evaluate a Vendor's AI Roadmap for the Close?

Evaluate an AI roadmap by what has already run in a customer's close, and treat everything else as a promise. In a June 2025 Gartner survey of 183 CFOs, 84% of finance organisations had implemented or planned to implement AI, yet only 7% reported high or very high impact.

Roadmap slides sell the first number, not the second. Many vendors now relabel robotic process automation (RPA) as agents. Our comparison of agentic AI versus RPA sets out the difference to test for on your own data.

Shipped, in Beta or Promised

Put each roadmap item into one state, then ask for matching evidence.

State

Evidence to ask for

How to treat it commercially

Shipped

Dated release notes and customers in production on the feature

Price it, and test it in the pilot on your data

In beta

Which customers run it, and in which close it last ran

Price it only with a dated commitment to general release

Promised

A target date and the dependency it waits on

Give it no value unless it is written into the order form

Source: Stacks implementation practice, 2026.

Ask which month-end each feature last ran in and how many entities it covered. A vendor who cannot name the close is describing a demo.

How Model Changes Are Governed

A change to a model or its matching logic is a change to a control, and repeatability is why your auditor cares. We ran 8 real reconciliation datasets 10 times each through 3 general-purpose AI agents. In 22 of 24 cases, they did not give the same answer twice.

A reconciliation that changes on every run cannot be reviewed, let alone sampled. Ask each vendor:

  • Which calculations are deterministic, returning the same result on every run?

  • How much notice do we get before a model or matching rule changes?

  • Do you regression-test each change on our data before release?

  • Can we pin the current version until after year-end?

  • Where does a model change appear in the audit log our auditor samples?

How Should Finance Teams Assess Vendor Lock-In Risk With Close Automation Platforms?

Assess lock-in by asking where your accounting record and audit evidence will live, then price the exit before you sign. GOV.UK guidance on technical lock-in, written for public-sector cloud buyers, calls lock-in switching that is "difficult, time consuming and disproportionately expensive". It also accepts some lock-in where the value outweighs portability.

The record test is the sharpest. Where journals and sign-offs exist only in the vendor's database, an exit becomes a migration of your books. If your enterprise resource planning (ERP) system stays the record, you lose a workflow tool and keep the ledger.

Data Export and ERP-Side Records

Ask for an export of reconciliations, working papers, sign-offs and the audit log in open formats such as comma-separated values (CSV), with attachments intact. Include the data the tool pulled from subledgers such as Stripe, Adyen, Chargebee, Coupa, Pleo or payroll, since that is often the hardest part to rebuild. Then run that export during the pilot. An export you have opened and tied back to the trial balance is worth more than a line in a security questionnaire.

Map which records already sit in the ERP, since those carry no exit risk. Integration depth raises separate questions, covered in planning the ERP integration.

Contract Terms That Decide the Exit

The Data Act is European Union (EU) law, and it sets a useful floor for software contracts. The Bundesnetzagentur Data Act switching guidance gives the provider a maximum of two months' notice, then a maximum transitional period of 30 calendar days. Your data must then stay retrievable for at least 30 calendar days before erasure.

The three windows run in sequence: notice, then the switch itself, then retrieval. The 30-day transitional period caps the switch, and you can extend it once if you need more time to test. Where 30 days is technically unfeasible, the provider can propose up to seven months and must say so within 14 working days. Switching charges are limited to direct costs until 11 January 2027, and "from 12 January 2027, no switching charges at all may be imposed".

Contract term

Reference point

What to ask for

Notice period

EU Data Act: maximum two months

The same maximum, written into the order form

Data retrieval

EU Data Act: minimum 30 calendar days

A full export, including the audit log, inside that window

Switching charges

EU Data Act: none from 12 January 2027

No switching charges, whichever law governs the contract

Early termination and renewal

Outside the switching-charge ban

Disclosed termination fees, renewal notice dates and a renewal price cap

Source: Bundesnetzagentur Data Act guidance; Stacks implementation practice, 2026.

The Data Act covers any provider serving customers in the EU, wherever the provider is based, so contracts for your EU entities are in scope. For UK-only contracts, ask the vendor to offer equivalent terms by contract, and take legal advice before you sign.

What Support and Implementation Resources Should You Expect From a Close Software Vendor?

Expect named people who have run a close, a stated commitment from your team, a sandbox on your data and a timeline in close cycles. Put all of it in the order form, because sales-stage promises rarely survive the handover.

The mistake we see most often is a project staffed only by solutions engineers, who stall when a reconciliation breaks for an accounting reason. Ask:

  • Who sits with our team during the first close, and are they qualified accountants?

  • How many hours a week does our team commit, and in which weeks?

  • Will the first close run in parallel with our current process?

  • How long does hypercare (extra support straight after go-live) last, and does it cover the first quarter-end?

  • What are your service level agreements (SLAs) for response and resolution on close days?

  • Is training included for reviewers and approvers who join after go-live?

Then ask for a recent reference customer of similar scale. Ask them how many people the vendor actually staffed, and how long signature to first close took.

What Questions Should You Ask About How a Close Platform Handles Exceptions?

Ask who owns each exception, how it escalates, what evidence it leaves and what the AI does when it is unsure. The Financial Reporting Council (FRC) wrote its guidance on generative and agentic AI for audit firms, but its four categories of mitigation make a good checklist for finance teams: system design and development, certification, staff education and governance, and human-in-the-loop review and oversight. Exception handling sits in the last of these.

The EU AI Act requires high-risk AI systems to let a person "disregard, override or reverse the output". Most close tools won't be classed as high-risk, but it is the right design benchmark for exception handling. On the call, ask:

  • Who is the named reviewer for each exception type, by entity and account?

  • Which threshold or age triggers escalation, and what happens to an item still open at close?

  • Does each suggestion show its reasoning, and can the reviewer reject or undo it with the action logged?

  • When the AI is unsure, does it say so and route the item to a person?

In the demo, feed in a bank transfer with no reference that the AI cannot match, and watch where it goes. It should reach a named reviewer with its reasoning attached, instead of posting as a plausible match. That division of work matches our view of agentic close software defined, where agents prepare and named reviewers own exceptions and sign-off.

Why Choose Stacks for Your Close

Stacks connects directly to the systems you already run (see all integrations), so there is nothing to migrate and nothing to replace. Stacks reads from your ERP without changing its configuration, and posts back only the journals your team has approved, with every sync and posting logged. Your ERP stays the accounting record.

Agents prepare the work, and your team reviews exceptions and signs off. Every suggested match carries an explanation, and users accept, reject or undo it inside the close management workspace.

The ERP connection is installed in week one, and standard agents reach production in 4–8 weeks, so within your first one or two month-end closes. Custom agents built with forward-deployed accountants take 2–4 close cycles. Your team gives 1–2 hours in week one, then about 2 hours a week in weeks two and three, in a sandbox with your data.

Epidemic Sound signed on 29 August 2025 and went live four weeks later, across 102 bank accounts and 13 legal entities on NetSuite. The project covered single sign-on (SSO), user acceptance testing (UAT) and training before the team closed on Stacks on 6 October 2025. That first close ran 713 tasks at 100% completion.

Stacks holds System and Organization Controls (SOC) 2 Type II and ISO 27001:2022 (the international information security standard) certification. Stacks enforces segregation of duties, and we comply with the General Data Protection Regulation (GDPR). Put every question above to us too, including support and hypercare terms.

FAQs on Evaluating Close Automation Vendors

Is Some Vendor Lock-In Acceptable?

Yes, when the value outweighs a known cost of leaving. Keep the accounting record in your ERP and lock-in shrinks to the workflow layer.

Does the EU Data Act Apply to UK Contracts?

Only where the service is provided to customers in the EU, such as your EU subsidiaries. For UK-only contracts, ask for equivalent switching terms by contract and take legal advice.

How Long Should Close Automation Implementation Take?

Ask for the timeline in close cycles, with the ERP connection and a sandbox on your data in the first weeks. Stacks puts standard agents into production in 4–8 weeks, inside your first one or two closes.

Should Roadmap Features Be Written Into the Contract?

Yes, if they affect your decision. A promised feature with no date and no remedy in the order form carries no commercial weight.

Who Should Own Exceptions in an Automated Close?

A named reviewer for each exception type, with escalation to the controller by threshold or age. The AI prepares and explains, and the reviewer decides and signs off.

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

"Stacks has transformed how our finance team operates... it's saved us time and reduced frustration."

Graham B., SVP of Finance at Volt

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.

"Stacks has transformed how our finance team operates... it's saved us time and reduced frustration."

Graham B.

SVP of Finance at Volt

Trusted by fast-growing companies including:

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.

"Stacks has transformed how our finance team operates... it's saved us time and reduced frustration."

Graham B.

SVP of Finance at Volt

Trusted by fast-growing companies including:

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.

"Stacks has transformed how our finance team operates... it's saved us time and reduced frustration."

Graham B.

SVP of Finance at Volt

Trusted by fast-growing companies including: