How to evaluate CECL credit loss estimation software

What this software does

Credit loss estimation software calculates the allowance for credit losses on loans, held-to-maturity securities and off-balance-sheet credit exposures under the current expected credit losses (CECL) methodology. It loads period-end portfolio data and loss history, groups exposures into pools, applies one or more estimation methods with your economic forecasts, and records the qualitative adjustments, reviews and approvals that turn a calculation into the figure your institution reports.

Products differ most in how much they do beyond the calculation: reconciliation to the general ledger, sensitivity analysis, movement analysis between periods, disclosure and regulatory report data, and, for groups that also report under IFRS 9, staging and 12-month losses. Decide early whether you need a calculation engine, a governed close process, or both.

The rules that shape the work

CECL is an accounting standard, not a software requirement. FASB’s Accounting Standards Update 2016-13 created Accounting Standards Codification (ASC) Topic 326, which replaced the incurred loss model with an estimate of expected credit losses. The banking agencies’ Interagency Policy Statement on Allowances for Credit Losses, issued by the Federal Reserve, FDIC, NCUA and OCC in 2020 and revised in April 2023, describes how they expect institutions to apply it. Points that bear on software:

• Who and when. As the policy statement notes, Topic 326 took effect for fiscal years beginning after December 15, 2019 for SEC filers other than smaller reporting companies, and after December 15, 2022 for all other entities. Every bank and credit union reporting under US GAAP now applies it.
• What the estimate considers. Expected credit losses are measured over the contractual term, considering past events, current conditions and reasonable and supportable forecasts. Beyond the forecast period, the estimate reverts to historical loss information. Topic 326 prescribes no specific forecasting method, no minimum or maximum forecast period and no specific reversion technique: these are your institution’s judgments.
• Pools and methods. Assets that share similar risk characteristics are evaluated collectively, and segmentation should be reviewed on an ongoing basis. FASB does not specify an estimation method; the policy statement mentions loss-rate methods such as weighted-average remaining maturity (WARM), vintage analysis and the open pool method. Collateral-dependent assets have their own measurement guidance.
• Qualitative adjustments. Adjustments may increase or decrease the estimate, and should cover information not already captured, not repeat what the model already reflects.
• Governance and documentation. Management should have written policies and procedures and should document calculations, qualitative adjustments and changes made in internal review. The board or a committee reviews management’s assessments. Adjusting the allowance only to match a peer median, a target ratio or a budgeted amount is inappropriate.
• Validation. The loss estimation process, and any changes to it, should be validated periodically by someone with the right expertise who is independent of credit approval and of the estimation process, such as internal audit, an independent risk unit or a contracted third party. Third parties involved in the estimate need sound risk management.

The policy statement points to SR 11-7 for models, including vendor models. For the Federal Reserve, FDIC and OCC, that guidance was replaced in April 2026 by SR 26-2, so check which framework your model risk policy now follows.

Who should be on the evaluation team

• The controller or chief accounting officer, who owns the allowance and signs off each period.
• Credit risk analysts who build pools, loss history and forecasts today, often in spreadsheets.
• Allowance accountants, who reconcile data and prepare the committee pack every quarter.
• Model risk or independent review staff, who will validate the methods and need to reproduce results.
• Data owners for the core, loan servicing, collateral and general ledger systems.
• Internal audit, and an early conversation with your external auditors about what evidence they will expect from the new process.

Questions to ask and things to test

Test with your own portfolio, not the vendor’s sample data.

• Load an anonymized extract of one portfolio and its loss history, and compare the estimate with your last approved figure. Ask the vendor to explain every material difference.
• Reconcile a period’s data to the general ledger and see how differences are listed, explained and signed off before any estimate runs.
• Check method choice per pool. Can each pool use a different method, and is the reason for each choice recorded?
• Change the forecast period and reversion setting and look at the effect. Are the settings versioned, with who changed them and when?
• Enter a qualitative adjustment and follow it through review and approval. Is it tied to a policy factor, with the rationale and evidence attached?
• Run an adverse scenario and review how the result would be shown to your committee.
• Rerun a locked period and confirm it reproduces the same figures from the same data and settings.
• Confirm role separation. Who can propose, review and approve adjustments and the allowance? Can one person do all three?

Common limits to probe

• The estimate is not the allowance. Software produces and documents an estimate; the amount in your financial statements is your institution’s judgment, approved by your own controller and committee.
• Forecasts. Many products do not produce economic forecasts. Ask where scenarios come from and how their weighting is recorded.
• Thin loss history. Where your own history is short, peer or proxy data has to be sourced and justified by you.
• Posting and filing. Check whether the product posts to the general ledger or exports a journal entry, and whether it files regulatory reports or only produces the data.
• Scope. Fair value measurement, hedge accounting and available-for-sale securities impairment may be outside it.
• Validation and audit. Using a product does not replace independent validation or your auditor’s testing.

Implementation risks

• Data mapping. Missing risk ratings, dates or collateral values weaken every pool. Reconcile record counts and balances with the source systems.
• Carrying over a spreadsheet’s logic. Rebuilding your current workbook in a new tool can carry over its errors. Document the methodology first, then configure it.
• Parallel periods. Run old and new processes side by side for enough quarters to explain differences before relying on the new figures, and agree with your auditors how the change will be evidenced.
• Changes in method. A change in pools, methods or forecast approach needs the same review and documentation as any other change, and may affect your disclosures.
• Historical records. Keep prior periods’ data, workpapers and approvals reproducible after the old process is retired.
• Ownership after go-live. Name who maintains pools, forecasts and qualitative factors, and who approves changes to them.

Related reading on this site

Requirement pages: Model risk management guidance (SR 26-2).

In the glossary: CECL, SR 26-2, Model risk management, Audit trail.

This guide describes the category, not any one product. It is general information, not legal advice: for what a rule requires of your organization, read the rule itself or ask your counsel.