The Benchmark Is Part of the Evidence: A Reviewer’s Checklist for Treasury Peer Comparisons

The benchmark is not background — it is part of the evidence

Bank of America’s 28 September 2026 launch announcement for Payments Insights is a useful current example. The bank says the CashPro capability gives corporate and commercial clients insights into payment efficiency, cross-border flows and working-capital performance, and that it uses proprietary peer benchmarking and comparisons with industry standards. That confirms the feature and the intended use of peer context. It does not by itself tell a reviewer exactly how a Payments Insights peer group is constructed. In the official public materials reviewed for this article, fields such as sample size, exact inclusion/exclusion rules, refresh cadence, methodology version and a numeric aggregation threshold were not disclosed. That gap should stay visible rather than being filled with inference.

Separately, Bank of America’s general Global Transaction Services terms describe financial benchmarking reports made available to some treasury-services customers. Those terms say the reports may compare a client with anonymous companies sharing similar characteristics and use de-identified, aggregated peer-group data; they also list examples such as NAICS, geography, balances, account counts and payment volumes. That is useful adjacent evidence about Bank of America financial benchmarking generally. It should not be presented as the disclosed methodology for Payments Insights unless Bank of America explicitly links the two.

The broader benchmarking literature points to the same reviewer problem. APQC recommends standardized definitions, relevant peer cuts, normalized measures, percentile distributions and sample size. Official statistical-quality guidance from the UK Office for National Statistics and the U.S. Census Bureau likewise emphasizes comparability, source population, reference dates, methodology, model assumptions, coverage limitations and material changes over time. The details differ by system, but the governance lesson is consistent: a comparison needs metadata about how it was made.

What a peer comparison actually contains

A reviewer should separate seven concepts that are often collapsed into one dashboard label:

APQC’s guidance is especially useful on the last two distinctions. A high percentile is not automatically better: for cost or cycle time, lower can be stronger, while for output or quality, higher may be stronger. And performance benchmarking can show where a gap exists; it does not, by itself, establish why the gap exists.

Why a company can stay the same while the benchmark moves

Suppose a treasury team’s own KPI is unchanged between two review dates. Its relative position can still change if any of the following changes in the reference object:

• peer membership;
• the distribution of peer observations;
• the observation period or seasonality window;
• the normalization denominator;
• source coverage or the mix of contributing data;
• definitions, classifications or inclusion/exclusion rules;
• model assumptions or transformations;
• weights, if the benchmark uses weighting;
• outlier treatment;
• methodology version or restatement.

This is not an argument against benchmarking. It is an argument for versioning it. ONS revision guidance explicitly treats new data, improved methods, changed definitions/classifications and rebasing as reasons published statistical values may change. A treasury reviewer does not need a national-statistics process, but the same basic discipline is useful: show what changed in the reference object and when.

Reviewer failure modes

1. Wrong peer group. Industry may be similar while size, geography, payment mix or operating model is not.
2. Changing denominator. “Per payment”, “per supplier”, “per $1m revenue” and “per active account” can tell different stories.
3. Period mismatch. A trailing-12-month company metric compared with a calendar-quarter peer snapshot is not like-for-like.
4. Stale population. A benchmark can look precise while representing an old peer population.
5. Selection or coverage bias. An opt-in, product-specific or incomplete population may not represent the decision population a reviewer assumes.
6. Outlier sensitivity. A mean, percentile or model can react differently to extreme observations.
7. Observed/modelled mixing. Estimated fields can be useful, but they should not be indistinguishable from observed transactions.
8. Aggregation differences. One benchmark may use legal-entity data while another groups subsidiaries, accounts or regions differently.
9. Methodology drift. A vendor can improve a benchmark in a way that breaks comparability with prior versions.
10. Restatement without lineage. Replacing an old benchmark silently makes it difficult to reconstruct the decision that was made at the time.

Comparison dimensions a reviewer should inspect

Treasury-specific note. Industry, size/revenue and region are well-established benchmarking dimensions in APQC materials. Payment mix, channel, currency/corridor and operating model are Treasury Desk extensions for treasury/payments review; they should be used when material, not treated as a universal external standard.

Treasury Benchmark Evidence Card

The following is a Treasury Desk reviewer aid, not an external standard. It converts the research into a reusable evidence object that can travel with a benchmark screenshot, export or memo.

Reviewer checklist

Before using a peer comparison as evidence in an internal review, ask:

• Can I define the company metric in one sentence, including its denominator?
• Can I define the peer population without saying only “similar companies”?
• Are the entity scope and observation period aligned?
• Is the peer cut appropriate for the business question?
• Do I know the source population and material coverage limitations?
• Are modelled/estimated fields distinguishable from observed data?
• Do I know which statistic is being shown — mean, median, percentile, score or model output?
• Is the benchmark fresh enough for this decision, and can I identify its method/version?
• If the benchmark changed, can I tell whether the company changed, the reference object changed, or both?
• Is the decision-use boundary explicit — including what the benchmark does not prove?

If several answers are “no,” the right response is not necessarily to reject benchmarking. It is to downgrade the comparison from evidence to context pending documentation.

Worked example — synthetic

Assume a treasury team tracks a synthetic KPI: cross-border payment exception rate = 2.4%. The company’s definition, denominator and observation period do not change.

• Review A: peer median = 2.1%. The company sits above the peer median.
• Review B: peer median = 2.6%. The company now sits below the peer median.
• Company KPI: still 2.4%.

Nothing in those three numbers tells the reviewer why the peer median moved. The peer population may have changed, new observations may have entered, the period may have rolled, the method may have been restated, or another documented reference change may have occurred. Until that lineage is known, the correct conclusion is narrow: the company metric did not move; its relative position did.

The numbers above are synthetic and demonstrate mechanics only. They are not Bank of America data, market data or a recommended performance threshold.

What this does NOT prove

A peer benchmark does not, by itself, prove:

• causality;
• safety or treasury suitability;
• management quality;
• future performance;
• legal, tax, accounting or compliance status;
• that a vendor’s methodology is appropriate for every decision;
• that an anonymized or aggregated dataset is legally permissible for every use;
• Treasury Desk customers, adoption, revenue, product-market fit or production readiness.

Those questions require separate evidence and, where relevant, qualified human review.

Practical next use

Use the Treasury Benchmark Evidence Card above alongside the comparison before it enters a material internal review. The card is not an approval certificate. Its job is narrower: help a reviewer decide whether the benchmark documentation is sufficient to enter review, what remains missing, and when the reference object needs to be checked again.

Final caveat

Peer benchmarking is useful precisely because it adds context that a company metric lacks. That extra context is also an extra evidence dependency. The more a decision relies on relative position, the more important it is to preserve the population, metric definition, period, source coverage, methodology, freshness and version history that produced the reference value.

Sources and notes

1. Bank of America — “Bank of America Launches Payments Insights for CashPro Clients” (28 Sep 2026); retrieved 4 Oct 2026.
2. Bank of America — CashPro Data Intelligence; retrieved 4 Oct 2026.
3. Bank of America — Global Transaction Services Terms & Conditions (01 Jun 2026); retrieved 4 Oct 2026. Used only as adjacent evidence about general financial benchmarking, not as the disclosed Payments Insights methodology.
4. APQC — “What Benchmark Values Best Support Comparison With Peers?” (created Sep 2026); retrieved 4 Oct 2026.
5. APQC — Open Standards Benchmarking; retrieved 4 Oct 2026.
6. Office for National Statistics — Quality defined; last updated 8 Oct 2020; retrieved 4 Oct 2026.
7. Office for National Statistics — Revisions Policy and Correction of Errors Policy; last updated 9 Aug 2024; retrieved 4 Oct 2026.
8. U.S. Census Bureau — Statistical Quality Standard D2: Producing Estimates from Models; last revised 16 Dec 2021; retrieved 4 Oct 2026.
9. U.S. Census Bureau — Statistical Quality Standard F2: Documentation for Transparency; last revised 16 Dec 2021; retrieved 4 Oct 2026.
10. Office for National Statistics — Statistical disclosure control; retrieved 4 Oct 2026.

Treasury Desk Brief is educational public-source analysis and decision-support. It is not investment, legal, tax, accounting or compliance advice.