In voice processes, quality is mostly about how a conversation was handled. In back-office work, such as insurance claims processing, invoice entry, KYC document checks or order management, quality is mostly about whether the output is correct. The metric is accuracy, and it is usually written into the contract as a number like 99% or 99.5%. What that number means depends on how errors are counted, and very few people outside the quality team look closely at that.

A month of audit data

An invoice-processing team completes 2,400 invoices in a month. Each invoice has 25 fields that the associate captures or validates: vendor code, invoice number, date, tax amounts, line totals, cost centre and so on. The quality team audits a random sample of 200 invoices. They find 9 invoices with at least one error, and 12 field-level errors in total across those 9 invoices.

There are two very different ways to report this.

Transaction accuracy counts an invoice as wrong if any field is wrong: (200 − 9) ÷ 200 = 95.5%.

Field-level accuracy (sometimes called opportunity accuracy) treats every field as a chance to make an error: 200 invoices × 25 fields = 5,000 opportunities, with 12 errors. Accuracy = 1 − 12 ÷ 5,000 = 99.76%.

The gap between 95.5% and 99.76% is not a rounding difference. Against a 99% contractual target, one method says the team is failing badly; the other says it is comfortably ahead. Both are calculated correctly from the same audit.

Which one is right?

It depends on what the client experiences. If one wrong field in an invoice means the vendor gets paid the wrong amount or the invoice is rejected by the ERP, then the invoice is the unit of failure and transaction accuracy reflects reality. Field-level accuracy would flatter the team, since an invoice with 24 correct fields and a wrong bank account number is still a serious problem.

Field-level accuracy is useful for a different job: improvement. It shows which fields cause errors. If 8 of the 12 errors are in the cost-centre field, the fix is probably a better lookup list or a validation rule, not a general retraining of the team. It also makes processes with different numbers of fields more comparable, which is why it appears in Six Sigma style measures.

DPMO and the Six Sigma view

Defects per million opportunities (DPMO) is the field-level error rate scaled to a million: 12 ÷ 5,000 × 1,000,000 = 2,400 DPMO. Six Sigma practice converts DPMO into a “sigma level”, with the well-known figure that six sigma corresponds to 3.4 DPMO under the standard 1.5-sigma shift convention. DPMO is helpful for comparing processes of different complexity, but it has a known weakness: the answer depends on how many opportunities you define. Declaring 50 opportunities per invoice instead of 25 would halve the DPMO without a single change in work. Opportunity definitions should be fixed and documented before the metric is used for targets.

Not all errors weigh the same

A misspelt vendor name and a wrong payment amount are both errors, but only one causes financial loss. Most back-office quality frameworks separate errors into categories such as:

  • Critical or fatal: financial impact, regulatory breach, or data sent to the wrong party.
  • Major: affects downstream processing or needs rework, but caught before harm.
  • Minor: cosmetic or formatting, with no downstream effect.

Contracts often set a separate, stricter target for critical accuracy. If 2 of the 9 erroneous invoices had critical errors, critical accuracy in the sample is 99.0%. A team can meet an overall target while failing the critical one, and the critical one is usually what the client’s auditors care about.

From sample to the whole month

The audit checked 200 of 2,400 invoices. If the sample was random, the best estimate of the error rate for the whole month is the sample rate: 9 ÷ 200 = 4.5%, which suggests around 108 of the 2,400 invoices contain at least one error. Most of those were never seen by quality and went out to the client.

The estimate has uncertainty. With 200 samples and a 4.5% error rate, the approximate 95% margin of error is about ±2.9 percentage points, so the true transaction error rate could plausibly be anywhere from under 2% to over 7%. That uncertainty matters when results sit close to the target.

The word “random” is doing a lot of work here. If auditors pick invoices from experienced associates because they are easier to check, or audit only the first hour of the day, the sample does not represent the month. Stratified sampling, where each associate or each invoice type is sampled in proportion, gives a fairer picture. New joiners are usually audited at a much higher rate, sometimes 100%, until they reach a stable accuracy.

Internal accuracy versus client-found errors

Many clients run their own audits or track errors found downstream, for example invoices rejected by the payment system or claims reopened by policyholders. These “external” error rates often look worse than internal audit results, because the client finds errors in work the BPO never sampled. A persistent gap between the two numbers is a sign that the internal audit sample, the error definitions or the audit checklist need revisiting.

The most durable way to raise accuracy is to move from detection to prevention: field validations in the workflow tool, checklists for high-risk fields, maker-checker controls on high-value transactions, and fixing the source documents or templates that cause repeated confusion. Audits measure accuracy; process design is what changes it.

Written by the RP Strativa Team. Figures in the examples are illustrative; the margin of error uses the standard normal approximation.

Leave a Reply

Trending