Choosing Countermeasures in Accounts Payable
The accounts payable shared-service center processes about 3,400 supplier invoices a month with six clerks. Average cycle time from receipt to approved-for-payment is 11 business days against a 5-day target. Rework affects 14% of invoices. Data show 38% arrive without a purchase-order reference, 22% of PO invoices fail three-way matching, duplicate approvals occur on 9%, and the team works from a shared email inbox with no visual priority. Two approvals are required for invoices over EUR 5,000. Month-end volume is 1.8 times the normal daily rate. The manager has proposed adding two clerks. IT has offered an RPA bot to key invoice data into the ERP. A clerk has suggested a kanban board and 5S for the shared drive. A Yellow Belt is supporting the Black Belt and has been asked to help choose countermeasures for the Improve phase. The sponsor wants a decision this month and is worried that automation will be seen as the safe choice.
1. Evaluate the three proposals (two extra clerks, RPA, kanban/5S) and recommend a countermeasure sequence tied to the data.
Read the worked answer
None of the three proposals is wrong in itself, but each addresses a different cause. Adding clerks adds capacity; with 38% no-PO invoices, 22% match failures and duplicate approvals, it would likely process waste faster and may not reduce the 11-day cycle time. RPA can reduce keying time, but automating an unstable process with missing PO references and unclear match rules risks scaling rework; it should come after standard work and exception rules. Kanban/5S addresses the shared inbox and visual priority, which is useful but not sufficient if the upstream PO and approval causes remain.
Sequence: (1) Reduce incoming defects: require PO at the supplier portal or use a no-PO exception route with clear ownership; work with top suppliers on PO compliance. (2) Simplify and mistake-proof matching: define tolerances, fix master data, and use a hold/exception queue for mismatches instead of rework loops. (3) Remove duplicate approvals with a threshold and delegated approval rules. (4) Implement standard work and a visual kanban for the email queue. (5) Then automate the stable, high-volume keying step. Use an impact/effort view and pilot on one supplier segment or invoice type.
2. Design a four-week pilot and define the metrics and operational definitions you would use to judge it.
Read the worked answer
Choose one controllable slice: for example, invoices from the top 20 suppliers and one clerk team, or one invoice type. Baseline four weeks before the pilot using the same definitions, then run four weeks. If possible, keep a comparable non-pilot group to reduce seasonal and month-end confounding; at minimum, stratify results by month-end and normal days.
Metrics:
- Cycle time: median and 90th percentile business days from receipt timestamp to approved-for-payment timestamp.
- First-pass yield: % of invoices approved without rework or a second approval cycle.
- Rework rate: % returned for missing PO, match failure or duplicate approval; define each reason.
- Touch time: minutes of clerk handling per invoice, from time logs or a sample.
- Backlog and overdue count: invoices past 5 days.
- Supplier PO compliance: % of invoices with a valid PO at receipt.
Set a decision rule in advance: proceed if median cycle time falls by at least 2 days and first-pass yield rises without increasing touch time. Review weekly, plot run charts, and check data quality. Do not judge on a single month-end week.
3. How should the Yellow Belt involve the clerks and address the likelihood that RPA will be seen as the safe choice?
Read the worked answer
The clerks know where invoices actually stall, so involve them early in process mapping and error-reason coding. Ask them to identify the top three handoffs that create rework and to co-design the kanban and exception rules. This turns the project from an IT imposition into a shared improvement and gives the Yellow Belt credibility. Explain that RPA is a candidate, not the default: it will be judged against the same cycle-time and first-pass-yield metrics. Show the data: 38% no-PO and 22% match failures mean automation alone cannot fix the flow.
Address the 'safe choice' concern with a decision matrix: impact on verified causes, effort, risk, time to benefit and durability. Put RPA in the sequence after PO compliance and standard work, and define the conditions under which it is worth doing. For resistance, use a pilot with volunteers, share results weekly, and separate data from blame. Escalate to the sponsor only if a policy barrier (approval thresholds, supplier contracts) blocks the team. The Yellow Belt should facilitate and document, not overrule the Black Belt or the manager.