AI Prompts for Operations: SOPs, Process Maps and Vendor Evaluation

A SIPOC lists the suppliers, inputs, process steps, outputs and customers of a process on one page. A value stream map adds the time each step takes and the time spent waiting before it, so the delay can be located.

• 9 hours of work, 52 hours elapsed: value-added time is 17% of lead time. The other 43 hours are queues between steps, and that is where the delivery promise is lost.
• 19 hours at the dock: the old courier collected at 11:00 and 16:00; the new one collects once, at 16:00. Orders packed after the cut-off, 38% of them, now wait overnight.
• 11 handoffs: the order passes between customer service, the warehouse, quality and the courier eleven times, and address changes are keyed in twice.

A standard operating procedure (SOP) is the written, approved way a task is done, with numbered steps, owners and checks. A work instruction is the short version of the critical steps, used at the workstation.

• 9 days to a credit: with no procedure, returns waited for whoever had time. Clinics waited an average of 9 working days for a credit and withheld payment in the meantime.
• 1 credit in 5 was wrong: each person priced returns differently, and clinics disputed $180K of invoices in two quarters.
• One exception with a safety cost: temperature-controlled items had been restocked after return. The procedure now quarantines them at receipt, with no exception.

The Theory of Constraints holds that the output of a process is set by its bottleneck, the step with the least capacity against demand. First-pass yield is the share of units completed correctly the first time, without rework.

• 250 lines an hour short: verification handles 1,050 lines an hour; demand between 10:00 and 14:00 is 1,300. Four hours of shortfall leaves about 1,000 lines waiting each day.
• More people everywhere would not help: picking runs at 71% and packing at 62% of capacity at the peak. Extra pickers would only lengthen the queue at verification.
• 93.8% first-pass yield: 6.2% of 26,000 orders a month, about 1,600, leave with the wrong item or quantity. Most pass through verification during the peak.

A Pareto analysis ranks causes by frequency to find the few that account for most failures. A fishbone diagram groups possible causes by category, and the 5 Whys asks why repeatedly until the answer is a cause in the process.

• 76% from three types: wrong item 41%, wrong quantity 22% and missed line 13% of the 1,612 errors logged in September.
• The override is the process cause: verification lets a picker skip a failed scan with a manual confirm. At the peak it is used on 18% of lines, and error rates are the same for new and experienced staff.
• $103K a month: each error costs about $64 in reshipping, credit processing and labour. 1,612 errors in September cost $103K, about $1.24M a year.

Failure mode and effects analysis (FMEA) lists the ways each process step can fail and scores each on severity, occurrence and detection. Multiplied together, they give the risk priority number, used to decide which failures to prevent first.

• 31 parcels, $43K: each lost temperature-controlled parcel costs about $1,400 in product, reshipping and credit. 31 in a quarter is $43K, before any cost to patient care.
• 315 from a calendar: a 48-hour cold pack cannot survive a weekend. Sending cold-chain parcels only until Wednesday 16:00 removes the failure mode rather than inspecting for it.
• 240 from a missing check: nothing stopped a parcel leaving without its cold pack. Scanning the pack before the label prints makes the error visible at the point it happens.

Lean process redesign removes the eight wastes from a process: transport, inventory, motion, waiting, overproduction, over-processing, defects and unused skills. The future-state map shows the process after the changes.

• Waiting is the largest waste: a second daily collection removes the 19-hour overnight wait for 38% of orders, and brings lead time from 52 to 22 hours.
• Defects are the most expensive waste: errors cost about $1.24M a year. Removing the override and re-slotting 120 look-alike products targets an error rate of 1.5%, saving about $0.94M a year.
• The bottleneck gets capacity only at the peak: a second verification lane from 10:00 to 14:00 costs $38K a year, instead of adding people across the warehouse.

Total cost of ownership (TCO) adds to a supplier’s price every other cost of using it: failures, rework, switching and management time. A weighted scorecard fixes the selection criteria and their weights before any bid is opened.

• $414K saved, $2.2M lost: the switch cut the $4.6M delivery bill by 9%, $414K a year. Credits alone ran at $1.1M in two quarters, $2.2M a year: a net loss of about $1.8M.
• Weights fixed first: on-time performance 30, cold-chain capability 25, cost 25, service and claims 10, systems 10. The March decision had used price only.
• Bid C at $4.8M: dearer than today’s price by $0.2M, cheaper by $1.6M on total cost, and the only bid with dedicated cold-chain vehicles.

A service level agreement (SLA) defines the service a supplier must deliver as measured targets. A service credit is the reduction in the supplier’s charge when a target is missed.

• $0 of $1.1M recovered: the March contract promised “best efforts”. Brennock paid clinics $1.1M in credits and could claim none of it from the courier.
• A credit sized to matter: Bid C’s monthly charge is about $367K. Each point below 97% costs the courier about $7.3K a month, up to a cap of $55K.
• A way out: three consecutive months below 94% gives Brennock the right to terminate, with a 90-day handover the supplier must support.

A supplier performance review compares a supplier’s results with its contract, calculates what is owed for misses and agrees corrective actions. A quarterly business review (QBR) is the meeting where this is done.

• $8.8K in January: on-time delivery was 1.2 points below the 97% target. At 2% of the $367K monthly charge per point, the credit is $8.8K, and it is claimed.
• The miss has one location: 8% of clinics are on rural routes, and those routes account for 61% of late deliveries, most of them on Tuesdays.
• An action with a date: the courier adds a second van on rural Tuesday routes from 15 February, and the February scorecard shows whether it worked.

A control plan lists the measures that keep a process on standard, their limits and the response when a limit is breached. Statistical process control uses control charts to separate normal variation from a real change.

• Six days against eleven weeks: on-time delivery fell below 95% six days after the courier switch. Without a limit or an owner, it was noticed after eleven weeks and $1.1M of credits.
• 2.1% is calculated, not chosen: the upper control limit comes from eight weeks of daily error rates since the redesign. Above it, the warehouse manager starts a review the same day.
• Six measures, six owners: each measure on the dashboard has one owner and a written response. Nine other measures the teams used to report were dropped.