From SAP to AGVs, everything went live and the line got slower: none of the 8 systems owns takt time

Six months after every system went live, the shop floor handed in an improvement list with 35 items.

One of them: switch off automatic order grouping driven by the production plan, go back to grouping orders by hand, and release them one at a time as the line actually progresses.

The first thing people asked for after an expensive automation project was to go back to manual.

1. First, the distinction: 8 systems, each working in a different unit of time

A complete chain looks like this:

• SAP — the enterprise planning system; plans and books by the day
• SRM — manages supplier deliveries
• MES — the shop-floor execution system; breaks the plan into work orders
• QMS — the quality system; incoming material can’t be used until it passes inspection
• LES — the in-plant logistics system; breaks work orders into delivery orders for each workstation
• WMS — the warehouse system; picks and stages stock in batches
• WCS / RCS — equipment control; dispatches the stacker cranes in the AS/RS (the machines that move loads up and down the racking) and the AGVs (automated guided vehicles) on the floor

Each one does its own job correctly. The problem is that they don’t share a unit of time: SAP works by the day, LES by the delivery order, WMS by the picking batch, the equipment by the individual command.

This isn’t a flaw in any vendor’s software. In ISA-95, the common layering standard in manufacturing, the planning layer runs in months, weeks and days, and the shop-floor layer runs in shifts, hours and minutes.

An assembly line runs on takt — every fixed number of minutes, the next unit moves forward one station. Of the 8 systems, not one uses takt as its unit.

2. Takt has three upstream inputs, and every one of them keeps changing

The system chain only passes messages along. The three things it passes down are unstable to begin with:

The plan. In our chain, the scheduling system’s column read “to be planned later.” The plan is made by the day, and the floor inserts and changes orders every day. LES groups and stages material from the plan, so once actual progress drifts, it goes wrong at both ends: parts for the next unit aren’t staged, while parts for units further out sit on pallets taking up line-side space.

Material planning has a name for this: “system nervousness.” Change the top-level plan once and the change amplifies level by level down the BOM. The research reached its conclusion long ago: a frozen period that locks the plan works better than more safety stock.

The BOM — the list of parts that go into one product. When a design changes, you have to decide from which unit the new part takes over; in the trade that’s the effectivity breakpoint. Between engineering changing the drawing and the shop floor changing the part sit several systems — design, planning, execution — and if any one of them doesn’t sync, the delivery order still carries the old part. Get the breakpoint wrong and old and new parts sit at the line side together, and the assembler has to stop and ask.

Supplier deliveries. One assembled unit needs hundreds of parts, and one missing part means it can’t start. This is multiplication: with 300 parts each on time 99% of the time, and assuming they’re independent, the rough odds that all of them arrive together are about 5%; at 99.9% each, it’s still only about 74%. And arriving isn’t the same as usable: until incoming inspection is done, the part is frozen in the system.

Large companies can’t escape this either. In a 2022 response letter to the US SEC, Caterpillar wrote that transportation shortages hit its suppliers and delayed production at its plants, and that it had to change assembly processes and carry more work in progress.

The 8 systems pass the changes in these three places straight down the chain, and each layer adds its own delay.

3. Automation took away the floor’s “sponge”

Warehouses in the manual era weren’t accurate either, but they ran. The storekeepers were absorbing the variation: whichever line was urgent got served first, and if a part was missing they went and found it themselves.

A stacker crane only takes commands; it doesn’t know “this line is urgent.” We ran into this: the same material was spread across several storage zones, so the conventional racking zone held up the AS/RS; and loads coming out of the AS/RS went straight to the picking stations, where machines were fast, people were slow, and a queue formed.

The fix in the end wasn’t more equipment but more buffer, less variation. Once the staging logic was fixed and a buffer area was added, outbound throughput caught up with the design takt.

Behind this is Kingman’s formula from queueing theory: the busier a step and the more variable its input, the faster waiting time grows — near full load it climbs almost vertically.

4. So what do you do: treat takt as its own acceptance criterion

Information systems solve for accurate records; takt needs accurate timing. I work to four rules:

① Stabilize the inputs first. Give the plan a frozen period with no changes for a few days; fix BOM changes to a specific breakpoint unit; move the kitting check to the day before a unit goes on the line. Test: no more stops for missing parts on the day itself.

② Pull by actual progress, don’t push by plan. Release deliveries according to where the line actually is, in workstation order. Test: line-side pallets that won’t be used today approach zero.

③ Leave buffers at the seams on purpose. Between the AS/RS and the picking stations, and at the line side — and they must be calculated. Test: you can say why each buffer is the size it is.

④ Add one cross-system metric to acceptance: how often, and for how long, workstations wait for material. And have a manual issuing procedure ready for when the systems go down.

---

I haven’t fully worked out rule ③ myself: how many units’ worth of material at the line side is enough? Too much and it’s inventory; too little and the line stops the moment takt breaks.

How do you size the line-side stock where you are?