Supply Chain Action Points
Read this first — the conclusion, and the moves to make:
- Run four weeks of timestamped throughput at receiving, putaway, picking and pack-out, in units per hour, before any vendor visit, and use the lowest of the four as the number you negotiate against.
- Write the acceptance test as end-to-end order cycle time at peak-day volume instead of aisle rate, and hold 10% of contract value until it clears for three consecutive peak days, measured eight weeks after go-live.
- Name one owner for the decant and orchestration layer in the contract, with a single queue and a single interface, and make the batching rule a deliverable with a target of 10% more mixed orders per labour hour by month six.
- Rebuild the payback at 13% market growth instead of 17% and require it to clear at half the forecast volume; if it only works at 17%, the forecast is carrying the decision rather than the equipment.
On 8 July, Italy's Comau announced it had completed a 100% acquisition of Brazil's Invent Smart Intralogistics Solutions, adding case-handling, order fulfillment and orchestration software to its Automha pallet and AS/RS business. The intralogistics market is forecast to grow about 13% annually over the next three to five years; the U.S. and Brazil are seen at 15% and 17%.
The Analysis
An acquisition announcement tells you what a company bought. It rarely tells you where the seam lands, and the seam is the only part a shipper ever feels.
On 8 July Comau completed the purchase of 100% of Invent Smart Intralogistics Solutions, a Brazilian business built around case handling, order fulfilment and the software that sequences both. It goes into Automha, the pallet and automated storage side of the group. The forecast attached to the announcement puts intralogistics growth at roughly 13% a year for the next three to five years, with the United States at 15% and Brazil at 17%.
Drawn as a system, the shape changes. Pallet storage and case picking are two queues with different service times that happen to share one dock door. Owning both ends does not make either end faster. It makes the handover between them one company's problem, which is worth more than the hardware and considerably harder to write into a contract.
Model a distribution centre as four queues in series: receiving, putaway, picking, pack-out. The throughput of the line equals the throughput of the slowest queue, and it stays that way no matter how much money goes into the other three. Automation business cases that fail tend to fail here rather than at the technology layer. The machine does what the spec sheet says. The line does what the worst station allows.
The awkward detail is the unit change. Pallet storage moves pallets an hour. Case handling moves cases an hour. Between them sits a decant or de-palletising station where one unit becomes the other, and that station is exactly where two vendors' performance numbers stop being comparable. A crane rated at 60 pallet moves an hour and a cell rated at 900 cases an hour cannot be added together. Neither vendor is usually contracted on the number that decides your service level, which is dock-to-stock time for a mixed order.
The one line item in this deal that can change throughput without changing steel is the orchestration software. It decides sequence: which order goes to which station, in what order, with what batching. On identical machines and identical labour, a good batching rule and a bad one produce different orders per hour, and neither shows up in a capital line. That is also why buying software from the same vendor as the steel is not automatically safer. Integration is a scope question, not a brand question. What you want in writing is one queue, one owner, and an interface nobody can disown.
The 13, 15 and 17 numbers deserve one more step than they usually get. At 17% compound growth an installed base doubles in about four years and five months. At 13% the same doubling takes about five years and eight months. Brazil therefore doubles roughly fifteen months ahead of the global average, with the United States in between. Fifteen months is not a demand signal. It is a staffing signal, because commissioning engineers, integrators and maintenance technicians do not compound at 17%.
Treat 15 and 17 as a range rather than two separate facts. A three-to-five-year forecast that splits two markets by two points is inside its own error band, and the useful question is not which number is right but what your project looks like at the bottom of both. Build the case at 13 and let 17 be upside. A project that only clears at the top of a forecast is a bet on the forecaster, and the forecaster is not the one paying for the concrete.
Who feels this first is not who you would guess. A large third-party operator with an existing warehouse control layer absorbs the acquisition as vendor consolidation and notices nothing until renewal, when there is one fewer bidder at the table. A mid-size importer running a single distribution centre feels it at the factory acceptance test, where the cell runs at its rated rate while orders still queue at the dock. A small importer feels none of it directly and pays for it indirectly, because integration labour that used to be available at short notice is now booked months ahead.
Timing gets misread in a predictable way. Nothing hurts on signing day. The first real signal is the factory acceptance test, when the acceptance criteria turn out to be written per machine instead of per order. The second is around month nine, when aisle rate is at or above specification and dock-to-stock has not moved. By then the capital is committed and the payback model lives in a spreadsheet owned by somebody in finance, and the discussion is about whose scope the decant station was.
Here is the arithmetic I would put in front of a buyer, with every assumption labelled as mine. Say a centre moves 8,000 cases a day over two shifts, six productive hours each. Receiving takes 45 pallets an hour, the automated store 60, case picking 900 cases an hour, pack-out 1,000. Those four numbers are my assumptions, not figures from the announcement. The ceiling is 900 cases an hour, which is 5,400 a shift and 10,800 a day, so volume looks comfortable. Now let the order mix turn multi-line, which is what happens when a cell is specified on single-line orders and then fed mixed ones, and effective pick rate drops to 750 an hour. The ceiling falls to 4,500 a shift, 9,000 a day, and 8,000 now depends on overtime. A second pallet crane would have moved neither number.
Run the counterfactual the other way. Buy the cell, keep two control systems and two queues, and the bottleneck moves thirty metres downstream while the payback model still assumes the whole line improved. Capital spent, constraint intact, and the only thing that changed is that the constraint now sits inside the new equipment instead of in front of it. That is a more expensive version of the problem you started with, and it is harder to explain at the next budget review.
The trap is buying somebody else's reference site. A reference site was specified around one order profile, commissioned by the vendor's own engineers, and has had two years to shake out. Yours will have a different profile, a different shift pattern, and an integration team learning the software on your volume during your peak. Measure your own four queues for four weeks before you sign, put the slowest one into the contract as the acceptance number, and hold money against it. Optimisation without a baseline is just a purchase with better vocabulary.
↑ Back to the key points