A packing process that depended on the judgement of whoever reviewed orders that day. The judgement was good. The dependency was the problem.
Verified customer review · published with permission
Keto Chow
Outcome
Average label cost
$9.33 → $8.49
First month, net
$10,880
Packing decision
Holds without the expert
Also covers · Rate Selection
Keto Chow makes meal-replacement shakes and sells direct to consumer. The catalogue combines in a lot of ways: multiple flavours, multiple sizes, ordered in quantities that change which box actually works. Growth had been quick enough that fulfilment volume had outgrown the process that carried the operation to that point.
This is a common position for an operation of this size. Nothing was broken. The process worked, and it worked because someone competent was running it.
Packing depended on whoever was reviewing orders that day. That person knew, from experience, how many of which products fit into which package, and roughly which service came out cheapest for the resulting parcel. Two decisions, made by feel, hundreds of times a day, mostly correctly.
The exposure was not accuracy. It was that the knowledge existed in one place and could not be relied upon to be present. It held until that person took holiday, was ill, or order volume rose past what one person could review carefully.
An operation in this position does not have a software problem yet. It has a single point of failure that happens to be a person, and a cost line that quietly moves depending on who is working.
ShipStation handled the standard cases correctly. Its rules engine applies a package and a service consistently, every time, once you can state the rule — and for a large number of operations that is genuinely sufficient.
The difficulty here was that the rule could not be stated. The knowledge being applied was which combination of these particular products fits into which box, evaluated against an order profile that changes with every order and a catalogue that changes with every new product.
A static IF/THEN engine can express what to do once you already know the answer. It cannot derive the answer per order. That is not a configuration failure or a missing feature — it is the boundary of what a rules engine is for, and no amount of additional rules moves it.
What replaced the rules was not a bigger set of rules. It was a calculation performed per order, in which three questions are answered together rather than in sequence:
The reason these interact is that the cheapest service depends on which box you use, and the best box depends on what the resulting parcel costs to send. Answered separately, each decision looks correct while the pair is wrong. It is one decision with two outputs, and the only way to get it right consistently is to evaluate it as one.
The average label fell from $9.33 to $8.49, and the first month returned $10,880 net of the cost of the software.
The more durable change is harder to put a number against: the packing knowledge stopped being a person and became a decision made identically on every order, including the orders processed on a Tuesday in August when the person who used to make it is not there.
Implementation required ShipStation access and accurate dimensions and weights for both products and packages. From there it was iterative — the team reported what needed adjusting and the logic was tuned against real orders rather than assumed ones. That is the normal shape of this work, and it is worth budgeting for.
"We get a pretty constant string of 'hey we will save you money on shipping' pitches but none of them ever saved us anything because our process was already really great. String had a different angle: it would look at all of the options available in ShipStation and pick the least expensive — I felt that was worthwhile to look at since it wouldn't require us to change our fulfillment process or train employees on new software."
What another operator can take from this review, whether or not their operation looks anything like this one.
If one person's absence changes your postage cost, the knowledge is real but the process is not. Cost variance by shift is a measurable symptom.
Correct package and cheapest service are a single decision. Optimising them separately produces two locally sensible answers and one worse outcome.
Packaging cost belongs inside the rate comparison. A cheaper label in a more expensive box is not a saving.
An operation that already runs well is not evidence that nothing is leaking. Competence and consistency are different properties, and only one of them scales.
Other operational reviews
Rate Selection
4 min
The team knew their rules were mis-allocating orders when they built them. They built them anyway, because the alternative did not scale.
Rate Selection
3 min
A carrier relationship that solved a real reliability problem, then kept being applied to orders it was no longer the right answer for.
Exception Monitoring
4 min
Rules made fulfilment faster and, at the same time, removed the human check that used to notice when an order never actually shipped.
Start here
A short conversation, an export of your order data, and a String Operational Review you keep — along with an honest answer about whether String can meaningfully improve your fulfillment operation. If it can't, we'll tell you. That answer requires a detailed review by us and could save you a year of building the wrong solution.
