The cart line that decides the sale.
Someone buying a fridge online looks at three things: the price, the photos and the shipping line. If that line shows a high price, a long wait or "we don't deliver to your postcode", the sale ends there.
Behind that line there are a lot of rules: how much the product weighs and how much truck space it takes, which carriers serve that postcode, how much each one charges, whether the order came from the website or a marketplace, which regions the store would rather not serve. Every store has its own.
WebContinental used Intelipost, a shipping system rented by subscription. It worked for the common case, but the business's specific rules did not fit, and every adjustment waited in the vendor's priority queue.
Four sums that did not add up.
The problem was not that shipping was expensive to compute. It was that the store could not tell the system how to compute it. When the store's rule does not fit, the system falls back to the generic rule, and the quote comes out different from what the store wanted to offer.
Situation before
A shipping engine with the house rules.
We built a shipping calculation and management engine from scratch, in Go. A quote goes through a sequence of steps, all configurable by the store:
Billed weight. Carriers charge the larger of the actual weight and the "volumetric weight", the truck space the box takes converted into kilos. A light, bulky fridge pays for its space.
Who serves the postcode. A carrier registry with served areas, restrictions and exceptions by postcode.
The right table. Different prices per channel and per customer, read from tables the store imports.
Plan B. If no carrier serves the address, the engine offers the alternative the store defined, instead of simply losing the sale.
History. Every quote is recorded, with the rules in force at that moment. You can explain later why the customer saw that price.
67 days, R$ 150 a month.
Before · rented system
After · in-house engine
In practice: the store swapped a subscription for a small fixed server cost and started deciding on its own how shipping is computed. The build investment existed and, according to the project's own analysis, pays for itself in a few months; the exact price of the old subscription is not public.
“We paid a lot for a SaaS that didn't do what we needed. In 67 days we had our own system that costs R$ 150/month and does exactly what we want.”
Management, WebContinental (translated)Build or rent your shipping?
Building pays off when
- shipping rules are part of the strategy (channel, region, product type);
- quote volume is high and the subscription weighs;
- the store has people to maintain the system, or a partner who does;
- quote history matters for audits or for negotiating with carriers.
Renting pays off when
- the store is starting and the common case is enough;
- nobody will look after the system once it is built;
- the rules are the same as any store in the segment.
And there is a new reason to keep shipping under control: the AI assistant that shops for the customer also asks about shipping. If price and delivery time are in a data sheet machines can read, the agent can compare and pick the store; if they only appear after clicking a button, it does not see them. An in-house engine makes it easier to publish that information the way agents understand. That is our reading, not part of the 2020 case.
The space the box takes in the truck, converted into kilos. The carrier charges the larger of it and the actual weight.
Each carrier's price list by region and weight band. A store can have different tables per channel.
The alternative the engine offers when no carrier serves the postcode, instead of blocking the sale.
Where this could be wrong.
The project's own numbers
The 67 days, the R$ 150 a month and the millisecond answers come from the 2020 project, as reported by the team and the store, with no external audit.
"Thousands" is not a number
The old subscription price is not public. The case says "thousands a month"; the exact saving cannot be computed.
2020 is not today
Rented shipping systems have changed since then. The comparison holds for what existed at that time.
Building has an upkeep cost
An in-house system needs someone to look after it: new carriers, new tables, rule changes.
Want to see your store the way an agent sees it?
The free reading shows, within 24 hours, what an AI assistant finds when it visits your store, including shipping and delivery time. (Service in Portuguese, for Brazilian stores.)
← Agent-ready e-commerce · stickybit.com.br
- Original WebContinental case (2020): replacing Intelipost, 67 days, R$ 150/month infrastructure, millisecond answers at checkout.
- WebContinental, home-appliance online store · Go, the engine's language · PostgreSQL for carriers and rules.