Get your e-commerce integration in order
Peter Nilsson 
A customer logs in and sees their contract price. When the item goes into the checkout, the price has been replaced by a general web price. The order goes through, but by then it is already too late. Sales has to issue a credit, customer service has to explain and finance has to correct the records in the business system.
In another case, the stock level looks right on the website, but the order gets stuck between the platform and the business system as load increases. The customer gets a confirmation, the warehouse sees no pick order and the business has to chase the error by hand through several systems. Situations like these are where e-commerce integration stops being a technology project and becomes a governance question about price, stock, orders and responsibility.
A simple sync of products, prices, stock and orders is therefore rarely enough. A working e-commerce integration needs a decided owner for every data set, clear business rules, measurable operations and a stated responsibility after go-live.
E-commerce integration is a governance question for the business
Many companies already have e-commerce that affects sales, logistics, customer service and finance at the same time. The website shows availability, the customer chooses a delivery method, the pricing logic decides what may be sold to whom, and the order has to land correctly in the business system without manual interpretation on the way. E-commerce is therefore part of the operational chain, even when the business system is still the hub for transactions, accounting and physical availability.
As volumes grow, manual transfers quickly become a risk. A file with yesterday's stock level can show sellable quantities that are already reserved. A price at checkout can be based on the wrong price list. An order status can stall in the integration layer without anyone noticing until the customer gets in touch. Every discrepancy creates follow-up work in several places, from investigation and correction to customer contact and financial follow-up.
The practical goal of an e-commerce integration is therefore to reduce the number of manual decisions in the order flow. The integration should turn business rules into a traceable flow between systems. It should give the business control over what is sold, at what price, from which warehouse and with what status at every step.
This has also become a broad business issue. According to Statistics Sweden's figures on ICT usage in enterprises, 36.7 percent of Swedish companies with at least 10 employees sold online in 2024. At the same time, PostNord's E-barometer for 2024 shows that Swedish e-commerce grew by 5 percent during the year. The web was the most common channel, and EDI was used by 11.2 percent of companies. In practice, that means the same item, the same customer and the same order logic often have to work in several channels at once, with different pricing rules, different ranges and different traceability requirements.
When companies underestimate this, the problems often end up at the boundary between functionality, data and responsibility. That is also where analyses of why replatforming gets priced twice usually start, because the cost rarely sits in a single screen. It sits in how rules, data and dependencies actually connect.
Master data needs a clear owner
A two-way sync sounds simple until two systems each believe they know best about the same item or the same price. Moving data between systems is then not enough. Every data set needs a decided owner, a direction and a rule for what happens when values differ.
In most cases, the business system should own item numbers, base prices, stock, orders and order status. That matches its responsibility for transactions, finance and physical availability. PIM should own enrichment such as product descriptions, attributes, documents and channel variants. The e-commerce platform should present the information, manage the search index and apply the channel's presentation, but it should not normally become the original source of the transactional facts that control what the customer can actually buy.
That principle matters most when you have customer-specific terms. A contract price must not be overwritten by a general web price just because the pricing logic happens to run in the wrong system or in the wrong order. In the same way, the platform should not become the owner of stock just because it displays the stock level. It shows a decision made somewhere else, and it needs to be able to explain where the value comes from and when it was updated.
The same reasoning applies to order status. If the business system is the one that books, reserves and moves the order forward in the operational process, its status also needs to be the one other systems relate to. The platform can show the status, but it should not invent its own truth that later has to be reconciled with the warehouse or finance.
The practical work should start in a field matrix. It documents who owns each field, which data type applies, when the field is mandatory, how the value is validated and what happens on an error. A matrix like this quickly reveals whether the organisation is trying to let several systems be the authority at once. It also makes it easier to see where a PIM really adds structure and control to product information, especially when several channels share the same range and the same core data, which you can read more about on the PIM page.
Syncing has to be sized by data type
A stable e-commerce integration does not treat all information the same way. Different data types have different requirements for freshness, load and consistency. Stock, reservations and order status normally need to be handled close to real time, because they affect the customer's buying decision and the business's ability to fulfil the order. Product descriptions, attributes, documents and media can more often be published in batches or through event-driven flows without harming the business.
When everything updates at the same frequency, the solution becomes expensive, unnecessarily heavy and harder to monitor. A common mistake is to let large product flows compete with time-critical order events for the same integration capacity. That leads to queues, longer response times and a greater risk that price or stock changes before the order reaches the business system.
Sizing therefore needs to start from actual load and from how quickly each data type has to be correct in each channel. PostNord's E-barometer for the first quarter of 2025 shows that Swedish e-commerce grew by 9 percent in the first quarter of 2025. Movement like that puts demands on capacity, queue handling and monitoring, especially when demand shifts between categories and order volumes change quickly.
Sizing the integration layer in isolation is not enough either. Dependencies in the business system, databases, warehouse management, payment flows and external APIs often set the real limit. That is why it makes sense to start from transaction responsibility and then decide which events the platform should consume, which is the core of an ERP-first integration strategy.
The flows have to handle repetition and errors
Order flows have to be built to handle the same message being sent more than once. Every message therefore needs a unique ID and a correlation identity that lets the recipient recognise the same order if it is sent again. This principle is called idempotency, which in plain terms means the same action can be repeated without changing the result. An order should not be booked twice just because the integration layer retries after an outage.
Temporary errors should be handled with retries and exponential backoff. That means the system waits a little longer between each new attempt, instead of hitting the same service again and again within a few seconds. Permanent errors should go to a dead-letter queue, a separate error queue where messages collect when they cannot be processed automatically, so they can be reviewed, understood and restored in a controlled way.
For monitoring to be useful, you need to measure both technical and operational signals. The share of successful syncs shows whether the flow works overall. Median and p95 latency show how long things normally take and how long the slowest five percent take, which matters when a few slow flows cause real customer problems even though the average looks fine. The number of manual corrections shows how much work the business actually has to carry. The difference between the website's and the business system's order and stock figures shows whether the systems still describe the same reality.
Load tests therefore need to cover both normal traffic and planned peaks with a safety margin. A test that passes on a quiet afternoon is not enough. The flow has to hold when campaigns, import jobs and order peaks coincide. Queue depth should be tracked continuously, because growing queues often signal problems long before the customer sees an error message.
Test data quality through the whole checkout
Data quality cannot be tested as an isolated question for a single system. A price that changes too late, a discount that disappears at checkout or stock that is not reserved correctly quickly becomes a business problem, even if each system works according to its own logic. Testing therefore needs to follow the whole sales chain, from master data in the business system and PIM to the order confirmation coming back from the business system.
The field matrix is central here too. For item, customer, price list, stock, delivery terms and order, it should state which data type applies, who owns the value, whether the field is mandatory, which rule decides whether the value is valid and which fallback to use when a value is missing or invalid. That documentation means a new mandatory field produces a clear error before production, so the order flow does not break without an understandable explanation.
Test data also needs to reflect the real business. That means Swedish company registration numbers, personal identity numbers where relevant, postcodes, VAT rules, SEK and Nordic addresses. Contract tests between the business system, the integration layer and the platform should make sure that changes in one system do not accidentally break another. When a field becomes mandatory, a value changes format or a pricing rule gets an extra condition, it should be caught before deployment.
It is also necessary to test what happens when the flow fails. A sold-out item, a price change between cart and checkout, an invalid discount, a partial delivery, a cancelled order and a timeout against the business system are typical scenarios that must give predictable outcomes. All affected systems need to show the same rule outcome when a customer segment, price list or stock status changes. Otherwise the discrepancy gets built into the process and only shows when the business starts correcting things by hand.
A pilot with a limited range makes it easier to isolate errors before the whole range or all customer groups go live. Follow-up should then focus on orders that need manual correction, validation errors per 1,000 transactions and the time from placed order to confirmation in the business system. Those measures say more about real operational reliability than green technical logs alone.
Security and continuity start when something goes wrong
More integrations mean more identities, more API keys, more permissions and more points of failure. That does not make the business more mature in itself. Integration security needs to be handled as part of ongoing operations. According to Svensk Handel's report on security in retail (in Swedish), many retailers say they lack enough support to protect themselves against digital crime. That makes it reasonable to treat integration security as a concrete operational responsibility that continues after launch.
When the business system is down, there needs to be a decided model for how orders are received, queued and resumed. In some cases, the platform can keep accepting orders into a secure queue, but only if there are rules for how long the queue may grow, how orders are prioritised when the system comes back and how each order is then processed exactly once, using the same idempotent principle as in normal operation. Without that model, the problem is only postponed and grows bigger when the systems restart.
It is also important to document RTO and RPO for the integration. RTO stands for recovery time objective and describes how quickly the flow should be back after an outage. RPO stands for recovery point objective and describes how much data loss the business accepts, for example whether a few minutes of order events must be recoverable or whether no loss at all is acceptable. These targets need to make sense to finance, sales and operations, because they directly affect the cost of monitoring, redundancy and recovery.
Permissions should follow the principle of least privilege. API keys should be rotatable without major disruption. Logs should link orders, messages and retries without exposing sensitive data. Recovery should also be tested in practice. A documented routine that nobody has run gives no real readiness when the pressure is high and several suppliers need to coordinate the next step at the same time.
Ask the supplier three questions before you decide
How are price, stock, order status and customer-specific terms split between the systems, and which system is the source when values differ? The answer shows whether the supplier understands your business rules and can turn them into a lasting master data model, beyond selling a technical connection.
What happens when a message fails, and how do retries, error queues, alerts and manual recovery work in practice? The answer shows whether the solution is built for operations and exception handling, or whether the business is expected to discover and sort out problems only once the customer is affected.
Who is responsible when the platform or the business system is upgraded, who tests compatibility and what does the contract cover when something needs fixing? The answer shows whether product ownership continues after go-live, or whether it effectively ends at handover and is replaced by new quotes for every change.
How Glanser works with integration responsibility
Glanser builds integrations around clear flows for products, prices, stock, customers and orders, using solutions from the Function Library with a fixed scope and fixed price where that suits the business. Integrations with Business Central, Monitor and Jeeves are common examples, and customer-unique parts can be kept separate so the rest of the solution is easier to upgrade and maintain over time.
For this to work, the customer and the supplier need to agree on data ownership, business rules and error handling before the integration is built. Glanser Care then takes responsibility for keeping the integrations compatible through upgrades, so the solution can keep developing without every version change becoming a separate integration project.
On the page about ERP and integrations, you can see how Glanser describes integration flows, the split of responsibility and examples of connections between e-commerce and business systems.

About Peter Nilsson
Peter Nilsson writes about integrations and product information, focusing on how business systems, PIM and the web can share the same data without duplicated work.

How to govern product information across every channel
How to structure product information so every attribute has an owner and a source.
Peter Nilsson
Switch e-commerce agency without losing orders or data
How to switch e-commerce agencies without losing data, orders or control.
Kenneth Lyngshede 
ERP and EDI that keep orders and invoices together
ERP EDI integration explained: which messages are included, how to validate data, and which metrics indicate whether the flow is working.
Peter Nilsson