Web maintenance after launch and what to demand
Kenneth Lyngshede
The website went live in the spring. Now sales wants to change a price filter, the product team wants to add an integration field and the CFO wants to open the site to a new customer group. Everything worked at delivery, but the changes affect several systems at once.
The quote describes what you bought. The web maintenance agreement needs to describe what happens now, who assesses each change and which work the monthly cost covers. Otherwise every adjustment risks becoming a new discussion about responsibility and price.
Web maintenance starts once the website is live
A new price filter can look like a small change on screen. Behind the filter sit product attributes, customer prices, the search index and rules for which products the customer may see. Maintenance starts with someone assessing that whole chain before the work is ordered.
Web maintenance covers several workstreams that need to be held together. Operations and security come first, with monitoring, certificates, backups and tested restores. Version tracking covers the platform, the content management system, the connection to the business system and third-party components. Bug fixing needs to be sorted by how the business is affected, while further development adds functions in a controlled way.
Content and product data also need their own place in the planning. New documents, product images and translations keep being published for the whole life of the solution. A project has a handover date. Maintenance continues for as long as customers, editors and salespeople use the solution.
The need for recurring checks shows in the Swedish Agency for Digital Government's monitoring of public digital services, which covered 1,111 public websites between 2022 and 2024. The material concerns the public sector and shows why reviews need to cover how the services work for users.
A practical maintenance calendar starts with weekly follow-up of incidents, failed syncs and publishing errors. Each month, you review versions, restore tests and prioritised changes. At every deployment, the affected user flows are checked and the result is documented. The calendar makes the work possible to plan and gives you a basis for checking when the supplier reports what has actually been done.
Service levels and what the agreement actually promises
A basic level can cover a monthly review, version tracking and handling of reported errors during office hours. An extended level can add continuous monitoring and set times for deployment. Business-critical flows need their own targets for orders, checkout and customer login, together with agreed on-call cover. These are possible agreement setups, with different demands on staffing and cost.
A service level agreement, often called an SLA, needs to separate response time, troubleshooting and restored function. An acknowledgement says the ticket has been registered. It says nothing about when customers can order again. The agreement should also state when the clock starts, which opening hours apply and how planned maintenance is treated. A useful SLA describes what the supplier does when something deviates and how you can check that the work was done.
As a hypothetical example, an availability target of 99.5 percent in a thirty-day month allows at most three hours and thirty-six minutes of downtime. The percentage needs to be judged against your sales hours and the functions included in the measurement. The web server can respond while customers are missing their contract prices. That is why the agreement needs to state the measuring point, the check method and which exceptions are excluded.
Priority classes should describe the impact on the business. Critical errors can stop purchases or hide order status that customers need. Serious errors may have a temporary workaround, while minor errors and improvements get planned fix times. A reasonable starting point is that a critical error gets a human response the same working day and an action plan within a working day or so. An outage that stops sales can require much faster cover.
On-call cover costs money even when nobody calls. It can be justified when orders come in outside office hours, but adds limited value if the business can wait until the next working day. You should be able to express your current agreement simply: what is measured, how often it is measured and what the supplier does when something deviates. Unclear answers on any of these points need to be sorted out before signing.
Responsibility is shared between supplier and customer and belongs in the agreement
The customer prioritises and orders the change. The supplier assesses scope, dependencies and risk. After testing, named people at the customer approve the change, and the supplier then deploys it, documents it and follows up the result in production. A working way of working makes every handover clear and possible to follow.
The agreement needs to name who owns the change queue, who approves deployment and who is responsible for product data and editorial content. Also name a deputy for when the regular decision-maker is away. Otherwise finished work can get stuck before launch, or be published without the business's approval.

An editor might, for example, rename a product field that is also used in a sync. The visible text change can then affect product selection or data transfer. Third-party services need the same kind of boundary: someone has to be able to renew access, receive error messages and decide whether the service should be replaced.
This matters most when AI is used to enrich product information. AI can be allowed to suggest descriptions, keywords and translations within set fields. Item numbers, prices, stock levels and order status should come from designated source systems and be protected from automatic rewrites. Safety information and product properties need human approval before publishing. Log the original value, the change, the model, the instruction and the approver. Otherwise an incorrect suggestion can spread to the website, customer portal and reseller channels at the same time, which makes the fix much bigger.
Upgrades need a test suite that runs before release
A successful connection test shows that the systems can talk to each other. When the item structure or an API contract changes, an error can still spread through PIM, staging storage, the search index and the order flow. Every connection can respond correctly while the customer gets the wrong price or the order gets stuck.
The test suite therefore needs to follow business events. Syncing should be idempotent, which means a repeated transfer gives the same result without duplicates. Check prices, stock levels, order status and permissions for the relevant customer groups. Also test what happens with timeouts, retries and transactions that only partly completed.

A synthetic flow test can create a test order from the product page to the confirmation in the business system. It runs in the test environment before deployment and compares the outcome with expected values. The questions about integration with the business system therefore need to cover the whole order, including rejected orders and missing mandatory details.
For technical maintenance, track the error rate per integration job, queue length, sync delay, successful retries and time to restore.
Version-lock the API contracts and test rolling back to the previous version in the test environment. A restore instruction needs to work even after data has changed. Major versions of the business system and major platform changes justify the full test suite. A limited bug fix can get a smaller check, as long as the dependencies are mapped. Schedule deployment when order volumes are low, and leave a real approval window for the business before the change is released.
Accessibility has to be checked with every change
Accessibility needs to be part of the work whenever content, templates and functions change. The Swedish Post and Telecom Authority describes the requirements of the Swedish Accessibility Act for services to consumers, including e-commerce, and the link between EN 301 549 and WCAG 2.1 at levels A and AA. The Swedish act on accessibility to digital public services applies to the public sector. Clarify which rules your service falls under before you decide the scope of the checks.
The requirements need to be turned into things editors and developers can test. A product image should convey relevant information even when the user cannot see it. The customer's route through checkout should work with a keyboard, and form errors need to be understandable and fixable.
The authority describes the principles of perceivable, operable, understandable and robust. Alt text helps the user perceive what an image shows. Keyboard navigation makes the interface operable, clear error messages make forms understandable and correct code structure helps assistive technologies interpret the page.
After launch, a new image can lack alt text, or a changed checkout can move focus to the wrong place. A form can show an error that the screen reader never announces. Embedded third-party components therefore need to be part of the same check as the rest of the page.
Run automated checks before publishing where possible. Add manual testing with keyboard and screen reader for larger changes, and track remaining errors per deployment. Use the agreed responsibility matrix to assign issues to the right person. Automated tools find some errors, while others require someone to actually complete the task. Even an accessible core platform can develop serious gaps as new copy, documents and components are published.
When the setup demands more than you expect, or does not fit
Ongoing maintenance gives the most value when the solution is used daily and keeps changing. An agreement that only covers server operations leaves questions about integrations, product data and content unanswered. The more dependencies you have, the more work it takes to keep an up-to-date picture of them. The discussion of composable commerce and system dependencies is relevant when you assess how many parts need to be coordinated.
The model does not fit everywhere. Take a smaller content website that rarely changes. There, the agreement should be limited to operations, monitoring and a channel for reporting errors, while changes are ordered and priced one by one. Such a clause might read: "The agreement covers monitoring of the domain, certificates and publishing environment, a response by the next working day when an error is reported, and an annual review of the platform version. New pages, template changes and integrations are ordered separately at a fixed price per function." You then pay for what actually needs to happen.
A limited one-off project follows the same logic. If nothing is going to change after launch, there is no maintenance work to divide up, and a large ongoing commitment becomes expensive without moving the work forward while ownership is still unclear. An agreement that does not name which functions are included is, in practice, hourly billing with an extra step. So write in both the functions and the service levels.
A heavier setup is justified when the solution carries orders, stock and customer data to and from the business system and changes every week. A channel for reporting errors is not enough then. The agreement needs to cover upgrades of the platform and the business system, plus priced functions for anything that is added.
Count maintenance together with the build when you assess TCO, the total cost of ownership. For the calculation, you can compare over three to five years. That is a chosen planning horizon, not a measured norm. Include your own staff time, third-party fees, testing and the consequences of skipped updates. Then map your current situation before you choose a service level, so the agreement matches the risks you actually have.
Three questions to ask the supplier before you sign
Which versions of the platform and the business system does the commitment cover, and how long is an older version supported after a new one is released? The answer shows whether there is a planned upgrade pace and clear limits to the support. Ask the supplier to specify what happens when a component reaches the end of its support period. It needs to be clear whether the work is included and which decisions you have to make.
What happens to a new function when the platform is updated, and who tests it against your integrations before deployment? The answer shows whether compatibility is handled through a documented process with real test results. A useful report links the test cases to the delivered function and lists any remaining deviations. It should also show which customer groups and business flows were tested. Ask for material from completed work. Test reports and incident logs show how the supplier applies the agreement when something needs handling.
How many critical errors were resolved within the agreed time last quarter, and who reports the outcome? The answer shows whether the supplier can follow up its commitment with a clear measurement method. If the period had no critical incidents, that should be stated, along with how the on-call readiness was checked.
Ask for an anonymised example if the material contains other customers' details. The answers should be backed by test reports, release notes and incident logs. Also check that the same definitions are used in the sales material, the agreement and the ongoing reporting. Then you can judge what the supplier actually commits to once the solution starts to change.
Decide who owns maintenance before you order
An order without a named owner for the period after go-live leaves upgrades, errors and improvements with nobody to receive them. When you compare e-commerce platforms and ongoing maintenance, assess the agreement at the same time as the technology.
After launch, we maintain the solution as a product through Glanser Care, a contracted responsibility at a fixed monthly cost that covers version tracking, compatibility through upgrades, maintenance of the integrations with the business system and bug fixing. Upgrades do not require a new quote. The line between agreed improvements and new functions still needs to be spelled out in the order.
Include a written split of responsibility in your next request for quotes: who decides on a change, who tests it against the integrations and who is responsible if it breaks a flow. With the same questions to every supplier, the answers become comparable and ownership is clear before you sign.

About Kenneth Lyngshede
Kenneth Lyngshede writes about business models, offers and strategy, and how companies can plan their digital development with predictable costs.

ERP systems as the hub for your website and portal
Business systems consolidate finance, orders, inventory, and product data into a single source. This is then connected to the web, portals, and EDI.
Peter Nilsson
PIM for wholesalers and who owns the product data
PIM for wholesalers starts with deciding who can change what. How to split ownership between ERP and PIM, choose stable keys and set limits on where AI can enrich product data.
Peter Nilsson
Ordering portals: how they work and what to demand
An ordering portal lets customers see their contract prices and stock levels and place orders themselves. How the order flow reaches the ERP system, when EDI is the better fit and what to demand from your vendor.
Kenneth Lyngshede