A material study in bright daylight. A closed ring of clear turquoise glass, its surface combed into fine parallel grooves, with pearl silver bands spiralling around it, resting on a pale reflective floor.
  1. Home
  2. Insights
  3. Care and Optimization
  4. The year after launch

Decision guide

The year aftera site launches.

A website is not a delivered object. Over twelve months its dependencies age, its content goes out of date, its connected systems change at the other end, its forms fail silently and its performance drifts. Almost none of that is visible from the front page, which is why most of it is found by a customer.

The table below ranks nine failure modes by how long each one typically takes to notice rather than by severity, because time to notice is what decides the cost. Written 23 September 2026.

HUREAL / Material studies

The inversion

The expensive failuresare the quiet ones.

A site that goes down is fixed within the hour, because everybody knows immediately. It is loud, it is embarrassing, and it is therefore cheap. A contact form that stops delivering is discovered about six weeks later, by somebody in sales wondering why it has been quiet, and by then the cost is a month and a half of inbound enquiries that were received, accepted, and thrown away.

That is the inversion that makes the first year after launch expensive. Ranking maintenance work by severity puts uptime at the top, where it is already handled. Ranking it by how long the failure takes to notice puts the things nobody watches at the top, which is where the actual losses are.

Everything in this article follows from that. The most valuable maintenance is not the work that fixes serious problems. It is the work that shortens the gap between something breaking and somebody knowing.

What actually changes

Nine things,ranked by time to notice.

What changes under a site in twelve months, and how long each takes to be noticed
What changes Why it happens How it is usually found
Found immediatelyBecause they are loud
The certificate expires Certificates have finite lives and renewal is automated until the day it is not. Every visitor is warned by their browser. Public, humiliating, and fixed within the hour.
The site is unavailable A host incident, a failed deployment, a domain renewal nobody owned. Somebody notices within minutes.
Found in weeksIf something is watching
A connected system changes at the other end Every vendor at the far end of an integration ships changes on their own schedule. An alert, if one exists. Otherwise a person noticing that records stopped arriving.
Analytics stops recording A tag removed by a redeployment, a consent configuration change, a property that was reconfigured. A flat line in a monthly report, a month later.
The form stops delivering A mail provider policy change, a spam rule, a full mailbox, or a recipient address belonging to somebody who left. A salesperson observing that it has gone quiet. Typically six weeks.
Found by accident, or neverNothing is watching at all
Dependencies accumulate security advisories Upstream maintainers publish them continuously, for every library a site is built on. An automated alert if one was configured at launch. Otherwise nothing, until it matters.
Redirects rot A URL is changed, and an old redirect now points at a page that no longer exists. A crawl. There is no other way, and nobody crawls their own site by accident.
Performance drifts Every image, script, tag and embed added since launch, none of which individually seemed significant. A measurement against a recorded baseline. Without a baseline, it is unfalsifiable.
Accessibility drifts Every page published since launch by somebody who was not trained, plus every component added since. A barrier report, which means somebody was excluded first.

Scroll the table sideways to read it.

The pattern in the third group is that each one is invisible from the front page. A site with a year of unapplied security advisories, a broken redirect map, a doubled page weight and an accessibility regression looks exactly like a site with none of those, to anybody who opens it.

The expensive one

A form that stopsis the worst of these.

It is worth separating out because it fails more ways than anything else on the list and because every one of those ways is silent. The visitor sees a confirmation. The site logs a submission. Nothing in any dashboard indicates a problem. The only symptom is an absence, and absence is precisely what nobody monitors.

  • The recipient left the company

    An address that used to belong to a person now bounces, or worse, silently discards. This is the most common single cause and it is entirely organisational: nobody told the website when somebody left.

    Send to a group address that survives a departure, never to a person.

  • The mail provider tightened its rules

    Authentication requirements for sending domains have tightened repeatedly, and a notification sent from a server that is not properly authorised for the domain starts being filtered rather than rejected.

    Filtered mail produces no error anywhere. It just does not arrive.

  • The notification is being classified as spam

    Frequently by your own mail system, and frequently after an unrelated policy change by whoever administers it.

    Check the junk folder before concluding it was a quiet month.

  • The connected system stopped accepting the record

    A required field was added in the CRM, a credential expired, or a limit was hit. The submission is captured on the site and never arrives where anybody works.

    This is the one that also loses the record, not just the notification.

The prevention is a scheduled submission that goes through the whole path, end to end, from outside the network, and an alert when it does not complete. It is a small amount of work, it is cheaper than one lost month, and the number of sites that have it is considerably smaller than the number that should. If nothing else in this article gets implemented, implement this.

Content

Pages age relativeto what was published since.

Content does not decay on its own. It decays relative to everything published on the same subject since, which is why a page that was the best available answer two years ago is now the fourth best without a word of it having changed.

That makes it measurable rather than a matter of opinion. Compare search impressions and clicks per page across two comparable quarters, allowing for seasonality in businesses that have it. The pages that fell are the review list, and it is a short list with a reason attached to each item, which is what makes it something a marketing team will actually work through.

What to do with each one is a choice between three actions. Update it, where the subject is still right and the content is behind. Merge it into a stronger page, where two pages are competing for the same question and dividing the signals. Or retire it, which is the one with a rule attached.

The rule about deleting pages, and it has no exceptions

Nothing is deleted without a redirect. Deleting a page that ranks is the fastest available way to lose traffic you already had, and it is done routinely during content clear-outs because the page looked old in a list.

Every retired address points at the closest genuinely relevant page. Not the homepage, which is a redirect that satisfies a crawler and tells a visitor that what they wanted no longer exists. The redirect map is a document that lives as long as the site does, and a quarterly crawl is what keeps it honest.

The arrangement

What a monthlyarrangement should list.

Read the list rather than the title. A maintenance plan is a set of specific recurring actions, and where the description is a category rather than a list of checks, there is no way to tell whether any of it happens.

  • Dependency and security updates, applied and tested

    Applied, not reported. An advisory list with nothing done about it is a document rather than maintenance.

  • Uptime and certificate monitoring, alerting a named person

    Named, not a shared inbox. The difference between an hour and a week is entirely in who receives it.

  • A scheduled end to end form delivery test

    Through the whole path, from outside, with an alert when it does not complete.

  • A heartbeat on every integration

    Something that expects each connection to run and complains when it does not, because a connection that fails by never starting produces no errors.

  • A quarterly crawl

    For broken links, redirect chains, orphan pages, missing titles and anything returning the wrong status code.

  • Performance measured against a recorded baseline

    From a real device profile on a real connection. A number with nothing to compare it to is not a measurement.

  • An accessibility pass on what was published since the last one

    Keyboard, screen reader, contrast and headings, on the new pages rather than on the whole site every time.

  • Content changes, publishing, and the review list

    The actual work, driven by the quarterly comparison rather than by whoever asks loudest.

  • A report saying what changed and what it produced

    Short, and written so somebody who does not work in software can read it.

What an arrangement does not cover, and should say so. New features, a redesign, a new integration and anything that is a project rather than an upkeep are priced separately, and a plan that implies otherwise will produce an argument in month four. It should also be monthly and cancellable rather than annual, because an annual lock is a commitment made at the point of least information.

This month

Six thingsin the first month.

Whether or not anybody is on a monthly arrangement, six things are worth doing in the thirty days after a launch, and all six are cheap. Five of them are the baselines that make every later measurement mean something, which is why they are worth doing before anything drifts rather than after.

  1. Submit the form yourself, from outside the building

    On a phone, on a cellular connection, not signed in to anything. Then check that the notification arrived, that the record landed in the CRM, and that the acknowledgement the visitor receives is the one you intended to send.

  2. Confirm the analytics is recording the thing that matters

    Not page views. The conversion. Submit a test enquiry and find it in the reporting, because a tracking configuration that was correct in staging is not evidence about production.

  3. Crawl the live site against the redirect map

    Every address that was supposed to redirect, redirecting, once, to a page that returns successfully. Chains and loops are introduced during the last week of a launch more often than at any other time.

  4. Turn on dependency alerts

    A few minutes of configuration in the code repository, and it is the difference between finding out about a security advisory the week it is published and finding out about it from somebody else.

  5. Record a performance baseline

    Measured from a mid range device on a cellular connection, written down with the date. Without it, every future conversation about whether the site got slower is a matter of opinion.

  6. Put the first content review in the calendar

    Ninety days out, with a named owner, and with the quarterly comparison as its agenda. A review that is not in a calendar is a review that happens during the next redesign.

One thing sits behind all six and no supplier can provide it. One named person inside the business owns the site, with the authority to say that a page is wrong and to have it changed. A supplier can do every item on this page and cannot decide that the opening hours are out of date, that the service you stopped offering is still listed, or that the person on the team page left in March. A site with no internal owner ages regardless of who is maintaining it.

Questions

Questions peopleactually ask.

  • Do we actually need a monthly arrangement?

    You need the work done. Whether it is done by a supplier on a monthly arrangement, by somebody internal with the time and the access, or by a mixture, is a commercial decision. What does not work is nobody doing it, because none of the nine things in this article stops happening when the invoices stop.

  • What happens if we stop?

    The site keeps working, and it degrades on the timescale of the slowest item rather than the fastest. Security advisories accumulate, the content ages, an integration breaks at the other end and stays broken, and the first real signal is usually a customer or a quiet month in the pipeline. The arrangement should be monthly and cancellable precisely so that this is your decision rather than a negotiation.

  • Who should own the site inside our company?

    One named person with the authority to say that a page is wrong and to have it changed. Not a committee, and not the person who happens to have the login. A supplier can do all of the work in this article and cannot decide that the hours on the contact page are out of date, and that decision is the one that ages a site fastest.

  • How often should content be reviewed?

    Anything with a price, a person, an opening hour or a legal statement on it belongs on a quarterly list. Everything else is better driven by measurement: compare two comparable quarters of search impressions per page, and the pages that fell are the review list. That turns an unbounded task into a short one with a reason attached to each item.

  • Is a maintenance plan just hosting with a different name?

    They are separate things and they are frequently sold together, which is where the confusion comes from. Hosting keeps the site served. Maintenance is the dependency updates, the monitoring, the delivery tests, the crawls, the measurement and the changes. A plan that lists uptime and nothing else is hosting, and it is worth reading the list rather than the title.

Read next

Three that followfrom this one.

The service behind this article

Monthly,and cancellable.

The work in this article is what Care and Optimization is, and the arrangement is monthly and cancellable rather than annual. Stopping should be a decision you can make in a month, and the system stays yours and keeps running either way.

A performance audit reports where a site is now, before any arrangement exists. Dependency and certificate status, whether the forms deliver, what the connected systems are returning, what an indexer and an answer engine can read, and the measured performance from a real device profile. It is yours whether or not anything follows it.

Request a performance audit

More on this subject in the Care and Optimization index, and everything else at Insights.

A material study photographed in bright daylight. A tall fan of clear turquoise glass fins rises on the right of the frame with a polished pearl silver ribbon curving through it, standing in a shallow film of still water. The left of the frame is empty pale mint.

Software you own from day one.

HUREAL / Material studies