Silent failures, real costs

API Integration Stopped Working? Read the Status Code Before You Touch a Key

The website still takes the form. The thank-you page still says thanks. The lead just never reaches the CRM, and nobody notices until a salesperson asks why Tuesday was so quiet.

November 19, 2026 · 10 min read

acmeplumbingco.com · website to CRM lead sync
The lead path, checked hop by hop
Form works, CRM refuses: the key belonged to a departed user
  • PassContact form saves each submission
  • FailSubmissions accepted by the CRM401 since 09:14
  • FailAPI key still listed in CRM settingsdeleted
  • PassRate limit headroom
  • WarnWebhook address answers without a redirect301 to www
  • PassFailed leads queued for replay46 waiting
Illustrative: sample integration check on a fictional plumbing company's lead sync

The short answer

Find the failing request in the logs and read its status code. A 401 means the request lacked valid credentials, usually a rotated, deleted or revoked key. A 403 means the server knows the caller and refuses anyway. A 429 means too many requests, so pause and wait as the response says. Fix that cause, replay what failed, and check for duplicates.

On this page
  1. The integration that fails politely
  2. Three status codes, and the question each one answers
  3. The first hour, before anyone rewrites a line of code
  4. Keys rarely expire on a calendar. They get rotated, deleted or orphaned.
  5. When the other side moved: endpoints, redirects and missed events
  6. Restored the same day, once we can get in

The integration that fails politely

Most outages are loud. The site goes white, the phone rings, someone posts about it. A broken integration is quiet. The form on your website still saves the submission, so the visitor sees a success message. The order still lands in the store, so the customer gets a receipt. What stops is the handoff: the lead that should appear in the CRM, the order that should reach the warehouse system, the invoice that should post to accounting.

That makes the cost hard to see and easy to grow. Every hour the connection stays down, records pile up on one side and go missing on the other. The longer it runs, the bigger the cleanup, and the more likely someone fixes it by re-entering data by hand and creating duplicates along the way.

An integration that fails silently is not an outage. It is a leak, and leaks are measured in days.

If the trouble is that your website form itself stopped sending, start with our page on forms that stopped sending leads. If the broken link is a payment provider, read payment processing failures. This page is for the connection between two systems: your site or app on one side, a CRM, ERP, scheduling or accounting platform on the other.

Three status codes, and the question each one answers

Every API call comes back with a number. Integration logs, the vendor's developer dashboard or your own server logs will show it next to each failed request. That number tells you which side of the conversation broke, and it is the fastest way to stop guessing.

Status codes as the HTTP standards define them, read November 2026, with our practice for where to look first
What the standard saysWhere we look first
401The request "lacks valid authentication credentials for the target resource"; MDN says it means unauthenticatedWhether the key or token still exists, was rotated, or belonged to a user who was removed
403The server "understood the request but refuses to fulfill it"; MDN notes the client's identity is knownPermissions on the key, a plan change on the vendor account, or a new restriction on where the key may be used
429Too many requests "in a given amount of time"; the response "MAY include a Retry-After header"A retry loop, a bulk import, or a new automation hammering the same endpoint
5xxThe server "is aware that it has erred or is incapable of performing the requested method"The vendor's status page, or your own endpoint if the other side is calling you

Sources: the HTTP semantics standard, RFC 9110, for 401, 403 and the 5xx class; RFC 6585, section 4, for 429; and Mozilla's MDN status reference, all read in November 2026. The practical difference between the first two matters. A 401 says the vendor does not know who you are, so new credentials may fix it. A 403 says the vendor knows exactly who you are and is saying no, so a fresh key with the same permissions will fail the same way.

The first hour, before anyone rewrites a line of code

  1. 1

    Find one failing request

    Open the integration's log or the vendor's API dashboard and find a single failed call with its status code, time and endpoint. One clear example beats an afternoon of theories.

  2. 2

    Check the other side's status page

    If the vendor reports an incident, your credentials are probably fine. Leave them alone, note the time, and plan to replay the backlog when the vendor recovers.

  3. 3

    Ask what changed this week

    A password reset, a user removed from the CRM, a plan downgrade, an app reinstalled, a key rotated by IT, the website moved to a new address. Integrations break on the day something changes, and the change is usually small.

  4. 4

    Find out whose key it is

    Many integrations are connected under one person's login. When that person leaves and the account is disabled, everything authorized through it stops. Look in the vendor's settings for the connected app or key and the name next to it.

  5. 5

    Stop a retry storm

    If the log shows a wall of 429 responses, pause the job. Hammering a rate limit only extends it. If the response carries a Retry-After header, wait at least that long before trying again.

  6. 6

    Issue a new key the right way

    Create it under a service or admin account the business controls, limit it to what the integration needs, store it on the server, and delete the old one once the new one works.

  7. 7

    Replay the backlog, then hunt duplicates

    Resend the records that failed, in order, and compare counts on both sides. Records re-entered by hand during the outage are the usual source of duplicates.

Keys rarely expire on a calendar. They get rotated, deleted or orphaned.

People say the key expired. If a vendor sets expiry dates on its keys, its own documentation will say so; check it before assuming. But the Google Cloud page above, as read in November 2026, describes no expiration dates for its API keys. What it recommends is rotation: "Periodically create new API keys, update your applications to use the new API keys, and delete the old keys." It also advises you to "Retain only the API keys you are actively using." Good housekeeping, and it means a key can vanish on a Tuesday because someone was tidying up.

Glowing cyan line on a deep blue and violet grid running level, jumping in one sharp spike, then settling back to its old height
A broken integration rarely looks like an outage; it looks like a line that quietly goes flat.

The orphaned key is the other common story, and it is general advice drawn from AWS's and Google's account guidance rather than a rule either one states about integrations: keep account ownership, recovery email and phone on addresses the business controls, not on a single contractor or employee. When the integration was set up under a person who has since left, the fix is not only a new key. It is moving the connection to an account that will still exist when the next person leaves. Our page on a developer who disappeared covers getting those accounts back.

Webhook signing secrets are a related trap. Stripe's webhook documentation, read in November 2026, recommends rolling endpoint secrets "periodically, or when you suspect a compromised secret," and lets you delay the old secret's expiration "for up to 7 days" as of November 2026. Roll the secret without updating your server inside that grace window, and every event will fail signature checks the moment the old secret lapses.

When the other side moved: endpoints, redirects and missed events

Vendors retire old API versions and move endpoints. They usually announce it, in an email that went to whoever signed up for the developer account years ago. The signs are recognizable, even when the announcement was missed.

  • Calls that worked for years now return 404 or a message about an unsupported version.
  • Fields arrive empty or under new names, so records sync with blanks where phone numbers used to be.
  • Incoming webhooks stopped the day the website moved to a new domain or started forcing www or https.
  • Errors mention TLS or certificates after a server or hosting change.
  • The integration works for new records and fails for old ones, a sign that a required field changed.

The redirect case catches many businesses. Stripe's documentation says plainly: "We consider redirect responses to webhook requests as failures." Move your site, let the old webhook address redirect to the new one, and the events stop counting as delivered. The good news, also from Stripe as of November 2026, is that it "attempts to deliver events to your destination for up to three days with an exponential back off in live mode," its Dashboard resend "works for up to 15 days after the event creation," and its CLI resend "works for up to 30 days." A connection fixed early lets most missed events arrive on their own. Stripe also warns that endpoints "might occasionally receive the same event more than once," so the replay needs a check for event IDs already processed.

Restored the same day, once we can get in

Integration repair is the one emergency where we can name a day: connections to APIs and integrations are restored the same day once access is in hand. We respond the same business day, during the hours on our contact page. Access means admin rights on both systems, the integration's logs, and permission to create or rotate keys. Gather those while you wait and the work moves fast.

Fix it yourself if

  • The vendor's status page shows an incident and your code has not changed
  • The log shows 401s and you know which key was rotated or deleted
  • A built-in connector has a reconnect button and it works
  • The backlog is small enough to check by hand

Call us if

  • Nobody knows where the integration runs or who built it
  • The errors are 403s and the vendor says nothing changed
  • Hundreds of records are stuck, or duplicates are already appearing
  • Money, orders or patient data move through the connection

Emergency integration work is part of our emergency software repair service: from $575, depending on complexity, with a written estimate before work starts, as the emergency pricing page explains. If the fix turns into a better connection rather than a patched one, that is our API integrations work, and connections that kick off other tasks fall under workflow automation. If the database behind the integration is the real casualty, read the database emergency page; if the whole app is down, the web app triage page. Businesses with no one to own these systems day to day often bring in a fractional CTO. DataCram is a San Antonio web studio, building since 1999 (about), and the emergency repair hub lists every other emergency we handle.

Questions owners ask in an emergency

Something almost always changed, just not in the integration itself. A password was reset, a user was removed from the CRM, a plan was downgraded, a key was rotated, the vendor retired an old endpoint, or the website moved and now redirects. The status code in the failed request narrows it down: 401 points at credentials, 403 at permissions, 429 at volume.

A 401 means the request lacked valid authentication credentials, so the API does not know who is calling. New or corrected credentials often fix it. A 403 means the server understood the request and refuses it; MDN notes that the client's identity is known. A fresh key with the same permissions will usually fail the same way, so look at permissions and plan limits instead.

Stop sending requests for a while. RFC 6585 defines 429 as too many requests in a given amount of time, and the response may include a Retry-After header saying how long to wait. Pause the job, find what is sending the burst, such as a retry loop or a bulk import, and resume more slowly once the wait has passed.

Check your vendor's own documentation, because practices differ. Google Cloud's API key best practices page, as read in November 2026, describes none; it recommends creating new keys periodically, updating applications to use them, and deleting the old ones. In practice, a key that suddenly fails has often been rotated, deleted, restricted, or tied to a user account that was removed.

In live mode, Stripe's documentation says it attempts delivery for up to three days with an exponential back off. As of November 2026, the Dashboard resend works for up to 15 days after the event was created, and the Stripe CLI resend for up to 30 days. Stripe can deliver the same event more than once, so log processed event IDs and skip repeats.

We respond the same business day, and connections to APIs and integrations are restored the same day once we have access to both systems and their logs. The work is priced from $575, depending on complexity, with a written estimate before anything starts. Gathering admin logins and a sample failed request while you wait is the fastest way to help.

Keep reading

Free · No obligation

Get the two systems talking again

Tell us which systems are connected, what the logs say and when it stopped. We reply the same business day, and once we have access, we restore the connection the same day.

Which of these sounds like you? (pick any)

Rather talk now? Call (210) 346-0848.

CallGet a quote