Skip to main content
ComponentsEngineRUM JS
Availability2026.3 or later

What GermainUX Watches

Sessions, errors and releases per customer

Find the Cause

One customer to every customer

Alert and Act

Catch it before the next complaint

Set It Up

Portal, identity and MCP Server

Overview

A customer writes to support: “I tried three times to update my card on your portal, and it keeps saying something went wrong.” The support agent can’t see what the customer saw. They pass the case to engineering, engineering can’t reproduce it, and the case goes back and forth until the customer gives up. The case has the customer’s words, not what the application did. It doesn’t say which request failed, since when, or whether anyone else hit it. Without that evidence, a real bug looks like a one-off, and the customers who failed and didn’t complain stay invisible. GermainUX records what every signed-in customer does and runs into on your portal: clicks, the requests behind them, the error messages they see, rage clicks, feedback and releases. With the GermainUX MCP Server, the support agent asks an AI assistant such as Claude about that customer in plain language. The assistant finds the session, reads what happened, checks how many other customers hit the same error and which release started it, and links to the evidence in GermainUX. It then drafts the engineering ticket with that evidence, ready to file in Jira.
Customer ComplaintTheir SessionThe ErrorWho ElseWhat ChangedJira Ticket
This page uses a customer account portal, Claude and Jira as the example. The same approach works for any web or mobile application GermainUX monitors, any AI client that connects to the MCP Server, and any ticketing tool. When the question starts from a rise in cases rather than one customer, see Portal Case-Volume Analysis.

Questions It Answers

  • What happened to account 100482 when they tried to update their card?
  • Is it only this customer, or did others hit the same error?
  • Which customers are affected, and what do they have in common?
  • Since when, and did a release change something?
  • What should the engineering ticket say, and where is the evidence?
  • Is it fixed?

What GermainUX Watches

The assistant answers from the same KPIs the Workspace shows. Each signal is tied to the customer and the session it came from.

Find the Cause

The investigation runs from one customer out to everyone, in one conversation with the assistant:
  1. Find the customer’s session by their account number, and read its story: what they did, which request failed and which message they saw.
  2. Query the failing request for every customer over the same days. The assistant groups the failures by country, browser, page or any other field, and finds what the affected customers have in common.
  3. Look for a release that started it, and rule out what stayed normal.
  4. Ask for links to the evidence in GermainUX: the session replay, the analysis and the list of failures.
  5. Have the assistant draft the ticket with the evidence, file it in your ticketing tool, and ask again after the fix to confirm it worked.
See Example Prompts for how the assistant uses the MCP tools, and Available Tools for the full list. The example below comes from simulated data, and its numbers belong to the example.

Example: A Release Rejects Canadian and UK Postal Codes

At 09:00 UTC on a weekday, a portal release adds postal-code validation to the payment-method API. It only accepts US ZIP codes, so every card that a Canadian or UK customer saves gets an error. The portal shows “Something went wrong. Please try again.” That afternoon, account 100482 in Toronto tries three times, and the next morning their complaint reaches support.
1

Find the customer's session

Ask: “Account 100482 says they couldn’t update their card on Monday, Nov 16. Walk me through their session, with times in UTC, and include a link to the session replay.”The assistant finds the customer’s sessions by account number (query_kpi), then reads the one from 18:40 (view_session_story). The customer opened Payment Methods, added a card and clicked Save Card three times. Each time, the request to save the card returned a 422 and the portal showed “Something went wrong. Please try again.” Between the second and third try, they rage-clicked Save Card, and before leaving they gave negative feedback. The assistant adds a link to the session in Session Replay (link_to_session_replay).
Claude answer with a timeline of account 100482's session on Nov 16 in UTC: log in, Payment Methods, Add Card, three Save Card tries each returning 422 Unprocessable Entity with the message Something went wrong, a rage click, negative feedback, and a link to the session replay

One Customer's Session, Told by the AI Assistant - Claude

2

Check who else

Ask: “Did other Customer Account customers get the same error between the Nov 16 09:00 UTC release and Nov 17 08:00 UTC? What do they have in common?”The assistant queries the same request for every customer from the release to the next morning (query_kpi) and groups the failures by country (get_field_values). In this example, 33 card saves failed between 09:00 and the next morning, against 2 in the same hours on the previous days. Canada has 25 of them, from 9 customers, and the United Kingdom 7, from 5 customers. The one other failure is a US customer who mistyped the card number. Account 100482 is one of 14 customers who couldn’t save a card. Most of them tried two or three times, and every one saw the same message.The same comparison is in the Workspace. In Aggregate Analysis of the failed card-save requests, Leading Factors rank the United Kingdom and Canada by country, and Montreal, Toronto and Calgary by city. Chrome 141 and Mobile Safari 18 rank too, only because most customers use them.
Analysis of failed card-save requests from the release to the next morning: 33 failures against 2 in the same hours on the previous days, with Leading Factors ranking the United Kingdom with 7 failures and Canada with 25 by client country, and Montreal, Toronto and Calgary by city

Failed Card Saves Analysis - GermainUX

3

Find what changed

Ask: “When did it start, and was there a Customer Account release right before? Stay within the same window.”The first failure is at 09:04 UTC, a few minutes after the release starts. The assistant finds two deployments that started at 09:00 (query_kpi on Deployment Run): Customer Account release 2026.11.4 and an update of the payments-papi API. The API was rolled back at 10:15 and the failures carried on, so the Customer Account release is the change still in place.
Claude answer: the first failure at 09:04:51 UTC on Nov 16, with two deployments that started at 09:00, Customer Account and payments-papi, and the first failure about 1.5 minutes after the Customer Account deploy finished

Two Deployments Before the First Failure - Claude

In the Workspace, the release shows as a marker on the failed card saves. The failures start at the marker of release 2026.11.4 and stop at the marker of the fix, 2026.11.5, the next afternoon.
Failed card saves analysis with the Relationships chart from Nov 1 to Nov 20: Deployment Run markers for the releases on Nov 12, Nov 16 and Nov 17, and failures only between release 2026.11.4 on Nov 16 and the fix 2026.11.5 on Nov 17

Failed Card Saves Between the Release and the Fix - GermainUX

Hover the 2026.11.4 marker to see the release’s deployment ID, DEP-4117, and the files it changed.
Failed card saves analysis with the Nov 16 release marker hovered: Deployment Run Customer Account 2026.11.4 with deployment ID DEP-4117, status COMPLETED, application Customer Account, and details listing PaymentMethodController.java, AddressValidator.java and CardForm.tsx

Release 2026.11.4 Marker Details - GermainUX

Then ask: “Read the details field of the DEP-4117 deployment run.” The details list the three files the release changed: the payment-method controller, the address validator and the card form. The assistant names the address validator as the prime suspect.
Claude answer listing the files changed by deployment DEP-4117, Customer Account 2026.11.4: PaymentMethodController.java, AddressValidator.java and CardForm.tsx, with AddressValidator.java as the prime suspect

What the Release Changed - Claude

4

Rule out the rest

Ask: “Is anything else broken for these customers in that window?”In this example, everything else stays at its usual level. Canadian and UK customers log in, view invoices, update their profile and pay with a card already on file as usual. Response times match the day before: the card save fails, it doesn’t time out. The only change is the card save, for customers outside the US, since the release.
Claude answer: saving a card is the only thing broken for Canadian and UK customers, with 32 card saves rejected across 14 customers, while every other API call succeeded, response times matched the day before, and no crashes, freezes or dead clicks occurred

Nothing Else Broken - Claude

5

Draft the ticket

Ask: “Write a Jira bug for the Billing team with what we found, show it here ready to paste, and include the links.”The assistant gets the links from GermainUX (link_to_session_replay, link_to_analysis, link_to_drillthrough) and writes the bug, ready to paste into Jira. It names the release, the request and its 422 response, the countries affected, the number of customers, and the steps to reproduce with a Canadian postal code. Engineering doesn’t need to reproduce anything to start.The support agent files it in Jira. To keep the evidence in GermainUX too, they can add it as a Note on the customer’s session or on the analysis, assigned to the engineering team.
6

Confirm the fix

After the fix is released at 14:00 UTC the next day, ask: “Are card saves working again outside the US?”In this example, Canadian and UK card saves all succeed after the fix. Of the 22 customers affected over the 29-hour outage, 13 have saved a card since, including account 100482, and support can contact the 9 who haven’t tried again.
Claude answer: card saves work again outside the US after Customer Account 2026.11.5 (DEP-4121) at 14:00 UTC on Nov 17. Counted by session, Canada goes from 0 saved and 13 failed to 19 saved and 0 failed, the United Kingdom from 0 saved and 9 failed to 13 saved and 0 failed, and the United States stays at its usual level. Of 22 affected Canadian and UK customers, 13 have saved a card since, including account 100482, and 9 haven't tried again

Card Saves Before and After the Fix - Claude

In the Workspace, the failures stop at the 2026.11.5 marker on the failed card saves chart above. Hover the marker to see the fix: deployment DEP-4121 changed the address validator and its test.
Failed card saves analysis with the Nov 17 release marker hovered: Deployment Run Customer Account 2026.11.5 at 14:00 UTC with deployment ID DEP-4121, status COMPLETED, and details listing AddressValidator.java and AddressValidatorTest.java, with no failed card saves after the marker

Release 2026.11.5 Marker Details - GermainUX

The support agent replies to the customer with what happened and that it’s fixed. The engineering team gets a ticket it can act on, and the list of the other affected customers lets support reach out to them before they complain.

Alert and Act

Most affected customers never complain, which is the reason to catch this sooner:
  • SLAs and outlier detection on the error rate of key requests, such as saving a card, by country, so a release that breaks one segment raises an alert
  • Alerts when user-facing errors or rage clicks rise on a page, by email, Microsoft Teams or Slack
  • Synthetic user scenarios that save a card with a Canadian and a UK test account right after each release
  • Automated actions that roll back the release when the card-save SLA fires, with approval before they run
  • Percent change SLAs on negative feedback, so a rise in complaints on one page reaches the product team the same day

Set It Up

Each step adds value on its own. The assistant can answer more questions the more of the portal GermainUX monitors.
1

Monitor the portal

Deploy RUM JS on the portal so GermainUX records sessions, clicks, requests, errors and feedback. See Browser Monitoring for the other platforms GermainUX monitors.
2

Identify the customer

Configure a username extractor that sets the identifier your support team uses, such as the account number, on every session. Without it, the assistant can’t find a customer’s session. Review data privacy for what the sessions may contain.
3

Show releases on the timeline

Bring your portal releases into GermainUX as deployment events. The GermainUX team can help you connect your release pipeline. Then configure KPI relationships from your request and error KPIs to them, shown as markers. Give each relationship a constraint such as application.name == "Customer Account", or the releases of your other applications show on the portal’s charts too.
4

Enable the MCP Server for support

Enable the MCP Server and give the support agents a GermainUX account with access to the portal’s data. Every tool call runs as the signed-in user, so the agents see only what their account allows. See What users can see and do.
5

Connect the AI client

Connect the AI client the support agents use to the GermainUX MCP Server. The assistant writes the ticket as text, so it works with Jira or any other ticketing tool.

Additional Resources

MCP Server

Connect AI assistants to GermainUX

Analysis Dashboard

Aggregate and instance analysis

Use Cases

More Use Cases

Get Help

Every support team has its own portal, customer identifiers and ticketing workflow. The GermainUX team can help you identify customers in their sessions, bring your releases onto the timeline, set up the MCP Server for your support agents, and agree what a good engineering ticket includes.

Contact GermainUX Support

The GermainUX team helps you get up and running faster, empower your team, turn data into actionable insights, and optimize performance to deliver measurable business outcomes.

Can't Find What You Need?

Search all GermainUX docs with Ctrl + K, or use the feedback buttons at the bottom of this page to tell us what is missing.

The content of this document is proprietary to GERMAIN SOFTWARE.