Reference scenarioSplunk Observability CloudRUMTechnical · 3 min read
Users say the website is slow, but the back end looks fine.
Written by Fernando Mateo Ruiz · Splunk Enterprise Certified Architect · Splunk O11y Cloud Certified Metrics User · Reviewed on 28 October 2026
Reference scenario: it describes a common situation and how we tackle it. It is not a client project and includes no measured results.
In 30 seconds
The problem: Server dashboards show normal response times, yet users complain the site is slow and nobody can reproduce it.
What solves it: Real User Monitoring (RUM) in the browser, linked to back-end traces in APM, to see where the time goes: the browser, the network or a specific service.
What changes: Complaints become data: which page, browser, region and request is slow, and why.
Sound familiar?
- The server answers in 200 ms but the page takes seconds to become usable.
- It only happens to some users, regions or browsers.
- Front-end and back-end teams blame each other.
- Nobody knows how the last deployment affected real users.
The question to answer
For real users, which pages are slow, for whom, and is the time spent in the browser, the network or a back-end service?
Why it matters
Server metrics measure your systems, not your users. Without the browser's view, the problem that costs you sales stays invisible.
Signals needed
- EssentialRUM browser sessions and spans
- EssentialBack-end APM traces
- RecommendedWeb vitals
- OptionalSynthetic tests
Architecture
Reference scenario: the architecture is generic and does not belong to any client.
How it’s implemented
- Add the Splunk RUM browser agent (
splunk-otel-web) to the application, with an environment name that matches the one used in APM. - Instrument the back end with the Splunk distribution of OpenTelemetry. With
SPLUNK_TRACE_RESPONSE_HEADER_ENABLED=true, responses include theServer-Timingheader that RUM uses to link each request to its APM trace. - If the front end and the API are on different domains, review the cross-origin (CORS) configuration so the browser can read that header. The official RUM documentation lists what's needed.
- Review load times and web vitals by page, browser and region, and create detectors for the pages that matter.
OTEL_SERVICE_NAME=checkout-api
OTEL_RESOURCE_ATTRIBUTES=deployment.environment=prod
# Adds the Server-Timing header so RUM can link to the APM trace
SPLUNK_TRACE_RESPONSE_HEADER_ENABLED=true
Step-by-step investigation
An illustrative example of how it would be investigated, from signal to action.
- 01 SignalIn RUM, the load time of the basket page triples for logged-in users.
- 02 ContextThe slow part is a single request from the page:
/api/cart/summary. - 03 CorrelationIts APM link opens the trace: 2.8 seconds waiting for the recommendations service.
- 04 CauseRecommendations calls an external API with no timeout, only when the basket has items.
- 05 DecisionThe page shouldn't wait for recommendations: they're secondary.
- 06 ActionAdd a timeout, load that block after the page and create a detector on that endpoint's latency.
Common mistakes
- Measuring only the server and assuming that's the user's experience.
- Using different environment names in RUM and APM.
- Forgetting the cross-origin configuration when the API is on another domain.
- Collecting personal data in URLs or attributes without reviewing privacy settings.
What this case doesn’t solve
- It doesn't replace usability testing.
- It only sees users whose browsers load the agent.
Expected result
For each slow page you know which users are affected and whether the time is in the browser, the network or a specific back-end service.
Observed result
This is a reference scenario: we don’t publish measured results until it has been tested in our lab or in an authorised project.
Requirements
Platform
Splunk Observability Cloud with RUM and APM
Instrumentation
RUM browser agent and OpenTelemetry in the back end
Privacy
A review of the data collected from browsers
Related services
Questions we’d ask you
- Where do complaints about slowness come from today, and how do you check them?
- Are the front end and the API served from the same domain?
- Is the back end already instrumented with OpenTelemetry?