User journey SLO: measure what customers experience with RUM and synthetic

User journey SLO in Dynatrace: combine RUM and synthetic monitoring to measure what customers experience, then build the SLO and alert, SaaS and Managed.
SLO parcours utilisateur : mesurer ce que vit le client avec RUM et synthétique

All your services show 99.95% success, yet support keeps receiving complaints: payment fails in the browser, a third-party script blocks the “Confirm” button. A service SLO measures what the server sees; a user journey SLO measures what the customer experiences. This tutorial builds that second kind of objective by combining Real User Monitoring (RUM) and synthetic monitoring.

Goal: by the end you will have defined a critical journey (for example “search → basket → payment”), chosen two complementary SLIs (synthetic availability and success of real user actions), created the SLO and wired an alert. The tutorial covers Dynatrace SaaS (SLOs on Grail, DQL queries) and Dynatrace Managed (classic SLOs on metric expressions). RUM and synthetic exist on both; only the way you write the SLO changes.

For the general method of choosing indicators, see choosing your SLIs with DQL.

The problem: the server does not see everything

A purchase journey crosses the CDN, the browser, third-party JavaScript, then several services. JavaScript errors, slow DNS or a blocked resource never show up in the services’ error rate. The result: MTTR grows, because on-call searches the backend first, and the “official” error budget stays full while customers leave.

Prerequisites

  • A web application instrumented with RUM (automatic injection by OneAgent or a JavaScript snippet).
  • Key user actions marked for the journey steps. The exact menu label varies by version: see docs.dynatrace.com.
  • A synthetic browser monitor (browser clickpath) replaying the journey from at least two locations.
  • SaaS: access to Notebooks and to the SLO app. Managed: rights to configure SLOs in the environment.

Step 1: describe the journey in one sentence

Write the intent before the technique: “a signed-in customer can search for a product, add it to the basket and pay in under 10 seconds in total”. This sentence sets three things: the steps (hence the key actions), the success criterion (no error), and a latency criterion. Stick to three to five steps: beyond that, the SLO becomes impossible to diagnose.

Step 2: choose two complementary SLIs

SLI Source What it captures Blind spot
Journey availability Synthetic browser Full outage, even at night with no traffic Real diversity of browsers and networks
Key action success RUM Errors experienced by real customers Nothing to measure when there is no traffic

The two complement each other: synthetic gives a constant signal, RUM gives the signal that really matters. Create two separate SLOs rather than a weighted average, which is unreadable during an incident.

Step 3: create the user journey SLO on Dynatrace SaaS

On SaaS, RUM and synthetic data can be queried in Grail, but metric and field names depend on the RUM generation enabled in your environment. Start by finding them in a Notebook:

fetch dt.system.data_objects
| filter contains(name, "synthetic") or contains(name, "web")
| fields name, type

If this table is not available to you, use the metrics browser and filter on “synthetic” or “apps.web”. Once you have identified the synthetic availability metric, the SLI follows this pattern (replace AVAILABILITY_METRIC_NAME and the monitor ID):

timeseries dispo = avg(AVAILABILITY_METRIC_NAME),
  filter: { in(dt.entity.synthetic_test, "SYNTHETIC_TEST-0123456789ABCDEF") }
| fieldsAdd sli = dispo[]

For the RUM side, think in ratios: key actions without errors divided by total key actions. If the per-action error metric is not available in your version, a fallback SLI is to measure the backend services the journey calls, using Grail service metrics:

timeseries {
    total    = sum(dt.service.request.count),
    failures = sum(dt.service.request.failure_count)
  },
  filter: { in(dt.entity.service, "SERVICE-PANIER", "SERVICE-PAIEMENT") }
| fieldsAdd sli = 100 * (total[] - failures[]) / total[]

This fallback remains a server-side SLI: document it as such, and replace it as soon as the RUM SLI is in place. Then create the SLOs in the app, with a realistic target (for example 99.5% for synthetic over 30 days) and the output field the custom template expects.

Step 4: the same SLO on Dynatrace Managed

On Managed, create a classic SLO (Settings or the Service-level objectives page depending on the version) with a metric expression. For synthetic browser monitors, a per-location availability metric exists in the builtin:synthetic.browser.* family; an example expression:

builtin:synthetic.browser.availability.location.total
  :filter(eq("dt.entity.synthetic_test","SYNTHETIC_TEST-0123456789ABCDEF"))
  :splitBy()

Check the exact metric name and its dimensions in your environment’s data explorer before saving: keys can differ between versions. For RUM, web application metrics live under builtin:apps.web.*; pick those that count actions and errors, filtered on your key actions.

Step 5: alert on budget consumption

Set burn rate alerts rather than fixed thresholds: a fast alert for a hard outage detected by synthetic, a slow alert for gradual degradation visible in RUM. The multi-window method is detailed in SLO burn rate alerts in Dynatrace. Beware of RUM at night: with little traffic, a handful of errors is enough to make the burn rate explode. Require a minimum volume of actions before alerting.

How to check that it works

  1. The synthetic monitor shows successful executions from every location, and the matching SLO shows a value close to 100%.
  2. In a Notebook (SaaS) or the data explorer (Managed), the SLI query returns a series with no gaps over the last 24 hours.
  3. Deliberately break a step in pre-production (hidden button, endpoint returning 500): synthetic fails, the SLO drops and the fast alert fires.
  4. Replay a failed session with Session Replay to confirm the SLI really reflects the experience: see Session Replay on Grail.

Best practices

  • One journey, one owner: the product team owning the journey owns the SLO, not the team maintaining the monitor.
  • Few steps: three to five key actions, not the whole navigation.
  • Exclude noise: filter out bots and test traffic in the RUM SLI.
  • Read both SLOs together: a gap between synthetic (green) and RUM (red) often points to a browser, a region or a third-party script.
  • Make it visible: show these SLOs next to service SLOs, as in the SLO dashboard for leadership.

Frequently asked questions

Should a journey SLO use RUM or synthetic?

Both. Synthetic detects an outage even with no traffic; RUM measures the real experience. Either one alone leaves a significant blind spot.

What target should a user journey SLO have?

Start from the value observed over the last 30 days, slightly below it, then adjust with the product team. A target higher than the underlying services’ targets is unrealistic, because the journey accumulates their failures.

It shrinks the sample without making it useless. Measure a ratio rather than a volume, and check that the sample stays large enough during quiet hours.

Conclusion

A user journey SLO brings reliability back to what customers actually experience. Combining synthetic and RUM covers both hard outages and quiet degradation, and steers on-call to the right place from the first minute.

span.observer is an independent community blog, not affiliated with Dynatrace LLC. Dynatrace is a trademark of its owner. SaaS features evolve quickly: check their availability in your environment and in the official documentation.

Total
0
Shares
Leave a Reply

Your email address will not be published. Required fields are marked *

Previous Post
Migrer SLO Grail : passer des SLO classiques aux nouveaux SLO Dynatrace

Migrate SLOs to Grail: moving from classic SLOs to the new Dynatrace SLOs

Next Post
Postmortem sans blâme avec Dynatrace : modèle de document et requêtes DQL

Blameless postmortem with Dynatrace: a document template and DQL queries

Related Posts