product docs integrations use-cases changelog pricing Get API Key
// use-cases · one engine, six teams

Built for how your team
actually ships.

Growth tests prices. Engineering gates releases. Data joins assignments to the warehouse. Same API, same statistics, different levers.

Talk to an Engineer Read the Docs
crawltic --team growth
// growth · pricing-page-annual

Test price framing against revenue, not clicks.

Annual-first vs monthly-first plan display, measured on revenue per visitor and 30-day churn. The winner ships itself once both clear.

Rev / visitor
+12.7%
P(B>A)
97.4%
// engineering · search-v2 rollout

Ramp a release like an experiment.

1% → 5% → 25% → 50% → 100%, with p99 latency and error rate as guardrails. A regression rolls it back before a pager goes off.

p99 latency
142ms
Error rate
0.21%
// product · onboarding-v3

Redesign a flow, measure every step.

Each funnel step is its own metric. Crawltic reports where variant B wins, where it leaks, and whether the lift survives to activation.

Activation
+19.0pt
Step leak
8.2%
// data · warehouse sync

Assignments land next to your revenue.

Every assignment streams into Snowflake, BigQuery, or Redshift. Join on user id and model lift with the SQL you already trust.

Sync lag
42s
Rows / hour
1.8M
// marketing · weekly-digest-subject

Subject lines judged on revenue per send.

Open rate is a proxy. Crawltic follows each send through to purchase and picks the subject line that actually earns.

Rev / send
$0.47
Open lift
+49.1%
// platform · edge-cache-swr

Infrastructure changes are experiments too.

Cache policy, CDN routing, connection pooling. Latency, hit ratio, and cost per request are metrics like any other.

p99 · B
58ms
Hit ratio
94%
NIMBUSCLOUDPIXELFORGEHOLOSTACKQUANTALOOPVELLUMORBITALLATTICEWORKSNORTHWIND LABS NIMBUSCLOUDPIXELFORGEHOLOSTACKQUANTALOOPVELLUMORBITALLATTICEWORKSNORTHWIND LABS
// 01 · growth

Pricing tests that report in dollars.

Quantaloop split pricing-page traffic between monthly-first and annual-first framing. Crawltic scored both on revenue per visitor and 30-day churn, then promoted annual-first after 11 days.

  • Revenue, EPC, and LTV as first-class metrics
  • Stripe and Shopify order events joined automatically
  • Churn guardrail blocks a win that costs retention
pricing-page-annual.live
Visitors · A
0
Visitors · B
0
Rev / visitor · A
$4.12
Rev / visitor · B
$4.64
// 02 · engineering

Guardrails that roll back for you.

HoloStack gates every backend release behind a staged ramp. When p99 latency crosses the guardrail, Crawltic drops the variant to 0% and posts the incident, usually before anyone notices.

  • Staged ramps: 1%, 5%, 25%, 50%, 100%
  • Latency, error rate, and saturation as guardrails
  • Automatic rollback with a signed webhook to your deploy system
rollout.live — search-v2
ramping · 1%
Traffic on B
1%
p99 latency
140ms
Guardrail
p99 < 180ms
// 03 · product

Ship UX changes on your own release cycle.

PixelForge's product team tests onboarding copy, layout, and step order straight from their feature branches. Results come back through the API into the internal tools they already use.

  • Per-step funnel metrics with drop-off attribution
  • Server-side or client-side assignment, same results
  • Results queryable from Retool, Notion, or your own admin
onboarding-v3 — step 2 of 4
/onboarding/connectcontrol A
01
Signup
02
Connect
03
First test
04
Invite
// 04 · data

No analytics silo. Just rows in your warehouse.

LatticeWorks streams every assignment and event into Snowflake, then joins them with billing and product data. Their analysts model experiment impact with the same SQL and dbt models they use for everything else.

  • Native sync to Snowflake, BigQuery, Redshift, Databricks
  • Stable user keys for joins across first-party data
  • Raw events plus pre-computed daily experiment facts
warehouse.join — exp_facts
assignments
usr_4121 · B
usr_4122 · A
usr_4123 · B
usr_4124 · A
orders
usr_4121 · $84
usr_4122 · $41
usr_4123 · $112
usr_4124 · $0
exp_facts
B · rpv $98.00
A · rpv $20.50
B · n=2 · conv 1.0
A · n=2 · conv 0.5
select a.variant, avg(o.amount) as rpv
from crawltic.assignments a left join billing.orders o using (user_id)
where a.experiment = 'checkout-button' group by 1;
// 05 · marketing

Landing pages and subject lines, scored on revenue.

Vellum's lifecycle team runs every campaign as an experiment. Crawltic follows each send and each landing-page visit through to purchase, and the Bayesian posterior shows how sure they can be.

  • Email, landing page, and ad creative in one engine
  • Down-funnel revenue attribution per variant
  • Holdout groups to measure true campaign lift
posterior.live — revenue per send
A · control24,100 sends
Your weekly experiment digest
open 21.4%
B · variant24,100 sends
3 tests just hit significance
open 31.9%
P(B beats A)
50.0%
Sends analysed
400
// 06 · platform

Latency and cache hit ratio are metrics like any other.

Orbital's infra team tested stale-while-revalidate against a short TTL across their edge fleet. Variant B held a 94% hit ratio and cut p99 by 73%, verified with the same rigor as a product test.

  • Run experiments at the edge, in CI, or in the service mesh
  • Latency percentiles, error rate, and cost per request
  • Scripted end to end from the CLI
edge-cache-swr.live — p99 ms
B · stale-while-revalidateA · ttl 60s
Cache hit · B
94.2%
p99 · A
212ms
p99 · B
58ms
End-to-end example

A checkout button test, start to ship

Three steps. Every team above follows the same path.

Create

One CLI call defines variants, the primary metric, and the decision rules: crawltic experiments create.

Assign & log

Your app calls /v1/assign on page load and /v1/events whenever revenue occurs.

Ship

Once quotas and confidence clear, Crawltic promotes the winner and fires a webhook to your deploy pipeline.

zsh — checkout-button, end to end

See it fit your workflow.

Tell us which team is testing first and we will map the setup with you.

Contact UsRead the Docs