AWS observability What's slow, and what do we fix first?
Sherpa is an AI platform for AWS. It shows your metrics, logs, traces and errors from CloudWatch and OpenTelemetry next to your AWS cost, so engineering leaders can identify which services are slow, understand why, and prioritize what to fix first.
What's slowing our customers down this week?
checkout-api is the slowest, and failing most.
| Service | p95 | Error rate | Next step |
|---|---|---|---|
| checkout-api | 1.8 s | 4.2% | investigate in Automate |
| search | 940 ms | 0.6% | investigate in Automate |
| orders-worker | 620 ms | 1.1% | investigate in Automate |
Which AWS services cost more in the last 30 days, and which of our services got slower?
EC2 is up $1,120. checkout-api is the slowest.
Cost up
| AWS service | Change |
|---|---|
| Amazon EC2 | +$1,120 |
| Data transfer | +$340 |
Slowest
| Service | p95 |
|---|---|
| checkout-api | 1.8 s |
| search | 940 ms |
Which Lambda functions are failing?
3 functions, 2.1% errors overall.
| Function | Errors | Invocations | Next step |
|---|---|---|---|
| process-payment | 4.8% | 120k | investigate in Automate |
| send-receipt | 1.9% | 88k | investigate in Automate |
| resize-image | 0.7% | 210k | investigate in Automate |
Which pages load slowest for our users?
/checkout, 3.4 s.
| Page | Load time | Next step |
|---|---|---|
| /checkout | 3.4 s | investigate |
| /search | 2.6 s | investigate |
Why does finding a slow service take so long?
69%
of companies run six or more tools just to watch their systems.
Source: Omdia, 2026.
Performance data lives in one tool, the bill in another and security findings in a third, so diagnosing a slow service means stitching them together by hand. Sherpa puts all three in one platform.
What can we see in Sherpa?
-
Metrics
-
Logs
search and live tail
-
Traces
latency and errors by service
-
Lambda
memory, duration, invocations
-
Browser performance
page speed and errors your users see
Data comes from CloudWatch and OpenTelemetry, and your team decides which data types to turn on for its workloads.
What happens when a service is slow?
checkout-api · p95 1.8 s · 4.2% errors
Investigate the error spike first.
Investigate in Automate.
Automate gives your engineers a remediation plan that starts with diagnosing the cause and ends with the fix. Sherpa never applies a fix; your team does. Ask about application performance and AWS cost in one question to see both side by side.
What does monitoring data cost us?
CloudWatch data is billed to your own AWS account, so observability needs cost governance too. For CloudWatch metrics, Sherpa shows the monthly cost of each setting before you turn it on: off, daily, hourly or real-time.
Questions we get asked.
Does Sherpa replace CloudWatch?
No. Sherpa reads your CloudWatch data and presents it alongside your cost and security findings. Sherpa is an AI platform for AWS that complements AWS's own services rather than replacing them.
Do I need OpenTelemetry?
Not for CloudWatch metrics and logs. For distributed traces from your own application code, your team sends OpenTelemetry data from its own collector or SDK, authenticated with a Sherpa API key.
Does Sherpa send alerts?
Not today. Your team opens Sherpa, or asks it a question in plain language, whenever a service feels slow or error rates look wrong.
How is this different from an APM tool?
Sherpa places application performance next to your AWS cost and security findings. Automate then gives your engineers a remediation plan for the slow and failing services it identifies.
What does it cost to turn on monitoring data?
CloudWatch data is billed directly to your own AWS account. For CloudWatch metrics, Sherpa shows the estimated monthly cost of each collection setting (off, daily, hourly or real-time) before you turn it on.