# How Chiirp came back from Datadog and stopped opening dashboards

[case-study-intro]
text = "Chiirp is a lead-conversion and AI-driven communication platform built for home service and trade businesses. Its co-founder and CTO, Kevin Neubert, moved the company's telemetry to Datadog, spent a year sampling one log in twenty to keep costs predictable, then returned to Better Stack. Chiirp now ingests twenty times more log data for 62% less, and traces a customer-reported bug in under ten minutes."
name = "Kevin Neubert"
role = "Co-Founder & CTO, Chiirp"
[/case-study-intro]

[highlights]
[[highlight]]
value = "20x"
label = "More logs ingested"

[[highlight]]
value = "62%"
label = "Cheaper than Datadog"

[[highlight]]
value = "<10 min"
label = "To trace a customer-reported bug"
[/highlights]

## Sampling is a cost decision with an engineering bill

Observability pricing scales with volume, so teams that grow quickly end up choosing between the telemetry they want and a bill they can forecast. The standard answer is to sample: keep a representative fraction of events and discard the rest.

That works for aggregate monitoring. It does not work for answering questions about one customer at one moment, which is most of what a support queue consists of.

Chiirp sampled at one event in twenty, sometimes one in twenty-five, and kept 30 days of retention. At that rate an individual user's session frequently left no trace at all.

[quote]
text = "Every now and then, and often, a client would say this happened and we couldn't figure it out, because the data wasn't there."
name = "Kevin Neubert"
role = "Co-Founder & CTO, Chiirp"
[/quote]

Cost control was the reason for the sampling, and Kevin found the charges difficult to forecast in the first place.

[quote]
text = "Their entire spreadsheet of, we charge for this, and we charge for this, and we charge for this. Every time we turned something on, we didn't know we were going to get an extra charge."
name = "Kevin Neubert"
role = "Co-Founder & CTO, Chiirp"
[/quote]

## Why he came back

Kevin had been a Better Stack customer before Datadog, so when he reopened the evaluation he looked at us alongside the rest of the market.

Both pressures were moving in the wrong direction at once. The bill kept rising, and because he was sampling to contain it, the data he needed was often not there when a customer reported a problem. Paying more for less was what sent him back to the market.

The deciding factor was the MCP server. Kevin was early in adopting AI tooling and initially treated it as something that might prove useful later. It is now the primary way he uses the product.

[quote]
text = "I'd say your MCP is the best there is. They did a good job there."
name = "Kevin Neubert"
role = "Co-Founder & CTO, Chiirp"
[/quote]

[wide-image]
src = "https://imagedelivery.net/xZXo0QFi-1_4Zimer-T0XQ/d1e57cd0-4635-452b-dec3-62de1367bb00/orig"
alt = "Chiirp's coding agent connected to the Better Stack MCP, querying logs to trace a reported customer issue"
caption = "Chiirp's agent queries Better Stack through the MCP to trace a reported issue, then opens the fix."
width = 2960
height = 1760
[/wide-image]

## One pipeline, from error to commit

Kevin has not written application code himself since late 2025. He directs an agent, and most of the engineering work that reaches him arrives as a queue rather than a task.

Chiirp's services write structured logs into Better Stack, tagged by severity. Errors route to Slack, and from Slack into Monday as tickets. An agent on Kevin's side picks them up, investigates, resolves and commits, and he merges the resulting fixes through the day.

Better Stack is the layer that gets queried. When support staff file a bug they record the client, the user, the contact, the timestamp and the reported behaviour. The agent takes those parameters, queries Better Stack through the MCP, and returns the activity for that client at that moment.

He also configures Better Stack through the MCP rather than the interface, alerting included. The dashboard gets opened roughly once a week.

[wide-image]
src = "https://imagedelivery.net/xZXo0QFi-1_4Zimer-T0XQ/1813509d-a55a-40d1-9d6e-ee9457088c00/orig"
alt = "A Chiirp bug ticket with the Better Stack logs behind it attached by the agent, alongside the fix branch"
caption = "A customer-reported bug becomes a ticket; the agent attaches the Better Stack logs behind it before committing a fix."
width = 2960
height = 1760
[/wide-image]

## From hours to minutes

Tracing a single customer's reported issue used to take Kevin hours of manual work. It now takes under ten.

The same path serves live support. A colleague on a call with a customer messages Kevin about a problem the customer is describing; the agent queries Better Stack, and an answer comes back before the call ends. That was not possible at one event in twenty, because the relevant data had already been discarded.

## What it costs

Chiirp ingests roughly 10 TB of logs a month, about twenty times the volume that reached their previous provider, at 62% lower cost.

Kevin kept retention at 30 days deliberately. Issues that have not surfaced within a month are not, in his assessment, worth investigating, and a bounded window keeps queries fast enough for an agent to work through.

[quote]
text = "I felt like, it's not so expensive, I'm going to just push everything that I need to be able to find a problem and fix it. It used to take me hours to find that. Using your MCP, less than 10 minutes."
name = "Kevin Neubert"
role = "Co-Founder & CTO, Chiirp"
[/quote]

Working against our MCP shaped how Kevin thinks about Chiirp's own roadmap. He wants Chiirp's customers to be able to state an action in plain language and have the platform carry it out, without an interface in between.

[quote]
size = "xl"
text = "Don't even go to somebody else. I was with Better Stack, went to Datadog, came back, and I kick myself from time to time for that mistake."
name = "Kevin Neubert"
role = "Co-Founder & CTO, Chiirp"
[/quote]

[cta]
title = "Get predictable observability, like Chiirp."
text = "Logs, metrics, traces, uptime and incidents in one platform."
[/cta]
