Case Study
ClickFunnels × Honeybadger
Bots were testing stolen cards at checkout. Honeybadger helped ClickFunnels validate a way to stop them before rolling it out to customers.
"That's one thing that I've just totally loved doing with these tools: validating and gaining confidence, and then flipping the switch and removing the feature gates."
At a glance
Validate the change before enforcing it
ClickFunnels already relied on Honeybadger for error tracking and deploy verification. Nate Vick, VP of Platform Engineering, wanted to see how changes would behave before rolling them out to customers. Checkout was a sensitive place to start: stopping card testing—bots trying stolen card details at checkout—couldn't come at the expense of legitimate payments.
The team added bot detection to the checkout flow and recorded its results in Honeybadger without blocking requests. Keeping blocking behind a separate feature flag let them observe the behavior before deciding to turn it on.
Nate used a dashboard to see which checkout requests would fail the new check, broken down by customer workspace. He compared those results with records in their database before the team enabled blocking on their own checkout, where it helped reduce card testing. The protection is now fully rolled out to customers.
Before launching Build with AI, he used the same approach to investigate flaky behavior, make changes, and show the team the results on a dashboard.
From a question to a dashboard
Nate described what he wanted to investigate and worked with Claude to add the events and build dashboards through the Honeybadger CLI. He could then inspect the results and decide how to proceed.
Sample checkout data over three hours: 100 verifications, with 72 valid nonces and 28 invalid ones. With enforcement off, 18 requests would have been blocked. With enforcement on, 10 were blocked. The reason table breaks the invalid results into 24 missing parameters and 4 expired nonces. The line chart shows missing parameters ranging from 0 to 5 per 15-minute interval, and expired nonces ranging from 0 to 1.
| Reason | Verifications |
|---|---|
| nonce_valid | 72 |
| missing_params | 24 |
| expired | 4 |
Understand what customers are using
Nate also built an AI usage dashboard to understand what customers are using and what it costs. It shows LLM costs by workspace, input tokens by product line, and which models are in use. Errors appear alongside those metrics, so he can watch service reliability while examining usage.
Nate can narrow the dashboard to a workspace to understand what an individual customer is using and costing them. The team is using that data to consider plan allowances and overage pricing.
Sample AI usage over three hours: 68 LLM calls, 1,240,000 input tokens, 148,000 output tokens, and $6.84 total cost. The line chart shows cost fluctuating between $0.15 and $1.34 per 15-minute interval. Workspace costs are $3.42 for workspace 10482, $2.18 for 20816, and $1.24 for 31507. Input tokens by product are 580,000 for conversation summaries, 420,000 for contact filters, and 240,000 for page markup.
| Workspace | Cost ($) |
|---|---|
| 10482 | 3.42 |
| 20816 | 2.18 |
| 31507 | 1.24 |
| Product | Input tokens |
|---|---|
| Conversation summaries | 580,000 |
| Contact filters | 420,000 |
| Page markup | 240,000 |
With Honeybadger, Nate can show his team how a change behaves before rolling it out and understand what it costs once customers use it. That gives the team evidence for both engineering and product decisions, from enabling checkout protections to planning AI usage allowances.
Get monitoring done.
Get real-time alerts when code breaks, monitor performance, and fix errors in record time.