Workflow Spotlight on
Attribution Anomaly Detection with Claude
Learn how Grapevine Surveys and Claude work together to monitor your attribution survey for anomalies, catching sudden shifts in how customers say they found you before those shifts quietly distort your marketing decisions.
Claude 🤝 Grapevine Surveys
Claude is Anthropic's AI assistant, and it can now connect directly to your Grapevine survey data through the Model Context Protocol (MCP), a secure, open standard for linking AI assistants to the tools you already use. Once connected, Claude can list your live surveys, pull response summaries, and read the answer breakdown for any question, all in plain conversation. That turns Claude into a monitoring layer that sits on top of your attribution survey and tells you, in words, when something looks off.
The Workflow Objective:
Use Claude to monitor your Grapevine attribution survey and flag anomalies in the responses, such as a channel spiking or vanishing, a jump in "Other", or a drop in completion rate, before they mislead your budget decisions.
The Challenge
Attribution data is only useful if you can trust it. When a customer tells you they found you through a podcast, a friend, or a TikTok video, that answer feeds real decisions about where to spend. But the shape of that data drifts constantly, sometimes for good reasons and sometimes for bad ones. A channel might genuinely take off. Or an answer option might break, a survey edit might reshuffle the choices, a wave of bot responses might inflate one channel, or a spike in orders from a single campaign might skew the whole distribution for a week. From the outside, a real shift and a data-quality problem can look identical.
The operational problem is that almost nobody has time to eyeball the attribution report every week. So a shift can go unnoticed for weeks: "Instagram" quietly doubling, "Other" ballooning, or completion rate collapsing after a survey change. By the time it surfaces, you may have already reallocated spend on the back of numbers that were never real. What merchants need is a low-effort way to keep an eye on the data continuously and get a clear heads-up the moment something looks unusual, with a sense of why, rather than discovering the problem after the damage is done.
The Solution
The solution is to connect Claude to Grapevine through the Model Context Protocol and let it do the watching for you. Once the connection is authorised, Claude can read your attribution survey's response summary and its per-question answer breakdown on demand. You establish a baseline (what a normal week of responses looks like) and then ask Claude to compare the latest data against it.
Because Claude works with your live survey data rather than generic advice, it can surface anomalies in plain English: "Podcast responses have tripled week over week," or "'Other' free-text answers have jumped from 4% to 22%, worth checking whether an answer option is missing or broken," or "Completion rate dropped from 78% to 41% after the 3rd. Did the survey change?" You get a written flag with the likely cause attached, not just another chart to interpret. Run it whenever you want a gut-check, or set it up to run on a regular cadence so the monitoring happens without you having to remember.
To do in Grapevine
Make sure you have a live attribution survey running, typically a single "How did you first hear about us?" question with structured single or multiple-choice options and an "Other" free-text field to catch anything you haven't listed. Structured options are what make anomaly detection reliable: they give Claude a clean, consistent distribution to compare over time, where free text alone would be noisy.
Make a note of your survey code so Claude can find the right survey quickly. Claude can also retrieve this for you by listing your active surveys once it's connected.
Custom Connector in Claude Claude
Add Grapevine as a custom connector in Claude using the MCP server URL https://app.grapevine-surveys.com/mcp, then authorise it. Your existing Grapevine permissions apply, so Claude only sees what your account can see. Once connected, Claude has read access to three things: your list of surveys, each survey's response summary (response counts, impressions, and completion rate), and the per-question answer breakdown.
Prompt Claude
From there, a monitoring prompt is all it takes. For example:
"Pull the response summary and per-question breakdown for my attribution survey. Compare the answer distribution and completion rate for the last 7 days against the previous 7 days, and flag anything that looks unusual, such as big shifts in any channel, a spike in 'Other', or a drop in completion rate. Tell me the likely cause of each anomaly."
Run that ad hoc whenever you want a quick check, or automate it on a schedule so Claude reports back weekly without prompting.
How to setup this integration
Never miss a beat with constant ai monitoring of response data
How can this data be used?
Turn continuous monitoring into faster, safer decisions.
When Claude keeps an eye on your attribution survey, you protect the quality of the data underneath every marketing decision, and you spot real change while there's still time to act on it. This is the difference between attribution data you check occasionally and attribution data you can actually rely on.
Data Quality & Integrity
- Catch broken or missing answer options before they pollute weeks of responses
- Spot bot activity or unusual response spikes that would otherwise skew your distribution
- Confirm that a survey edit didn't accidentally tank your completion rate
Marketing Signal & Reaction Speed
- Notice a channel taking off (a podcast mention landing, an influencer post converting) while it's still worth doubling down on
- Get early warning when a reliable channel starts drying up, before it shows up in revenue
- Cross-check sudden shifts against your ad platforms and post-purchase survey data to separate real change from noise
You're not just collecting attribution data, you're keeping it honest, and turning it into an early-warning system for how customers are really finding you.
Start monitoring your attribution data
Your customers already tell you how they found you. Grapevine captures those answers, and Claude watches them for the shifts that matter, so a broken option or a rising channel never slips past you unnoticed. Together they give you a simple, low-effort way to keep your attribution data trustworthy and act on it faster.
Other Workflow Spotlights
Powerful survey workflows with Grapevine + your favorite apps
