27 prompts for Klaviyo in Claude

Klaviyo shipped a Claude connector recently.

The MCP exposes 32 tools and takes five minutes to set up.

The real question once it is running, is whether you know what to ask - like where the deliverability audit keeps getting postponed, the segment overlap quietly drains sending capacity for six months, and the flow fix that would recover $15,000 a quarter never gets prioritized because nobody has ranked the options by revenue impact.

I built a prompt library with 27 prompts across 6 categories to close that gap and tested against a real account running around 84,000 profiles.

A few that have become part of a regular cadence:

→ The subscribed-but-never-purchased slice: Usually 20% of a list, and the 30-60 day tier still opens at 28% - there is conversion value there before those profiles drift into a sunset flow. The 1-year+ group at 4% opens is what suppresses deliverability, and the two need to be separated before you can act on either.

→ The segment overlap check: A lot of accounts run "VIP Customers" and "High LTV" as separate segments that turn out to be 90%+ identical, billing against the contact count twice over with almost no difference in actual reach.

→ The flow fix-it list: Audits every active flow and ranks the top 3 fixes by estimated revenue impact - because without that framing, you get a long list of suggestions, and with it - you have three things ranked by ROI.

How each prompt is built

Every prompt ends with a format instruction that tells Claude to return structured text instead of conversational prose, because the output you actually want from a Klaviyo audit is something you can drop into Slack or a Monday brief, not a chat response you manually reformat to make it usable.

The playbook covers:

1. Reporting and Metrics

2. Flow Performance

3. Segments and Lists

4. Profiles

5. Campaigns and Drafts

6. Cowork Workflows

Each prompt is mapped to a pattern the connector actually exposes, with an operator note at the bottom of every prompt explaining what to watch for when the output comes back differently than expected. Sharing the full playbook, linked below.

If you want to see how to organize Claude across your team before you start running prompts, I have recorded the setup walkthrough.

Let's connect

If you run something like this and want to compare notes on what is actually working, I spend most of my week on agent orchestration, model routing decisions, and what is moving on Klaviyo and Shopify at scale.

DM me on LinkedIn

Or drop an email at ankit@atomz.ai

Share this with someone who can find it useful: howtoaiforshopify.com

- Ankit