Dandelion Chocolate’s customers were all getting the same campaign. We used OuterSignal to group them into six audience segments, wrote a version of each launch email for every segment, and tested it against a standard send. On the same audience size, the matched emails converted 97.1% better and brought in 94.6% more revenue. A follow-up send pointed to why: the lift came from the match, not the copy.
Facts
Client: Dandelion Chocolate
Industry: Craft chocolate (CPG, gifting)
List Size: 100K+
Partners: OuterSignal
ESP: Klaviyo
Focus area: Audience-matched campaign personalization
Results
- +94.6% revenue from matched sends vs. the control, on the same audience size ($9,869 to $19,203)
- +97.1% conversion rate per delivered email, with every one of the five segments significant at 95% confidence
- +23.2% click rate and +32.3% click-to-open, with average order value flat, so the lift is conversion, not order size
- 4.6x the conversion for the winning email on its matched segment vs. the general list
Client Insights
Dandelion Chocolate is a San Francisco bean-to-bar maker. Two ingredients, cocoa beans and sugar, single-origin bars, a confections line that changes with the seasons, and a gifting business big enough that a few weeks in December decide the year. Their email list is one of the most engaged we work on, with campaign open rates that regularly sit above 70%.
We’ve run Dandelion’s lifecycle email for over two years. This project was about something the account didn’t have yet: a way to write to the person opening the email, not just to the product being launched.
The Challenge
Every customer got the same campaign. Behavioral data, which is what email platforms run on, tells you what someone did: bought a bar, opened a newsletter, clicked a gift guide. It doesn’t tell you why they buy, and in a brand like this the why varies a lot: some buyers care most about how the chocolate is made, some about what goes into it, some about giving it as a gift. All of them open the same email about the same dragées.
The question we wanted to answer wasn’t whether a better email would sell more chocolate. It was whether matching the message to the person would, and whether we could separate that from the copy simply being stronger. That second part matters, because plenty of “personalization” wins aren’t personalization at all. They’re better writing in disguise.
Our Process
1. Six audience segments, built from the brand’s own customers
OuterSignal groups a brand’s customers into audience segments built from its own customer base, not a generic template. For Dandelion it found six. Customers were assigned to a segment from December 2025 onwards, and the segments synced into Klaviyo so each one became an audience we could send to.
2. One launch, six emails, one control
For each product launch we wrote a version of the email per segment. Same product, same offer, same send time. What changed was the angle: technique, ingredients, craft and packaging, the gift occasion or the sourcing story, whichever that segment responds to. A general version went to a control group drawn from the same audience. We pooled every segment and compared it with the control first, so the headline read on “does personalization work” couldn’t be rescued by one strong segment, and we didn’t call anything a win without a two-proportion z-test at 95% confidence.
3. Three tests, each bigger than the last
The first test, the Signature Dragée Trio in May, was the small one: conversion per click up 33%, and enough signal to run it again at scale. The second, the Italian Aperitivo collection in June, converted 29.8% better per delivered email, 54.3% better per open and 47.8% better per click, and it did that with a handicap: the control was running on Maverick’s inbox optimization and the matched versions weren’t. The third, the Fruit Dragée Trio on July 16, is the one in the results block.
4. The send that tested the mechanism
On July 17 we sent the winning email to the rest of the list, prospects and customers, with no segment match. If the copy alone had carried the lift, the general send should have come close. It didn’t: on its matched audience the same email had converted 4.6 times better than on the general list. The general list includes prospects, who convert less on any email, so it isn’t a clean control, but it points the same way as the controlled tests: the lift travels with the match, not the words.
The Results
The Fruit Dragée Trio test, July 16, 2026. Five audience segments pooled against a general control, every rate lift significant at over 99% confidence:
| Metric | Control | Matched | Lift |
|---|---|---|---|
| Click rate | 2.29% | 2.82% | +23.2% |
| Click-to-open rate | 3.86% | 5.10% | +32.3% |
| Conversion rate (per delivered) | 0.320% | 0.630% | +97.1% |
| Conversion rate (per open) | 0.539% | 1.141% | +111.6% |
| Conversion rate (per click) | 13.99% | 22.38% | +60.0% |
| Revenue | $9,869 | $19,203 | +94.6% |
| Revenue per recipient | $0.36 | $0.70 | +95.1% |
Every one of the five segments beat the control on conversion, each on its own at 95% confidence. The lift is conversion, not order size: average order value was the same on both sides, so the extra revenue came from more people buying.
Scaled to the brand’s full segmented audience, the Aperitivo-level lift alone projects to roughly $6,300 in incremental revenue per matched send, before the July result is applied, and that figure grows with every new customer added.
Where To From Here?
Campaigns were the proving ground, but they aren’t where the audience is. The bigger audience is in flows, where the same first-time buyers and lapsed customers pass through every day. The next layer is a segment-matched post-purchase touchpoint about two weeks after delivery, built around the customer rather than what they bought, and a segment-matched win-back that reconnects lapsed customers with the reason they bought in the first place. Segment matching is already routine on Dandelion’s launches: the September collaboration email went out in a story-led version and a product-led version, matched to the segments each one suits.
Two things would compound the results from here: extending OuterSignal segmentation to older customer profiles, which widens the testable audience, and applying segment targeting to the big seasonal launches, where audience size and purchase intent make the extra creative work pay off fastest.
