A marketer's best first personalization campaign, before they knew they needed one.

Whoever buys personalization is rarely the one who runs it.

A marketing leader signs for personalization, usually bundled with Optimizely's A/B testing. Running it falls to a marketer, who finds out afterwards: “oh, and by the way, we bought this personalization product.” They never chose it, and they still report to the leader who did.

492 accounts paid for personalization in June, and only 58 did anything with it.

Optimizely asked my team to explore what marketers could do with AI agents. I looked for what was broken first. Paying accounts grew every month from September to June, while the number using it stayed between 49 and 61.

Of the accounts that launch one campaign, 82% go on to launch three or more.

Marketers on completely different tools were stuck for the same three reasons.

I expected the problem to live inside Optimizely's product, so we started with the experts who build and sell it. Then marketers on other tools described the same stalls, so the problem was bigger than Optimizely's interface.

Ten in-depth interviews, with six experts and four marketers, gave three reasons.

The first campaign waits next to the A/B tests a marketer already runs.

Marketers didn't know when to start, so I put the start where they already work: a quiet band on the Overview tab. Most weren't even going into Personalization to set it up.

Each account sees which campaigns it can run today, and what the rest need.

Experts told us marketers assume they need a customer data platform first. That's true for some campaigns and not others, so each campaign is sorted by the data the account already has. When one needs the tracking snippet, the page names who installs it and drafts the note to send.

I asked for an honest read, and the rule built from it banned any verdict.

I chose to promise marketers an honest read instead of proof of a winner. The agents turned that into a rule: the page could never give a verdict. I thought the page answered the marketer's question, since it came from my promise.

The spec's first test was “No verdict, ever,” and the page passed. The spec even quoted a former customer asking “How do I know if my personalization campaign made me money or not?” as its reason never to answer.

A marketer has to tell their manager if it worked or not, so now the page does.

Two Optimizely experts walked the prototype and neither could say what the results page meant. That's when I found the rule I'd missed.

The rule was right that a page can't always prove a winner, but a winner was never the only answer. So I redesigned the page myself around a verdict a marketer can repeat, sized to the evidence: from enough data to act to a direction, not proof. The redesign is merged.

Four people and a set of agents, and the hard part was agreeing.

I built the knowledge base the four of us and the agents worked from.

Optimizely committed to building it. Next is the campaign nobody nudges.

I scoped the work to the first campaign, because that's where the problem was. The second starts differently: nobody hands it to you, and every prototype campaign begins with a nudge. Next, I'd design one built from scratch.