offers

New contract, and a whole new world. I'm at Marriott now, on the team that runs the credit card offers you see when you're logged in on web or mobile booking a stay, the co-brand cards they do with the banks. The stack is Adobe Target for deciding who sees what and AEM for the content itself, and my job is maintaining and building the targeting activities. After years of building the pages, I'm now working on the side that decides which version of the page you get.

what personalization actually is

From the outside, personalization sounds like the machine deciding your fate. From the inside, it also sounds like the machine deciding your fate. An activity in Target is roughly: for this audience, on this page, show this experience instead of the default. The audiences are built from attributes, member status, geography, whether you already hold the card, what you've browsed, and the experiences are content variants, a different offer, different art, different copy for someone who's already a cardholder versus someone the bank would like to become one.

The craft is in the details underneath. Audiences are only as good as the attributes feeding them, and some attributes come from profile scripts, little pieces of logic that run per visitor and compute things audiences can use.

// a Target profile script, roughly: runs on every request for a
// visitor, its return value becomes an attribute audiences can use
// user.cardMember.  This is pseudocode.
if (profile.get('crm.offerCode') === 'XXXX' && profile.isAuth()) {
  return 'member';
}
if (profile.get('page.lastOfferViewed')) {
  return 'shopping';
}
return 'prospect';

Get this wrong and a few hundred thousand existing cardholders see an acquisition offer for a card they already have, or someone not qualified for an offer is shown something we now legally are obligated to uphold. Most of my week is the unglamorous discipline around exactly that: QA across every audience and experience combination, checking the traffic splits, making sure the activity that ended actually ended, and the meat of it: setting up new campaigns for QA and stakeholders to review.

the reports again

The measurement lesson from the dealer locator years turns out to run this whole world. Every activity lives or dies by its numbers, conversion against bookings, and Target's reporting feeds decisions about real money. Which means the tracking has to be right, which means the same telemetry-first discipline applies, except now a broken beacon doesn't just make a dashboard wrong, it makes the company conclude an offer failed when actually the counting failed. I have watched a week of debate about a underperforming experience end with the discovery that a success metric was misconfigured in one of my production activities.

what I think i'm learning here

Two things I didn't appreciate from the dev side. First, content operations are the real bottleneck, not the targeting tech. AEM fragments, approvals, pushing changes back to stakeholders, the banks signing off. The tool can ship a variant in minutes, the organization ships it in weeks. Second, the marketing people I work with are sharp in a way engineering culture doesn't always credit. They think in segments and incentives the way I think in components and state, and the ones who are good are very good. The overlap between their world and mine, where the data flows and whether it can be trusted, keeps turning out to be where I'm most useful and feel like I understand things the best.