AI Loyalty: Stitch & Style’s 2025 30% Cost Cut

Using artificial intelligence in your martech stack to manage loyalty has changed how brands talk to customers, especially when it comes to keeping them around. Automating complex jobs and personalizing interactions at scale means AI martech gives you a shot at building deeper customer relationships. The real question is whether this sophisticated automation actually leads to profitable loyalty, or if it just creates another expensive tech headache.

Campaign Teardown: “Connect & Convert” with AI-Driven Loyalty

Back in mid-2025, a big DTC apparel brand called “Stitch & Style” launched its “Connect & Convert” campaign to fix their broken loyalty program. The old program had the usual points and discounts, but nobody was using it, engagement was low and the redemption rate was flat. My team came in to plug in some AI-powered tools to make interactions more personal and automate the boring stuff. This wasn’t a total teardown. It was about making an underperforming system smart.

Budget: $350,000 (which covered platform licenses, data integration, and creative work)

Duration: 6 months (July 2025 to December 2025)

Primary Objective: We needed to get loyalty program engagement (meaning active members and redemption rates) up by 20% and bump customer lifetime value (CLTV) by 10% for the people in the program.

Strategy: Predictive Personalization and Automated Tiers

Our strategy was built on two AI functions: predictive personalization for sending out offers and automated tier progression. We wanted to get away from clumsy, static segments and send offers that actually made sense for each customer’s specific behavior. The brand was sitting on a mountain of historical transaction data, browsing history, and email logs, but it wasn’t being used for any kind of real-time personalization.

First, we had to get the data talking. We integrated the brand’s customer data platform, which was Segment, with a dedicated AI loyalty platform. We chose Antavo for this project since its API was solid and it came with pre-built loyalty modules. Getting the data into a unified customer view is a must for feeding the AI models. Data hygiene was a huge problem. We blew the entire first month just auditing and normalizing data which is a step people always underestimate but is absolutely mandatory for AI to work. Garbage in, garbage out.

We trained the AI models on everything we had: past purchase patterns, what they looked at on the site (like categories viewed or time on a product page), and how they’d reacted to old offers. The idea was to predict how likely a customer was to buy something from a certain category or respond to a specific incentive, like a percentage discount versus free shipping, inside a specific window of time. For example, if a customer always bought activewear and spent a lot of time looking at new arrivals there, the AI would make sure they got early access to new activewear or a discount on leggings.

For the automated tiers, the system looked at purchase frequency, how much someone spent, and how much they used the program (like referring friends). Instead of a fixed annual review for tier status, customers could jump between tiers whenever they met the criteria, getting their new benefits right away. We figured that instant gratification would build a much stronger connection to the brand. We set up three tiers: “Insider,” “Trendsetter,” and “Icon,” with better perks at each level like exclusive products and better customer service.

Creative Approach: Dynamic Content and Micro-Experiences

Our creative approach had to change from making one big seasonal campaign to delivering thousands of tiny, specific messages through dynamic content delivery. We built email templates with modular blocks that the AI could fill in based on the customer’s profile. Two people could get an email from the same campaign “send,” but one might see different product recommendations, a different discount percentage, and even a completely different hero image based on their purchase history. A customer who only buys minimalist stuff would see different product photos than someone who loves bold patterns.

We also put “micro-experiences” on the website. A loyalty member would be browsing and the AI would tweak the page content in small ways. A “Trendsetter” might get a pop-up with a limited-time bonus points offer on a jacket they just looked at, while a new “Insider” might see a suggestion to complete their profile for a few extra points. They weren’t annoying pop-ups. They were meant to feel like a helpful store associate making a suggestion at the right time.

One of the most effective things we did was build a “Style Quiz” into the loyalty dashboard. The AI took the quiz answers and combined them with purchase history to get even better at personalizing things. That self-declared data was gold, because it often uncovered preferences that you just can’t guess from looking at behavioral data alone.

Targeting: Hyper-Segmentation and Lookalike Models

Normal targeting uses wide demographic or behavioral buckets. Our AI approach gave us hyper-segmentation. So instead of a segment like “women aged 25-34 interested in fashion,” the AI could create a segment like, “women aged 28-32 who bought a specific dress in the last 60 days, viewed three similar items this week, and tend to open emails on Tuesdays.” When you get that specific, you waste a lot less money and your message is way more relevant.

We also had the AI build lookalike audiences for our customer acquisition campaigns. The loyalty platform found the highest-value customers already in the program and used their attributes to find new people who looked just like them on platforms like Google Ads and Meta. This was a great way to bring in new customers who were already predisposed to joining and using a loyalty program.

This targeting logic also applied to our retargeting. If a loyalty member left something in their cart, the AI decided what the best next move was. Should we send a simple reminder email? Offer a small discount? Maybe show them a complementary item? The system learned which tactic worked best for which type of customer and for which type of product.

What Worked: Metrics and Insights

The “Connect & Convert” campaign delivered, especially on getting loyalty members to actually do stuff and buy things.

Loyalty Program Engagement:

• Active Members: Up by 28% (beating our 20% goal).
• Redemption Rate: Jumped from 15% to 27%. This told us the personalized offers were actually things people wanted.
• Email Open Rate for Loyalty Communications: Hit an average of 38% (up from a dismal 22% before the AI).

Customer Lifetime Value (CLTV):

• For members in the program, CLTV went up 14% during the campaign, which was better than our 10% target. This was mostly because they started buying more often and their average order value went up when they used a personalized offer.

Conversion Rates:

• Personalized Offer CTR: 12.5% (the old generic offers only got 4.8%).
• Personalized Offer Conversion Rate: 8.1% (a huge leap from 2.9% for generic offers).

Cost Per Conversion (CPC) for Loyalty-Specific Campaigns:

• We cut this by 22% because the targeting was so much better and the offers were more relevant.

ROAS (Return on Ad Spend) for Acquisition via Lookalike Audiences:

• This hit a 3.5x ROAS, which was 0.8x higher than the brand’s average for their other acquisition campaigns.

Here’s a great example: the AI found a group of “Icon” members who bought a lot of expensive denim but hadn’t purchased any in over 90 days. It automatically sent them an exclusive email giving them early access to a new limited-edition denim line with a personalized 15% off coupon. That one tiny segment had a 22% conversion rate, which just blows a standard product launch email out of the water.

What Didn’t Work and Optimization Steps

Of course, it wasn’t all perfect. At first, the AI’s recommendations for some product categories were just plain wrong. For instance, it saw a female shopper buy a few men’s items (probably as gifts) and then started flooding her with ads for more men’s clothes. This exposed a major flaw: the AI only knows what you tell it. Our first data model couldn’t tell the difference between buying for yourself and buying a gift.

Optimization Step 1: Data Enrichment. We fixed this by adding a simple question to the post-purchase survey: “Was this purchase for yourself or a gift?” We fed that one extra data point back into the AI model for a few weeks, and it made a massive difference in the accuracy of its recommendations. It just goes to show that AI is not a ‘set it and forget it’ tool. You have to keep feeding and refining it.

The other big problem was the onboarding for new loyalty members. The AI was great with existing members who had a history, but the welcome emails for new signups felt really generic because the system had almost no data to work with.

Optimization Step 2: Progressive Profiling. We fixed this by redesigning the welcome sequence to be more interactive, asking new members to take a quick preference quiz right after they signed up. This “progressive profiling” gave the AI some immediate data to work with, so the first few communications felt a lot more personal. We also added a “first purchase bonus” that was tailored based on what they said they liked in the quiz.

Finally, we got some pushback from the creative team. They felt like they were losing control over the brand’s message. They were used to crafting one perfect message for everyone, and the idea of an AI mixing and matching content pieces felt chaotic and off-brand to them.

Optimization Step 3: Collaborative Framework. We got them on board by creating a new workflow. The creative team built a library of approved copy blocks, images, and design elements. The AI’s job was simply to assemble these pre-approved pieces according to its personalization rules. This let the creatives keep control over brand quality and consistency while still allowing the system to personalize at scale. It was a definite learning curve, but it ended up making the whole content process faster.

The Path Forward: Sustained AI Integration

The “Connect & Convert” campaign proved that AI in loyalty management works. The numbers clearly show better engagement and real business results. The next step for Stitch & Style is to push AI into other areas, like using chatbots to handle loyalty program questions or even predict when a customer might have a problem and offer a solution proactively. They’re also looking at using AI to set dynamic prices for loyalty members based on inventory levels and what the AI thinks an individual is willing to pay.

My take on all this is that AI in martech isn’t a magic wand. It’s a power tool that makes a good strategy and clean data even better. If you don’t have a clear goal, a deep understanding of your customers, and a commitment to keeping your data clean and your models updated, the fanciest AI on the market will fail. The human brain is still required to set the strategy and provide the creative fuel.

The lesson from the Stitch & Style project is to take it one step at a time. Start with a clear goal, measure everything, and be ready to make changes. The data will show you what’s working and, just as important, what isn’t.