The combination of first-party behavioral data and artificial intelligence may transform ecommerce outbound marketing.
Called “AI individualization,” the goal is to create a personalized shopping experience tailored to an individual’s preferences, behaviors, and buying history.
The Perfect Send
“Internally, we strive for the ‘perfect send,’ when 100 percent of the people who get the message click or engage, and no one opts out,” said Alex Campbell, the chief innovation officer and co-founder at Vibes, a mobile marketing platform.
Campbell was discussing the potential for AI individualization (AI-I), Rich Communication Services, and mobile marketing in the retail sector when he described this 100% engagement, 0% opt-out scenario.
Ecommerce marketers might modify that definition, but the perfect send is when messaging meets a shopper’s need at the moment.
Shopper Expectations
Shoppers who opt-in to email, text, or push messaging want relevant offers.
“We do a customer survey every year…and we always ask a question like, ‘What would make you opt out?’ Two years ago was the first time we heard, ‘You are not sending me enough messages,’” said Campbell.
The folks surveyed had signed up to receive mobile marketing. They wanted to receive relevant and timely product notifications and discount offers.
AI-I can help.
First Party Data
Ecommerce AI-I is possible because online stores can collect first-party data — purchase history, browsing behavior, engagement data — without relying on third-party cookies or providers.
Humans cannot sort through all the data. Even rules and automations would struggle to reveal individual preferences in real-time.
An AI layer, however, can apply even during the deployment of the messages.
Not Merely Segments
Ecommerce marketers typically segment shoppers around common behaviors. A wine merchant, for example, might have a segment for “value wine shoppers” or “premium wine collectors.”
AI-I creates segments of one, such as a customer who buys red wine under $20, prefers Rhône varietals, responds to Friday sends, and often redeems mobile offers.
Composing the perfect send is much easier with a single segment.
Say the wine merchant implements an AI-I tool. This tool can send shoppers Rich Communication Services (RCS) messages and can access both the product catalog and shopper behavioral data.
Testing can lead to the perfect send.
The AI broadcasts an RCS message containing a product carousel. (RCS has app-like features.) The message has two offers: (i) an Argentine Malbec for $18, as recommended by AI based on the data, and (ii) a Portuguese red blend for $17, meant to introduce new wines to this shopper.
The shopper swipes, taps, visits the site, clicks a “Malbecs Under $20” filter, and ultimately makes a purchase. The AI adds the data from these touchpoints to the customer profile, recording the purchase under $20 or adding a note to test copy around value.
Each new message is an experiment, bringing the AI-I closer to discerning what a shopper wants and when.
That process is nothing new. Data scientists might describe it as “individualized multivariate tests” or a “contextual bandit.” It is an established way to identify individual preferences.
What is different is AI’s speed and scale.
Process Details
For the hypothetical wine shop, harnessing AI-I would require initial setup for more granular data collection, data normalization, and integration.
Once it’s up and running, however, the AI-I tool would likely follow a simple workflow for each new customer.
- Base segmentation. Start with broad wine categories based on the initial purchase, such as red or white, sparkling or still, and high-end or value.
- Early engagement. Begin sending messages and track, for example, whether the shopper clicks a Bordeaux at $40, ignores rosé, but buys a Malbec at $15.
- Individual testing. Generate shopper-specific messages. Each one is an experiment. Offer a Bordeaux at $35 or a Syrah at $18. Continue tracking engagement and behavior. Repeat.
- Refine the profile. Over time, the AI-I system identifies probabilities, such as “the customer is 70% likely to purchase when the price is under $20 and the varietal is bold red.”
- Balance with discovery. Introduce a “wild card” wine every few sends — perhaps a Spanish white or sparkling wine — to extend the system’s knowledge of the customer and prevent marketing fatigue.
- Feedback. All clicks, purchases, and opt-outs feed the AI model, both for the individual and to perfect the overall system.
With each iteration, the AI-I gets closer to the perfect send.
