Looking across the brand's products, data and interactions, her team asked where its users have stopped engaging and how to re-engage them. "It's micro-segmentation and it can be very small groups." They identified data points and estimated average monthly volume. "Some we knew would work but there was not enough volume to impact the business. Hence; Pitfall #1: Too much work for too little impact.

Using predictive analytics, The Knot Worldwide generated a welcome series and tested a general message against specific vendor listings and saw a 42% improvement in lead conversion to vendors. And Pitfall #2 reared its ugly head.

The recommendations can be low-quality, so you have to test your algorithms. Get the data team to take information into consideration in order to modify recommendations.

One of the challenges it faces is using its dashboard piece to feed back to users their recent activity. But if a user is not updating, the company has to decide when to take out a datapoint, leading to Pitfall #3: Dynamic data over time. Adjusting the message to account for the change.

It also sends messages to people who have searched for, say, wedding cake. With weeks to the wedding, sometimes it doesn't have all the information it wants yet can't ignore the customer. "So we're playing with this new message," said Grimstead, adding: "Don't ignore your dormant users. Pitfall #4 is talking only to the converted.

When users come to The Knot Worldwide, they are required to input a wedding date. Yet some may not know the real wedding date, so it's not a very reliable data point. "We're working on ways to sense when that date is going to be different," she said. Pitfall #5 is relying on unreliable data.

Pitfall #6 is putting too much focus on open and click rates. Pitfall #7 is not following up with hot prospects.