Customer Analytics: A Guide to Getting Started (Part 4)
October 2nd, 2008 by Doug Bright(Read the entire Customer Analytics: A Guide to Getting Started series)
Last time we left off, we put together a plan to help us to use customer analytics to understand which marketing campaigns were worth pursuing and which to kill. This is a valuable analysis that can yield real profit growth without requiring any advanced predictive modeling. But what if we want to do something with a little more sizzle that really boosts the bottom line?
What if we could predict the products our customers are most likely to be interested in buying at any given moment? Not only could we personalize your website and marketing materials to be relevant to each customer, but we could also use this predictive power to do effective price discrimination.
For example, let’s say that we have a lot of women’s sweaters left over from last season. We need to move this inventory fast. Typically, we would deeply discount the distressed merchandise in an effort to boost demand. In the best case scenario, we are able to sell all our sweaters but we make very little profit because of the deep discounts.
But what if we could offer the merchandise to customers who are likely to be interested in women’s sweaters at full price before offering it at a discount to everyone else? This would allow us to extract full value from our target customers while still ensuring that we have to leftover inventory.
So let’s put a plan together to do that. Just like last time, we’ll start with the result and work backwards.
Desired result: Send a marketing email to customers who are likely to be interested in women’s sweaters advertising the products at full price before offering a discount to the rest of the customer base.
3) Identify the customers who are likely to be interested in women’s sweaters. We could try to do this using some intuition. For example, we probably want to narrow the field to women and then maybe only to those that have bought a sweater in the past. Maybe we can look at each customer’s purchases at the same time last year and see if there was a women’s sweater purchase. Or we could do it the “right” way and use predictive modeling to know for sure. By using predictive modeling techniques our results will probably be 2-5 times as good as intution so let’s use that.
2) Figure out how to do the predictive modeling. If we’re a retailer doing under $200M a year, we probably don’t have any statisticians or applied mathematicians on staff. This means we can either
- Hire a modeler to do this in-house
- Contract a consulting firm
- Learn how to do it ourselves (eek)
- Use a tool (such as our targeted email tool or other predictive marketing solutions)
1) Assuming we’re going to do this in house or hire a consulting firm, we need to integrate the required data for the modeling process. Ideally, we’ll want to put the following information in separate database tables or csv files:
- Order data (customer id, date, product, discounts applied, price, referrer code, order channel, gift y/n)
- Customer data (age, gender, address/zip code for demographic merge if available, acquisition source, date of first order, date of last order, order frequenecy)
- Product data (product cost, product category, product style attributes, quantity available)
If we have internal IT resources available we can get started on the integration immediately. Otherwise we may need to hire a integration consultant or if we decide we want to use a third party solution to do this, ensure that it handles most of the integration for us.
Again, we’ve managed to break this seemingly mammoth task into two action items: get the data into three separate tables or csv files and then get someone or something to run predictive models on it. The result will be a list of each of your customers and what they are most interested in buying next. From there, we need only to query the list for any given product to get the customer ids we need to send our email campaign. Simple, right?
OK, maybe not, but it’s definitely manageable. And the results will quite noticeable.
Tags: Customer Analytics, getting started
Add Personalized Product Recommendations to Your eCommerce Site
Istobe is a powerful recommendation engine that makes it easy to add recommendations to your eCommerce site. How powerful? Shoppers who click on Istobe recommendations spend 20-50% more than the average visitor.
