If you’ve ever shopped at Amazon.com, you’re probably familiar with the many up-selling and cross-selling features of the Obidos ecommerce platform. Amazon frequently lets you know:

• Which similar products were actually purchased, instead of the current item.
• Which other products were commonly purchased, in addition to the current item.
• Which products were highly rated by other users who purchased products which you have rated highly.

If the Amazon recommendations system (read this paper if you’re interested in the clever mechanics of it) fails to find any products that interest you, then you probably have very esoteric tastes in books, movies, music, and kitchen appliances.

Over the last ten years, as Amazon has innovated, other providers of ecommerce software have included those innovations in their own shopping cart products. It’s rare to see a shopping cart anymore that doesn’t include some sort of statistical recommendations engine.

But I’ve got a hunch that very few brick-and-mortar retail businesses (even those that use computerized point-of-sale software) perform statistical analysis of their sales data to gather business intelligence.

Off the top of my head, here are some statistics that could be easily gathered from a transaction database at a typical retail store:

• Average total receipt (plus median, standard deviation, interquartile range, etc).
• Average number of items per purchase (plus median, stdev, IQR)
• Average item price (plus median, stdev, IQR)
• A rank-order list of highly correlated items (customers who purchase peanut butter are highly likely to also purchase jelly).
• A rank-order list of inversely correlated items (customers who purchase twinkies are highly unlikely to also purchase whole wheat bread).
• A histogram showing the frequency of repeat customer visits. (32% of this weeks customers have never shopped here before. 12% are returning within a week of their last purchase. 25% are returning within a month of their last purchase.)
• A rank-order list of the most profitable products (this assumes that the inventory database also includes wholesale prices, so that the profit margin of each product can be calculated).
• A detailed breakdown of redeemed coupons (96% of customers purchased without a coupon. 3% had a coupon from the September mailer, and 1% had a coupon from the October mailer.)
• All reports should be pivotable on a number of different axes: day-of-week, time-of-day, day-of-month, transaction type (cash, check, visa, mc), cashier name, etc.

I’d also produce a really reader-friendly report, with conversational bullet-points saying things like this:

The majority of your customers (68%) spend less than $10 per transaction, but you should pay extra attention to the minority of your customers (7%) who spend more than $100 per transaction, since those customers account for a disproportionate percentage of your total profit margin.

These big-spenders tend to purchase high-margin items (the margin on their purchases averages 45%, verses 22% for all other customers), and they’re generally more loyal than other customers (they’re 18% more likely than other customers to make another purchase within the next 30 days).

So treat those customers well.

I’d also include lots of pretty reports with colorful charts (with moving averages and trend lines to demonstrate the effectiveness of various promotional campaigns), giving small business owners a lot of deep insight into the purchasing patterns of their customers.

Basically, this idea germinated as an afterthought to my original WebDelve idea for small-business web analytics. Basically, what I’m envisioning here is web analytics for businesses that don’t exist on the web (as a bonus, it would be cool to integrate retail analytics across web and non-web retail outlets).

Market Analysis & Competition

IBM seems to be the dominant supplier of point-of-sale hardware. There are quite a few partner companies selling software solutions for the IBM PoS terminals, but I haven’t been able to determine which (if any) of those systems are dominant in the marketplace.

Conveniently, most of those tools seem to lack any business intelligence component, so the opportunities for providing a value-add product seem promising.

It’s a mixed blessing that the PoS marketplace is so fractured. On the one hand, it means that there’s no clearly dominant provider of retail analytics software. On the other hand, it means a possible integration nightmare, integrating with a zillion different PoS systems. (However, since integration could be handled through an abstract integration layer, it would probably be pretty straightforward to add compatibility with additional systems through a clean, modular architecture.)

Since this would probably be an expensive piece of software (and since small business owners are often pretty strapped for cash), I’d probably be most successful if I could identify a few of the major PoS software providers and partner with them, letting them take on some of the marketing burden. They could add a new feature to their feature lists, and I wouldn’t have to work as hard to convince customers of my legitimacy.

The ideal customer for this kind of software would probably be a company with at least a million dollars in annual gross revenue (or a company that can at least anticipate revenues in that ballpark). The tiny mom-and-pop stores are probably not of much interest, since they’re not likely to be tech-savvy enough to use computerized point-of-sale terminals with inventory and transaction management systems. Assuming a 25% margin, a million dollars in annual receipts would result in $250,000 of net revenues (after paying the wholesale suppliers), and that’s about enough cash to pay rent, plus the salaries of a manager and three or four full-time cashiers.

Of course, this software would not be targeted at massive retailers like Target or BestBuy. They already perform significant analytics about the purchasing patterns of their customers.

Pros:

• I think a lot of my existing analytics ideas (from the web-analytics problem domain) would apply equally well to retail analytics. The core analytic code for this software would be pretty straightforward for me to develop.
• The retail analytics market is not so clearly dominated by a few big players (Google Analytics, Omniture, WebSideStory, and a number of other companies already dominate the web analytics space.)

Cons:

• I have no experience managing a retail store. But I have a few friends who own a retail store, and I could consult with them for ideas. Also, since you can’t throw a rock without hitting a half-dozen retail stores, it probably wouldn’t be too hard to find a few companies willing to beta-test the system for free. Nevertheless, since I have no experience in the retail industry, I’m counting this as a disadvantage.
• Although I could probably develop the core analytic engine (and many of the reports) within a six month development schedule, the long sales cycle for a product like this would probably mean it’d be quite a while before I made my first sale.
• I have no idea what a product like this would cost, so I’m not very well equipped (yet) to evaluate where it would rank (compared to my other business ideas) in terms of profitability.

…

This is the seventh of 30 business ideas that I’ll be writing about over the course of 30 days. Some of the ideas leverage my experience in a particular industry. Other ideas apply some of my favorite analytic techniques to industries that are totally new to me. One of them will become a product over the next six months, and the foundation of my new software business.

This entry was posted on Thursday, June 22nd, 2006 at 2:59 am and is filed under biz. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.