Bad Filtering Kills Businesses

If your business model is based on information, and whose isn’t these days, then you need to be able to aggregate, filter and connect. While reflecting on the death of Borders Books, I thought of three stories of filtering in retail.

First Story: Tower Records

In the mid-80s, I went in to the Tower Records in Tacoma, looking for Stop Pretending, the new record by the Pandoras. I figured my odds of finding it were high, since there was a big promo display for the record up on the wall.

I went over to the “Rock – Misc P” and flicked through the records. No luck.

I went up to the counter and asked the clerk if they had it. He said no – they’d gotten one copy, and another guy that worked at Tower had bought it. I asked them why they had the display on the wall, and he told me that the guy that bought the record really liked it, so he made the display.

Then I asked if another copy was coming in. No. Why? Because for records from independent labels, the buying policy was to send one copy to each store. If they needed more than one copy, then it had to be special ordered.

There are five forms of filtering, and this is an example heuristic filtering.

Heuristic filtering is rules-based, and this is a great example of a dumb mechanical process. It’s dumb because there’s no learning (“hey, people in Tacoma seem to like the Pandoras, send them more copies of the record”).

This approach worked fine as long as Tower was still the biggest aggregator around. The boycott of Tower that I started in response to this didn’t really seem to hurt them, even though I bought a LOT of records back then.

However, as soon as a bigger aggregator came along – various internet-based options – the Tower business model was toast.

People say that the internet killed Tower Records, but I think it was killed by bad filtering.

Second Story: Borders Books

In the mid-90s, I bought Science as a Process by David Hull, which became one of my all-time favourite non-fiction books. I bought it at the Borders in Westwood, which at the time had a superb science section. Back then, buying was decentralized to each store. So the Westwood Borders, just down the road from UCLA, had a significantly different selection from the Studio City Borders, and every other Borders in LA at the time.

This was expert filtering. Each buyer knew the kind of people that were shopping in his or her store, and they stocked books appropriate to that market.

Unlike Barnes & Noble, which appeared to use heuristics to stock their stores, each Borders was unique.

When Borders came to Australia and New Zealand around 2000, they had individual store buyers then too, so each store was still unique.

After the chain got sold, the individual buyers disappeared – replaced by a central buyer. This was done in response to the threat of online booksellers. The only way to improve efficiency was to cut down on staffing costs.

So Borders went to dumb heuristic filtering.

And now they’re gone too – also killed by bad filtering.

Third Story: Pulp Fiction Bookshop

I while ago I was browsing through the shelves at Pulp Fiction Bookshop here in Brisbane. They specialize in Science Fiction, Fantasy and Mysteries. Their selection in these areas is among the best I’ve ever seen.

A guy walked into the shop and went straight up to the counter. He said “My wife really likes Iain Rankin and Donna Leone. I want to get her a birthday present – is there a similar author that you can recommend?” The owner of the shop said “Yes, there’s a South African author (whose name I didn’t catch) that’s writing really good mysteries, but no one has heard of him (or her) yet. Try that.” The guy bought two books by that author, and left, looking pretty happy.

That’s expert filtering – both in terms of stocking the store and in terms of helping customers.

Even though people can buy books on the internet, and the Australian dollar is really strong, and the parallel importing laws here making it nearly impossible to sell books successfully, Pulp Fiction seems to be doing pretty well.

They’re doing well, because they filter well.


Simply calling these filtering problems is probably too simplistic. And yet, bad filtering definitely played a role in the death of Tower and Borders. Both of them were pretty good at aggregating. Borders was pretty good at using expert filtering to connect people with books they might like in-store, while Tower was less consistent in this area. For a while, Borders was pretty good at filtering, and Tower was always fairly bad at it.

The problems started when the internet killed their aggregation advantages. This caused Borders to do away with the one thing that actually made them distinctive – their expert filtering. Expert filtering is something that Tower never had.

Neither store ever was able to connect people up with products in the way that Pulp Fiction does. This type of expert filtering & connecting is better even than the algorithmic filtering you get at Amazon or iTunes.

The problem is that it doesn’t scale. So it’s hard to have a Borders-sized bookshop with great expert filtering. It’s easier if you specialize in something, as Pulp Fiction does.

To succeed in an information-based business, you must be good at aggregating, filtering and connecting information. And you have to be able to do all three. The stories of Tower and Borders show you how bad filtering can kill a business.

Student and teacher of innovation - University of Queensland Business School - links to academic papers, twitter, and so on can be found here.

Please note: I reserve the right to delete comments that are offensive or off-topic.

4 thoughts on “Bad Filtering Kills Businesses

  1. I have a really hard time separating filtering and connecting in that regard Harold. Firms definitely have to connect effectively to succeed/survive – there’s no way around it. It’s the aggregating and filtering that gives them something worth connecting, I think.

  2. Thanks for the link John – I missed that on Blogging Innovation. Definitely an interesting post – I talk about algorithmic filtering in this post:

    but I think that refining that category as you have is useful. I still think that in an ideal world there needs to be a combination of algorithm + personal filtering. Not quite sure to get there though.

Comments are closed.