Searching for a Better Recommendation Engine

My first experience with recommendation engines was with Amazon in its early days. Then, when you bought a book, Amazon would tell you that people who bought the same book had also bought these other books. It was simple, brilliant, and most importantly, it worked. When Amazon later started selling CDs, the recommendation engine worked even better. I got to enjoy music I never knew existed, and Amazon sold more CDs. It’s a classic win-win, and you would think Amazon would put its substantial resources into making its recommendations even better. 

But apparently not. After buying an introductory book on Photoshop a while back, the recommendation engine started showing me every Photoshop book ever written (there appear to be hundreds of them), and crowded out every other book recommendation for nearly a year. These were lazy recommendations, and disproportionate to the one book I bought – ever – on a specific topic. And Amazon recommendations have gotten even lazier since then.

You may also recall the Netflix Prize, announced with great fanfare back in 2008. A $1 million prize was given to anyone who could improve the efficacy of the Netflix recommendation engine. It was an impressive commitment by Netflix, and it showed they deeply understood the importance and value of recommendations to their business. Fast forward to today. Having watched every single episode of Arrested Development on Netflix, how did I learn about the arrival of new episodes? I read about it in the newspaper. Has Netflix brought these new episodes to my attention? Not yet. Somewhere along the way, Netflix seems to have stopped caring about the quality of its customer recommendations.

Move over to the search engines – all of them. You may know that you can force a search engine to search for a specific phrase by putting quote marks around it. Typically, your first search results will be web pages containing that exact phrase. But then the search engines actually remove the quote marks and toss in results that have the requested search terms, but not necessarily together. Then they toss in pages that have some but not all of your search terms. Since I didn’t ask for these search results, I think it’s fair to consider them as recommendations. And they are (predictably) lousy. It’s as if the search engines assume I don’t know what I am doing, so they give me every possible type of result. Yes, more is better with search engines, but only if they are giving me more of what I want. 

Contrast this with the music service Pandora that I’ve been raving about since 2007. Despite a tough revenue model, Pandora has not forgotten that it lives and dies by the quality of its recommendations, and it’s built to over $1 billion in annual revenue by staying focused. Hopefully they'l maintain that focus as it continues to grow.

When companies get big, it’s very easy for them to get distracted and lose interest in what made them big in the first place. There are more voices now saying that Google search quality is in decline. And remember when Yahoo got bored with search and decided to outsource search while it chased bigger dreams? These distractions create opportunities for smaller players to do search better, and some are finding success.  

The Low Hanging Fruit Hiding in Plain Sight

One of the unintended consequences of the rapid shift to sales force automation tools, CRM systems and large-scale lead generation campaigns is that things only work well when you target prospects and they respond to your promotions. It’s an outbound world now. Pity the poor prospect who unprompted calls you to buy something!

I have recently been in that position, having to make sales inquiries to data companies on behalf of clients. At first, I simply bemoaned the quality of salespeople these days. But then I realized it wasn’t the salespeople who were the problem; it was me! None of these companies had put any thought into how to handle an unsolicited lead, probably because they assumed it was a non-issue. But it’s a big issue. I consistently fell through the cracks because none of these companies had made any provision to deal with me. I didn’t fit their workflow.

The first thing you learn about being a buyer in this situation is that you better not be in a hurry. Callbacks to unsolicited leads in my recent experiences ranged from two to four days. And when I did get a response, it was often by a screener, charged with determining if my business was worth a salesperson’s time. Indeed, after being screened by one major data provider, I received a surprisingly curt email informing me that the size of my potential order didn’t merit their attention, but that my name had been passed along to one of their distributors, and I would hear from them in due course. I’m still waiting after three weeks.

I’ve also learned that using the phone doesn’t accelerate the buying process at all. In fact, it makes things worse. Two of the data companies I contacted had automated attendants that would helpfully connect me … but only if I already knew who I wanted to talk to. In one case, I actually reached a live person who answered the company’s main number. When I asked to speak to someone in Sales, I got the response I hear nearly 100% of the time: there are no salespeople in the office. When I asked to leave a message for someone in Sales, I got a long pause, followed by a very hesitant and somewhat dubious “sure, if you really want to.” One receptionist actually made the mistake of connecting me to someone in the sales department. I say “mistake” because the person answering the phone said he “wasn’t allowed to talk to me,” but he’d have someone call me back. When I said I needed some basic product information first, he did in fact provide it, after swearing me to secrecy because “I could get in a lot of trouble for doing this.”

Since companies have clearly abandoned the telephone as means of inbound contact, you think they would pay close attention to incoming leads by email. If only that was true! After submitting my sales inquiries to three companies via the ever popular “contact us” form, proving that I was not a robot, and in some cases being asked the size of my budget (required field), I sat back and waited. And then waited some more. One company responded fairly quickly, but the salesperson was apparently so incredulous that a sales lead would be unsolicited that I had to submit to a grilling via email to confirm my interest and my bona fides.

The second company responded three days later, and apologetically asked for lots of information about my product requirements and me so that he could “get me in the system.” Once properly in the company’s lead stream, I had a satisfactory buying experience.

The third company? Three weeks and I am still waiting on a response.

You surely know where I am going with this: with so much technology and so many resources being devoted to lead cultivation, generation and management, we seem to have forgotten about the most valuable sales lead of all: the unsolicited inquiry. There is apparently no place for them in our automated workflows.

Not your problem? I challenge you: complete the form on your own company’s “contact us” page and sit back and wait, not with a stopwatch but with a calendar. If you want an even more dismal experience, call your own company’s main number and ask to speak to a salesperson. Yeah, it’s that bad ... which means the opportunity for quick increased revenue is that good!

Zagat: Down But Not Out?

This week, Google announced that it had sold its Zagat guide business, for which it paid a stunning $151 million in 2011, for “an undisclosed amount” to a company you’ve likely never heard of, The Infatuation.

It’s an ignominious development for the former household name brand, and true pioneer in the data business. Well before the Internet, Zagat had blazed new trails in the area of user-generated content and consumer reviews. Tim and Nina Zagat, the founders, proved to be creative and talented marketers and self-promoters. For many years, the name Zagat was synonymous with restaurant ratings in the United States.

But the Zagat empire was print-based. Moreover, the Zagat business model depended in large part on selling bulk orders to companies with their names and logos on the covers, to be given away as gifts. That made it difficult for Zagat to economically expand its coverage beyond the largest cities, so it never became a truly national data provider. And its attempts to expand into other segments of the hospitality industry where more competition existed, fell flat. But the Zagat brand transcended all these shortcomings.

And it is the brand that Google apparently paid so much to acquire. Everything about the Zagat business was at odds with the Google model and ethos. I trashed the deal at the time.

The irony is that Google couldn’t even find an effective way to leverage the Zagat brand. Fortunately, the $151 million purchase price is a rounding error for Google.

What’s the future of Zagat? I do believe the brand still has some life, and could be resuscitated. There’s a role for well-curated, tightly edited slightly snarky, user-generated content that helps you decide where to eat – especially when coupled with a restaurant booking engine. That’s why I would have been much more excited if Zagat had been acquired by, say, OpenTable.

Looking for New Product Ideas? They're Not All in Your Head

Part Three.

For many information and media companies, the preferred way to develop new product ideas is to brainstorm them internally. Get your best minds in a room and talk about the industry and its needs. You can conduct these sessions in a highly structured way or make them completely freewheeling and open-ended. Good, solid ideas can result.

Brainstorming sessions are both convenient and efficient. And if your staff is deeply engaged in your market, bringing them together to discuss new product concepts can yield powerful, even electrifying results. That’s because your staff is essentially reporting back what it is hearing and seeing in the marketplace. Synthesizing their different inputs, finding themes and conceptualizing solutions to problems is a great group activity, and resulting new product ideas can be very strong indeed.

Contrast that with companies that aren’t close to their markets. Their group brainstorming sessions will yield bigger product concepts (arguably bigger opportunities, but also riskier and harder to execute), incomplete concepts (based on lack of detailed market knowledge), and little certainty about market appetite. Perhaps most significantly, these product concepts, because they tend to be bigger, somewhat amorphous and without clarity as to market need, rarely get developed further.

My bottom line view of new product brainstorming is that it works, but the output can’t ever be better than the input. If your staff knows your market, they can effectively act as customer proxies, and the results can be compelling. If your staff doesn’t, brainstorming results in pipe dreams.

Workflow Elimination

The power of embedding one’s data product into a customer’s workflow is well understood by data publishers. Simply put, once a customer starts depending on your data and associated software functionality, it’s hard to cancel or switch away from you because the customer’s work has become designed around your product. It’s a great place to be, and it’s probably the primary reason that renewal rates for data products can sometimes verge on 100%.

But should workflow embedment be the ultimate objective of data publishers? This may depend on the industry served, because we are starting to see fascinating glimpses of a new type of market disruption that might be called “workflow elimination.”

Here’s a great example of this phenomenon in the insurance industry. A company called Metromile has rolled out an artificial intelligence system called Ava. What Ava does is stunning.

Auto insurers using Ava require their policyholders to attach a device called Metromile Pulse to their cars. As you may know, virtually all cars now have onboard computers that log tremendous amounts of data about the vehicle. In fact, when your local auto mechanic performs a computerized diagnosis of your car, this is where the diagnostic data comes from. Metromile Pulse plugs into this onboard computer. The device does two things for insurance companies: It allows them to charge for insurance by the mile, since the onboard computer records miles driven and the device transmits them wirelessly to the insurer. That’s pretty cool and innovative. But here’s what’s mind-blowing: if a policyholder has an auto accident, he or she can file an online claim, and Ava can use the onboard data to confirm the accident, re-construct the accident using artificial intelligence software, and automatically authorize payment on the claim if everything checks out, and all this can be done within a few seconds. The traditional claims payment workflow hasn’ just been collapsed, it’s effectively been eliminated.

How does a data publisher embed in workflow if there’s no workflow? That’s a problem, but it’s also an opportunity, because data publishers are well positioned to provide the tools to eliminate workflow. If they do this, and do this first, they’ll be even more deeply embedded in the operations of their customers. And doubtless you’re already thinking about all the subsidiary opportunities that would flow out of being in the middle of so much highly granular data on automobile operation.

“Workflow elimination” won’t impact every industry quickly if at all. But it’s an example of how important it is to stay ahead of the curve on new technology and always seeking to be the disrupter as opposed to the disruptee.