Saturday, September 13, 2008

SEM on five dollars a day

How do you build a new product with constant customer feedback while simultaneously staying under the radar? Trying to answer that question at IMVU led me to discover Google AdWords and the world of search engine marketing.

SEM is a simple idea. You declare how much someone clicking an advertisement is worth to you, and then the search engine does its best to get you as many clicks as it can at that price. There's a lot of complexity that I'm leaving out, naturally, because I want to stay focused on the simple idea at the center of SEM: that you can pay peanuts to have people come to your website. In a mature company with a mature product, the goal is to pay for lots of people to come to your website. But I think the genius of Google's innovation is that it allows you to pay for just a few people. Think of it as micropayments for beta testers.

My first AdWords campaign was limited to five dollars a day, and we were buying clicks at five cents a click. That yields 100 clicks a day, every day. Probably if I had been an experienced marketer, I would have known that tiny volume to be insignificant, and I would have been embarrassed. Luckily, I didn't know any better. 100 clicks might not sound like very much, but look at it this way: it meant that every single day, 100 human beings were coming to our website and being offered our product.

We expected that many of those people would buy our product (that's why we charged from day one). But anyone who had done direct response marketing before would have known better. At first, zero people bought anything from us. We tried tinkering with the payment system. Still zero. Maybe the problem was helping people find the payment system. Still nothing. We kept working our way backwards, until we realized that nobody was even making it past the first landing page. Oops. Slowly, over time, we optimized (or eliminated) each step in the process of becoming a customer by giving us money. And one day a remarkable thing happened: we started making more than five dollars a day in revenue.

In the process, we had also built a simple cohort-based analytics system. Its simplicity made it effective - everyone in the company could use and understand it. It just answered this question: for any given time range, for the 100% of people who registered in that period, what percentage of them downloaded? chatted once? chatted five times? bought something? That simple funnel analysis became our scorecard, and helped us refine our product with constant customer input.

After we were making more than five dollars a day, we could take the profits and reinvest them in raising the budget. As we would find more keywords to bid on (always bidding the minimum five cents per click), sometimes the increase in volume would drive our funnel percentages down. When that happened, we'd stop raising the budget, and keep optimizing the product. In this way, we gradually built out a more and more mainstream product.

So how do you find those initial 100 clicks a day? We started by using features of our product as keywords, but this was of limited volume. Eventually Steve Blank, one of our early investors, suggested a technique that not only increased the number of clicks, but also started to use AdWords as a learning and discovery tool. We ran ad campaigns against every single product we could think of in an adjacent market space to ours. We tried obvious competitors as well as long-shots. Since we were only paying per click, it didn't cost us anything to cast a wide net. We would pretty much bid on any phrase that was "[name of competitive product] chat" and variations like that. And then we would use that simple analytics system I mentioned to monitor the conversion rates of customers from each campaign. Those rates gave us a map that told us a lot about our customers; insights that proved stable even when the company grew orders of magnitude bigger.

Only much later did I realize that this was an application of customer development to online marketing. It's now a technique I recommend for any web-based startup.


  1. I just came across this post. You mention

    "we had also built a simple cohort-based analytics system"

    Is tihs a commercial/open source systems.

    Where can one get it, and for how much?

    Thank yo u

  2. Have AdWords rates increased too much for most common phrases like this to have the numbers work?

  3. @BillSeitz - yes, but that was already true back in 2004. The absolute essential key is to bid only on low-volume low-value keywords. We're only looking for 100 clicks a day, remember, so by bidding on a couple hundred keyword phrases (long ones are great, since they are uncommon) we can get the desired volume by finding keywords that only have a click every other day.

  4. Thanks so much for sharing this information. I've been working on a small project with some friends that we'd like to make into a sustainable business. There is much work that I need to do (the only developer so far) before we have something customers can use.