很庆幸，可以访谈到连网易CEO丁磊都推荐的《精益数据分析》作者之一Alistair Croll。一眼看去，Alistair 绝对是那种久经战场的“老兵”。谈话间，你会被他的睿智、经历所折服，为不时的玩笑和测试感到惊讶！
Alistair Croll（阿利斯泰尔·克罗尔）， 企业家、作家、演讲家，哈佛商学院的访问执行官。
从业以来，花费大量时间研究各种规模的组织如何使用数据做出优良决策、加速创业过程。职业生涯的大部分时间都是技术公司的产品经理，同时，在美国、加拿大等地运营孵化器和加速器，热衷于组织会议与活动，包括Enterprise Cloud Summit、the International Startup Festival、Strata、Pandemonio、Interop & Cloud Connect，是大数据、云计算、人工智能等领域全球顶尖会议的联席主席。
1999年，他创办用户体验管理先锋公司Coradiant，后被BMC收购。先后担任Google、DHL、Recruit Holdings、BBC、华为、Capital One、the White House、Unisys、Naspers、Teradata等公司的战略顾问。
著有畅销书《精益数据分析》、Complete Web Monitoring、Managing Bandwidth。
I know you've been involved in the understanding of how organizations can use data to make better decisions for such a long time.
Alistair: I would say I've been doing it as an employee, then entrepreneur, most recently as an advisor running accelerators. From a very early age, I was raised by scientists. So I always look at the scientific method: run an experiment, collect data and change behaviors. I think it's only in recent years that we have replaced opinion with analytics, because we have so many data cannot be ignored anymore.
How to guarantee data's accuracy?
Alistair: I teach a course at Harvard Business School called “Big Data and Critical Thinking”. People want to know data veracity. If you collect data yourself, it's probably accurate. After building instrumentation into your mobile device for your application, you'll know how many people are using it. But there're many cases where false data has been used to make big decisions. Two of these examples I used in the course are here.
When Greece entered the European Union, they provided a lot of financial information to EU. Someone analyzed the data using Benford's law and concluded all the data was false. Nobody listened to him at first but years later they found he was right.
Benford's law is so simple that you think it can't be right, which tells there're more numbers that start with one than those with two, more numbers that start with two than those with three…in many naturally occurring sets of data. If I give you all the transactions people have paid in accounts, the size of animals and their population, it'll not be random. You expected there to be the same number, but it's not the case. So you can just look at data to know if it's natural data sets or it's made up by Benford's law.
I give you the second example. The city Boston wants to measure where speed bumps are on the road. They created an App called “street bump”. You put “street bump” on your phone and place it on the seat-chair next to you. As you driving to work, it records where the bumps are. Then Boston gets a map of bumps, but what could be wrong with these data?
All the data just gives a map of where rich people live. Only rich people have cars with spare passenger seat, drive to work and have motivation to leave phone on. Even when your data is collected accurately, there may be kind of sampling bias. What you really should do is to think critically whether it's collected and processed in biased way. I think the least concern is like someone's feeding you with fake data.
Humans are definitely not rational. For example, we often buy products recommended by friends or family members even though we don’t need it. Rather than rigid analysis, we go through our day relying on past experiences or wisdom. How to balance data analysis and human elements when making decisions?
Alistair: The data you collected should tell you that humans are stupid. It will show whether decisions are rational or not. I don't think you have to balance data and intuition. Intuition is how you get a good experiment or idea. Then you use data to test it. There's no excuse to ignore facts. The question is what kind of experiment should you conduct based on facts. If you believe humans make bad decisions, get me data and show me the better way to conduct marketing campaign.
In order to find out the right metric, how to kick off those “vanity metrics”?
Alistair: You can actually turn a vanity metric to a real metric. If you concern about numbers of visitors to the website (supposed it's a vanity metric), ask yourself what you would want visitors to do. To buy something! Can you prove to me there's correlation between vanity metric (visits) and real metric (conversions)? The one you want should be something that drives your business model.
One of my favorite examples is about Nigerian spammers. Many people think if those mails were better written without word “Nigeria” in them, people will be more likely to believe. It's so easy to tell those mails are tricks.
There's a researcher from Microsoft who went to Africa and interviewed many spammers. The reason he found out why spammers insisted on putting “Nigeria” in the mail is that “Nigeria” helps identify the prospects they want. So they sent 1,000 mails and got 2 or 3 responses. But there was 70% chances that those people will send money. The vanity metric for Nigerian spammer will be open rate; the actual metric is “idiots” who send money to Nigeria.
Nigerian spammers are actually much better at marketing than I am, because they pass vanity metric of open rates to business metric of people who send money to Nigeria. If you were to measure the right metric, you would be able to tell how it has changed your business model. If anybody is focusing on vanity metric, then their job is to prove its correlation with business metric.
Could we use mathematical formula “business model + growth stage=real metric” to summarize lean analytics?
Alistair: The summary of lean analysis is if you know business model and stage of growth you're at, you can determine which metric is most important to you right now.
Is it enough with only these two dimensions?
Alistair: We think it's only those two, but that's probably oversimplification. Startups need some oversimplification, so I'm being OK with simplicity. What after that is asking yourself how to do experiment with that metric.
Except those six business models mentioned in your book, are there any other models?
Alistair: O'Reilly has asked Ben and I if we want to write another edition. We feel surprised and flattered by its success. One of the reasons that we might write another edition is that there're some new business models. Actually there's a chapter we didn't make into the book, because it's pretty thick to “Lean” books.
One of the models is crowd funding, like Kickstarter or Indiegogo, a place where you offer a product and many others back it. Charity and donation is a different business model. We think there's going to be a model around chat box. You could have conversations with AI or any other things. There's also a different model for virtual and augcumented reality. But we have specific ideas about what ecommerce would look like in virtual environment.
Other emerging business models could be like smart agent. It makes suggestions to me on my mobile device. What I'm hoping is that it could tell me useful things I want, not tell me things I don't know, and not fail to warn me about stuffs. We could do analysis on how effective the smart agent is.
So we think there are a few emerging business models, crowd funding, donation, virtual and augmented reality, smart agent that would be worth writing about.
Do we need some psychological background to conduct analysis?
Alistair: The field of behavior economics is fascinating. The book Thinking Fast and Slow by Daniel Kahneman talks about human's two models of thinking. If I say to you that I have a baseball and a baseball bat, together they cost a dollar and ten cents, and the bat costs a dollar more than the ball, how much does the bat cost?
You're puzzling. But most people will go immediately that the ball is ten cents and the bat is one dollar. That sounds like the easy answer. Then reality is the bat costs one dollar and five cents. In this situation, you're probably thinking more about it, for you don't feel pressure like being killed if you don't give the right answer. Daniel Kahneman in his book calls human's two ways of thinking system 1 and system 2. System 1 is about making impulsive and fast decisions based on guts and pressures. System 2 is very rational. Having understanding of two thinking models changes kinds of experiment you may run and ideas about bow to make market.
That's different. It doesn't mean you shouldn't use data science.
As an advisor, what kinds of founders do investors look for in your opinion?
Alistair: The first thing is to be able to bang out a prototype, because you want to build an experiment fast.
The second is not just focusing on products but on go-to-market strategy. I've run big Tech companies and know lots of engineers. If I tell you someone has given me 100 million dollars to build the next WeChat or Tencent, do you think I'm gonna succeed?
No, because no one is gonna care! The risk is “attention” and the unfair advantages that will get you grow. Here in China, Alipay and WeChat threw Hongbao into the chatrooms, which caused 100 million bank accounts added in one week. That's not a technological plan, it's a marketing plan.
The third one is to have founders who are subversive. That means they look at the normal way and find out other secret ways, which is usually a lazy way like shortcut.
Many big companies that we know today did something sneaky if you look at their secret past. Facebook send messages telling you someone has mentioned you on the internet. That's their hack. So I always ask people “can you iterate quickly and experiment quickly?” “Are you able to look at products and market?” “What's your sneaky unfair trick to launch products?” I f they don't have a good answer for those three, they're probably not going to be a successful entrepreneur.