Turing Lecture 2013
Suranga Chandratillake toured the BCS/IET Turing Lecture in February 2013 speaking at the universities of Cardiff, Manchester and Glasgow as well as at the IET in London © IET
How to build a technology company and take it to market
When the story is told of how a product came into being, it is often given from the point of view of the inventor. Suranga Chandratillake FREng, Founder and Chief Strategy Officer of blinkx plc, thinks it equally important to describe the journey of the entrepreneur. He went on a tour of UK venues in February 2013 to deliver the Turing Lecture describing how technologists (engineers, scientists and programmers) can become part of the process that takes products to market.
A couple of years ago, Lord Alan Sugar said on BBC Televison’s The Apprentice that he’d “never met an engineer who can turn his hand to business.” His statement seemed to ignore the 15% of FTSE 100 companies that have engineers on their boards. And, if he had looked towards Silicon Valley, where I am currently based, he would have seen a lengthy list of influential technology companies that were founded by engineers who built the technology and stayed on to run the business as CEOs.
It is clear that technologists are perfectly capable of developing and running companies. In fact, in the case of innovative, technology-led business, there are compelling data that suggest they are the best people to do this. Yet many shy away from the path. In the UK, our education system does an excellent job of equipping technologists with the rigorous, analytical skills necessary to enable them on the professional journey of being an inventor. But, given how few of us end up doing it, appears to do a significantly less successful job of equipping them for the journey of the entrepreneur.
I read computer science and worked in low-level research and development before being one of the team that founded blinkx. After a brief spell as the Chief Technology Officer, I took the role of CEO and, over the following seven years, built a publicly-listed, global company employing 300 people. At first, I was reluctant to take on this leading role, but through the process of building the company and taking it to market, it became clear to me that technologists and businesspeople share many of the same approaches to solving problems.
Google worked out that web navigation was all about links. The links that existed between pages are a measure of authority on any given topic. If there are 10 web pages that mention the word ‘dog’, but one of them is linked to 100 times and the others are linked to just once each, then the one that is linked to 100 times is probably a significant web page about dogs. Google then used that simple observation to create an algorithm which could trawl the web and used, essentially, human-created links to infer a sense of relevance
THE CHALLENGE IS SET
When consumer broadband internet became fast enough to distribute video content in on-demand, streaming form, the web exploded with internet video sites and content. Between 2004 and 2006, hundreds of companies sprang into existence, providing users with a plethora of videos on just about every topic imaginable. The problem, however, was that this content existed in a fragmented reality, spread across thousands of sites with little standardised organisation or order. From a user experience perspective, it was hard to know where to go to find a video. This fragmented video content issue was very similar to the fragmented content issue that was posed on the text web, with normal textual web pages.
The solution to the text fragmentation problem turned out to be search. Web search and, in particular, Google’s groundbreaking approach of paying attention to not just the words on a webpage but also the links between those webpages, created a navigation tool that allows users to find the text content they want.
The Google search model didn’t happen overnight. People initially started by trying to manually categorise web pages into a taxonomy which gave rise to Yahoo!. Later methods involved creating simple reverse look-up indices, like Infoseek and Altavista, that had little concept of relevance or importance. It took years of non-optimal ideas to develop a search engine with the intuitive relevance that Google built up and still provides today.
Video search was no different. The first generation depended entirely on manually generated, editorial descriptions of what videos were about. This worked well for professionally made shows that already had Radio Times-style editorial written for them, but really badly for amateur self-uploaded content. The next generation used textual metadata provided by the uploaders such as titles, descriptions, tags and date of upload. The popularity of the video based on the number of times it had been viewed was also used as a measure of relevance. These approaches failed because, ultimately, they were relying on a secondary piece of evidence, the text, rather than considering the content itself, the video.
A NEW APPROACH
blinkx changed the market by introducing ‘deep-indexing’. Our technology uses speech recognition and visual analysis to allow technology to watch and listen to the videos it has indexed, extracting information, word-by-word and frame-by-frame from the video itself. Our speech recognition uses a hidden Markov-based model, using statistical pattern recognition, which can transcribe and create phonetic transcripts of speech in around 30 different languages at an extremely high rate of accuracy. It can do that regardless of accent and regardless of content of the actual speech, so that blinkx can recognise speech across different types of jargon.
The system also uses visual analysis with computers watching the videos, frame by frame, so that we know when scenes start and stop. This is similar to creating paragraphs in prose and is a neat way of breaking down long videos. In addition, blinkx can read the text on the screen and uses face recognition, so that well-known individual faces can be flagged up and identified.
As well as powering the video search, we flipped this algorithm on its head and used it to solve the advertising problem: Which advert is most relevant to a particular clip of video? The search and advertising innovation we built at blinkx enabled the company to be a user success. The business success came from applying good entrepreneurial practice to the smart innovation.
HOW I BECAME AN ENTREPRENEUR
blinkx was initially developed out of research being done at the Cambridge-based enterprise search and knowledge management software firm Autonomy, before it was split off into a separate entity. The Autonomy management team and Mike Lynch OBE FREng in particular – the founder and CEO of Autonomy and himself an engineer – asked me whether I would like to be the CEO. I argued that my qualification of a computer science degree and time as a researcher didn’t equip me for the role. I kept saying no, in increasingly creative ways, for about half a year, but Mike is very persuasive and eventually I agreed to do it.
I took the role with one important caveat: experts from Autonomy would have to be on hand to advise me on finance, marketing, sales, human resources, and other areas I saw as being key to being a good CEO. Autonomy agreed as they were going to keep a stake in the business – the company still owns more than 10% of blinkx.
As a result, I enjoyed a unique career experience. I got to be a CEO and run my own company while, at the same time, behind the scenes, I had access to a grade-A global team who could provide me with their expertise.
One of the first areas I stumbled into, and felt ill-equipped for, was marketing. In the early days of the company, before we had fully focused on video searching, we actively marketed a product called blinkx Pico. This was a piece of downloadable software installed on a person’s Mac or PC. It would sit in the background and, when a person went to a web browser and typed in any URL or web address and went to that page, a small toolbar would appear in the top right hand corner. Pico would identify the web page the user was reading and automatically fetch them associated news articles, web pages, encyclopaedia articles and so on. These links were supplemented with ads relevant to the article they were reading.
We were able to give Pico away free because we made money from the ads. We found we could show about four of them a day on average and based on how long a user typically kept the product, how many ads they clicked on and the average value of those ads we calculated the lifetime revenue per user (LRPU) to be $2.20.
The primary advice from Autonomy’s marketing gurus was that while loss-leaders were acceptable during the launch of a product, ultimately the marketing costs had to be less than your sales price. Our sales price, effectively, was $2.20, so we were free to experiment with different ways of marketing to gather users as long as the aggregate cost of the people we found was less than that amount.
The marketing for Pico involved three main thrusts: PR, banner ads, and search or keyword ads. We found that PR was extremely expensive. There are large, up-front costs which included hiring agencies and paying for press tours. Even though we ran a successful campaign with almost 100 articles being written and many people making it to the Pico website, the cost per user acquired was over double that of sales – PR-driven users cost us a staggering $5.60 each.
Banner ads appeared to be much, much more effective. We bought ads that appeared on hundreds of websites at a cost of less than a cent per ‘impression’ with users who clicked on the ads coming to the Pico site. Initial estimates looked very promising: we were getting users to our site at a cost that was equivalent to around $2 per user, which made it a profitable exercise.
However, it turned out that the site conversion ratio was significantly lower than in the case of PR. Only 10% of the users who came to the site actually downloaded the product. Why was that? It turns out that when people read a PR article, they get to know a good deal about the product and by the time they get to the website, there’s a good chance that they’ll like the product, they are, in marketing jargon, well-qualified prospects. However, a banner ad that is viewed for a few seconds can’t convey a sufficient amount of detail and many people would click onto it and decide it wasn’t what they really wanted, and so would not download it. The actual cost per user was therefore $5, which was pretty horrific and almost as poor for us as PR.
We then tried search advertising. This is the advertising that sits next to Google search results. Here, a company doesn’t pay for the impression or an agency but per click – 45 cents per click, in fact. However, the site conversion ratio is really high. This is because the user has already searched for something in particular, then found an ad that was relevant to that search and consequently clicked on it. That is already someone who is self-selecting to be a very likely customer. In our case, we found that around half the time they would download Pico and therefore our cost per user was just 81 cents – making search advertising very profitable.
Stepping back from the day-to-day marketing process, I quickly realised that marketing is actually just a fairly complex multivariate modelling problem. In fact, today, the most effective marketers are people with a high degree of understanding of statistics and even probabilistic modelling. They will run many tests with small changes in variable settings, logging the difference in outcomes and, over time, refining the marketing ‘spend’ towards the channels that are working the best. It turned out marketing had a lot in common with computer science.
Interestingly, as a footnote, PR ended up being very profitable because of the non-mathematical reason that PR can gain access to important people. Through PR we came upon a particular expert at the Ask Jeeves search company. He came to us, having used our software, and said, “You guys should do video search. We need help in that area.” That ended up setting the course for blinkx to connect online video viewers with content publishers and distributors, using advertising to make money from those interactions. This product and business model ended up being worth much, much more than the Pico product line and, over time, became the foundation for blinkx becoming a growing, successful company.
One of the most complex areas in finance is raising money. There are a bewildering number of ways that this can be done, one of the most complex of which is to go for an Initial Public Offering (IPO), or stock market launch, which blinkx undertook very early on.
While the process appeared daunting at the outset, it turned out that going through an IPO follows a number of fairly well-structured steps. First, one needs to know how much one wants to raise. We calculated that we needed about $50 million for blinkx to reach the point of profitability, or break-even, based on how much product engineering we needed to go through, how we needed to test it, and the size of the sales force we needed to deploy the product at a large enough scale to start making money.
The next step is to value the business. Valuation is set by analysts who work in the City for investment banks. In our case, they looked at blinkx’s business, which at the time was primarily a video search engine that made money by placing ads viewed during or after the videos. Then they started to project, based on current numbers and growth rates, how the number of video searches and the revenue we generated per search would change over the coming years.
A variety of reality-checks were undertaken looking at similar companies – other video companies, other search companies – and comparing their growth rates with ours. The analysts were able to use these growth rates to calculate our likely sales revenues each year for the following five years. They then subtracted the costs we were expected to have in those periods and were able to generate our profit profile over time. This illustrated which years blinkx could be expected to generate losses, profits and what scale those losses and profits would be.
It turns out that the primary method of valuation of a business is known as the Discounted Cash Flow (DCF) – Wikipedia gives a good explanation of this. The maths behind it is fairly straightforward and the audience at each Turing Lecture I gave this year was able to derive the method from first principles in about five minutes. While the DCF is not a complex tool, especially for those with a quantitative background, it is one that most technologists seem to be unaware of.
In our case, the analysts examining blinkx arrived at a valuation of around $250 million, meaning that we needed to sell 20% of the company to raise the $50 million required. We held meetings with people who have access to lots of money – typically pension and hedge fund managers as well as people who work in the City – and told them about the company. After doing that for a few weeks, the bank gathered responses, and found that enough of these investors were willing to buy sufficient shares at the price blinkx had valued them, so the stock market float could go ahead. I am happy to say that we had a very successful Initial Public Offering and sold 56 million shares, smoothly raising the $50 target.
What the experience taught me was that my computer science degree had prepared me perfectly well for the financial aspects of being a CEO; I just hadn’t appreciated that I had this aptitude and preparation. As with the marketing, once in the trenches, the rigorous, analytical, quantitative background of my degree had left me in excellent stead.
From finance to marketing, to HR and even organisational management, computer scientists, engineers, mathematicians and natural scientists are significantly better equipped for the needs of modern business roles than they sometimes realise. In the end, the skills that I needed to be an inventor were the skills I needed to be an entrepreneur. Computer science degrees teach you to be quantitative, to be rigorous and to be analytical. It turns out that these are exactly the fundamental skills you need to be a really good businessperson.
There is no mystery to business and it is really all the same sorts of things, simply packaged in different ways and applied to a slightly different toolkit. However, the fact that I refused to take the CEO role for almost a year and the fact that Lord Sugar can make remarks like the one he made show that there is a gap.
While technologists are ready for these business roles, society at large (certainly in the UK) and the technologists themselves don’t appear to be aware of this. I call this gap of awareness the Boffin Fallacy. This is the fallacy that technologists do not understand business, and that it is something that they can’t do – something they won’t enjoy and something which, occasionally, they even believe is a bit beneath them. I think the Boffin Fallacy – more a societal and psychological issue than a gap in hard skills – is the key reason why many inventors fail to cross the divide into becoming entrepreneurs and is a problem that needs to be overcome if more are to do so.
I ended the lectures on this positive note: there is no doubt that the UK with its great education system has, for generations, been the birthplace of amazingly talented inventors whose creations mean the country continues to punch above its weight in innovation and technology. The tragedy is that, for a very long time now, we have failed to help these inventors follow through, to help them create and run the businesses that exploit these inventions. Data show that these technology/founder CEOs do better than their hired ‘businessperson’ counterparts, so when the Boffin Fallacy gets in the way and stops them from making this leap, we end up limiting the positive commercial, societal and even technological impact of their ingenuity. We end up selling innovation short.
To see the IET/BCS Turing Lecture, in conjunction with the Oxford University Press, go to http://tv.theiet.org/technology/infopro/16066.cfm
Keep up-to-date with Ingenia for freeSubscribe
Other content from Ingenia
- Environment & sustainability
A young engineer’s perspective on the good, the bad and the ugly of COP27
- Environment & sustainability
- Issue 95
How do we pay for net zero technologies?
- How I got here
Electrifying trains and STEMAZING outreach
- Civil & structural
- Environment & sustainability
- Issue 95