Autonomy and IBM Watson – clever marketing doesn’t always win in the end

This week brought two news items that reminded those of us with long memories of some of the more egregious examples of over-marketing search technology. One is about a long running legal case that may result in a UK billionaire being extradited to the USA to face charges of fraud, the other the last gasp of a promising AI project from a global IT giant: the common factor is where the reality of implementation failed to live up to the hype. Yes, I’m writing about Autonomy and IBM Watson, two fallen giants of the search space.

Most of those in the search business will have heard of Autonomy, founded over 20 years ago in Cambridge, U.K. I’ve followed the company since the beginning, known many ex-employees and heard many stories, and more importantly worked on several projects around replacing the Autonomy software. Although Autonomy professed to being much, much more, it was at heart just a search engine (I once had the opportunity to confirm this with one of the original developers, someone you’ll never have heard of which is probably to his benefit). However, magical powers were often ascribed to the Autonomy software, for example it “understood concepts” – no, it really didn’t (remember this was many years before the current excitement around NLP, BERT, vectors and other ways to encode the meaning of language). It was certainly expensive, and clients told me it was hard to set up and difficult or indeed impossible to scale, and we successfully replaced some installations with open source equivalents.

Autonomy’s rise was undoubtedly a triumph of marketing, leading to Hewlett Packard’s astonishing $10bn purchase of the company. I remember the collective shock of everyone in the search business at this figure and the collective shrugs as the deal was written down by $8bn or so the next year – of course, none of the people involved in the deal had thought to ask us in the search business what we knew about this software and there was a general feeling that HP had bought a handful of magic beans. Many years later this has led to the Autonomy’s founder’s current legal troubles which he looks increasingly unlikely to escape.

My second example is IBM Watson, which famously won the US gameshow Jeopardy in 2011. I remember some of the early talks given by the Watson technical team, during which they explained how a combination of open source software (I think Apache Lucene was one component), custom coding and lots and lots of person-hours of training had made this amazing feat possible. A year or so later, marketing was in charge and the talks had all changed into promises of magical AI – and the technical team or indeed the details they gave were nowhere to be seen. Armed with these magical promises, IBM attempted to break into the notoriously hard healthcare market, but this week the remains of Watson Health have been sold off at a discount.

I should make the point that marketing is important, marketing is necessary, and at its best marketing is vital to explain why a company should adopt a particular technology. I head up the marketing team here at OSC and spend most of my time explaining search engines and how we think you should build and tune them. However, marketers need to remember that if you over-promise eventually someone is going to find out – the beanstalk goes nowhere and you should probably have kept the family cow. At worst, you or the people who believe you (caveat emptor applies of course) could lose billions of dollars or end up in court.

Open source is different

One of the advantages of working in the open source world is that if you don’t believe what we say about search technology, you can go and look under the hood and find out for yourself, or easily find someone else who has done this for you. I also believe that most of us working in open source search care about honesty, proof and what our peers think – so we avoid hype unless we can justify it. Publishing examples, showing our working and providing the code so you can run it yourself are all important.

Even if you don’t run on open source code, publishing open APIs also means that others can go and test your claims – here’s a great example of how a researcher recently proved how using a commercial API (confusingly from a company named ‘OpenAI’) could cost up to one thousand times more than the competition.

Gratifyingly, it’s far harder these days for companies in the search space to get away with the same promises as people might have believed ten or twenty years ago – and the stories of Autonomy and IBM Watson should be seen as salutary tales.

If you need help choosing the right technology and process to build amazing search, get in touch.

Image by Robot Vectors by Vecteezy