Powerset, the much-hyped natural-language search company, has finally launched a public product: a showcase for its search technology that "enhances the Wikipedia experience." It's live right now on its homepage, so go check it out.
Are you back? That sound you heard is the technology world shrugging in unison. For all the hype Powerset has gotten over the last year and a half this showcase leaves a Chicxulub-sized gap between expectation and execution.
Even ignoring all the press, it's not that impressive on the face of it. Using their example queries as templates it took me about five seconds to find queries which not only returned appropriate results on Google but simultaneously returned nonsense on Powerset.
On a personal note, I really wanted to like Powerset. The people working there are all super-smart and I know they've put a lot of blood and sweat into this launch. But I have to be honest. If someone over there reads this just know I do it because I want to see the company launch a great product.
A Failure of Execution
Let's start by diving into a little Google vs. Powerset one-on-one.
| Query | Winner | ||
|---|---|---|---|
| Who is on Google's board? | Powerset | Powerset | |
| Who is on both Google and Apple's board? | Powerset | ||
| How did Hitler die? | Powerset | ||
| How did Adolf Hitler die? | Powerset | Tie | |
| What is the longest suspension bridge in the United States? | Powerset | ||
| When did the United States enter Iraq? | Powerset | Powerset | |
| Who was the tenth President of the US? | Powerset |
You get the idea. I tried to pick questions that Powerset is designed to answer, i.e., fact-based trivia easily found on Wikipedia.
Some of the failures are pretty egregious, honestly. Searching for "Who was the tenth President of the US?" fails to return a single relevant result in the first page of Powerset whereas Google's entire first page is relevant, even limiting Google to searching just Wikipedia.
In other cases it appears Powerset has a poor understanding of synonymy, returning irrelevant results for "How did Hitler die?" but returning the correct answer for "How did Adolf Hitler die?" Of course, Google returns the correct answer in both cases.
Guys, this is supposed to be the exact area where you excel. What gives? You're not even living up to your new, lowered expectations.
A Failure of Marketing
Marketing is a two-edged sword. The net effect of good marketing is to solidify your brand in the minds of consumers. This is good if you execute, but can also make it difficult to change course later.
Powerset fell into this trap. They started with ambitions of being a Google-killer, but reading their about page now it sounds more like they're aiming to be the Google-enhancer. This is a respectable business, of course, but it is hard to swallow after a year and a half of being promised a revolutionary new search paradigm.
They might be repudiating the Google-killer label now, but here's an excerpt from a February, 2007 press release:
”The time is right to tell the world about the game-changing technology we’ve created,” said Ron Kaplan, Powerset Chief Technology and Science Officer, who previously created and managed the Natural Language Research Group at PARC. "I am glad to join Powerset’s team of world-class linguists and search engineers to help this technology revolutionize the way people access information."
Too much marketing before a product launches can back you into a corner and in Powerset's case it will be difficult at this point to avoid being compared directly to Google in the press.
Conclusions
Powerset was started in 2005 and has been using Xerox PARC's natural language processing technology for over a year, now. In that time they've been pumping out press releases talking about how they will revolutionize not just search but the way humans and computers interact.
What do they have to show for it? Not much, judging by their latest product. As a search tool it is more interesting than useful, shining in only a few, pre-selected cases. The advantages over Google are so minimal and the defects so large that I would never consider using this as my main means of searching Wikipedia, let alone the Web at large.
To me this product smells like a tech demo, not a fully-featured product launch, intended to convince someone outside Powerset that they really are producing something amazing. There are rumors that Powerset is looking to sell or raise another round of financing, and have recently hired David Wehner, a managing director at Allen & Co.
This launch might be enough to convince investors to re-up or buyers to fork over the dough, but speaking as an end-user I'll take another look at what Powerset has to offer when it can tell me who John Tyler was.
Wow, great analysis!
What piqued me the most about this launch was how unexciting the demo video was. I’m sure there’s some amazing software at work here…it’s just not blatantly apparent (from the video or casual use).
Thanks for writing,
Ethan
[…] Today, Powerset launched an impressive PR volley that generated incredible noise through out both the mainstream press and the blogosphere alike. The HackerNews front page was littered with the requisite comparisons, pitting Powerset’s wikipedia search against Google and Hakia. But at EOD, the reaction was a resounding “meh.“ […]
nice analysis. thank you.
[…] service only uses Wikipedia and Freebase …ReadWriteWeb - http://www.readwriteweb.com/|||Powerset Launches, Verdict: Meh.Powerset, the much-hyped natural-language search company, has finally launched a public product: a […]
Thanks for the detailed write up. We have a few UofC grads here too.
I think you missed two key parts of Powerset, though. First, when you type in a regular topical query, Powerset extracts Factz from across Wikipedia. This is a unique feature enabled by our semantic processing that ya just can’t get in a regular keyword engine. Also, take a look at our enhanced Wikipedia pages. Note that we also have Factz on the page, which often provide a concise summary of the topic.
I signed up for your KallOut beta. I want an invite =)
Mark,
I know. In fact, I interviewed there last year, on the day of Scott’s demo at sfBeta. Adonomics was acquired before I finished the process. Say “hi” to Tim, Linda, and Andrew for me.
For the sake of argument let’s ignore the fact that you’ve been hyping yourselves as a revolutionary new search paradigm for the last two years and look at your “enhanced Wikipedia experience” on the face of it.
Factz just isn’t that compelling. I’ll even use one of your hand-picked examples, Henry VII.
Ok, so I’m in a mindset where I want to learn about Henry VIII. The first thing I’d normally do is type “henry viii” into Google. The first result is the Wikipedia page. I look at the Wikipedia page and get all the salient facts I could want.
With Powerset the only real savings I can see is that I’m spared the click over to Wikipedia. Is that revolutionary? In fact, it requires I change my search behavior to get value, which is an almost sure-fire way to damn a product.
There’s some value above Google, but remember: success equals execution minus expectations. Coupled with the fact that plenty of fact-based queries (e.g., “Who was the tenth President of the US?”) return absolute gibberish, there’s just no way I’d use Powerset.
From my perspective it feels like this was a tech demo designed to secure additional financing, not a fully-featured product in its own right. Maybe you should ask Barney what the strategic purpose of this release was.
Also, I no longer work at KallOut, so I can’t get you in on the beta.
[…] loudly launched an Alpha this week, covering Wikipedia and Freebase content. Some were unimpressed, some have covered the entire hoo-haa in gigantic, but brilliant, blog […]