Interview: Jeff Raikes - the Microsoft man behind Office and tablet PC software

"It even turns out with US versus UK English that we've found some degradation of [hand-writing] recognition accuracy."

Jeff Raikes is a 21-year Microsoft veteran and head - again - of the company's cash cow Office business. Here he is interviewed by our colleagues in California about the tablet PC, handwriting recognition and Office, among other things. Q: Microsoft Office has lots of features people never touch. Isn't it true, though, that you're still putting in more than what people can absorb? There's almost a disconnect.
A: There is a disconnect, but the disconnect is that people are supposed to use all those features. That's never been true and never will be true. We can say there's only about 10 per cent or 20 per cent of the features that we'll use, but your 10 per cent is going to be different from my 10 per cent. We learned a long time ago that what customers tend to really want is a set of tools that can meet their needs and be similar to the tools that others are using as well. Regardless of how much of the program's functionality they use?
The operative piece of information isn't whether you're using 20 per cent or 40 per cent or 100 per cent. The operative information is whether that the product does a great job for the 20 per cent and, with the new release, is there something in there that's important to me?
The trick isn't whether we can get people to use a bigger percentage [of the program features]. The trick is whether we're building the kind of things that are compelling enough to get them to move to the next release. But isn't it also true that Office remains a resource hog?
In terms of 30GB drives? Let's get serious here. But Microsoft makes software with big footprints. Doesn't it make sense to come out with smaller versions that use just fraction of the total features now found in Office?
That's where customer education comes in - we have to do a better job on that part. When it comes to footprints, there was a point in time when I would have agreed with you. But the size of hard drives has outstripped our ability to fill them up. You've been quoted recently as saying you expect to grow the Office business ninefold by 2010. Are you still sticking with that?
I don't really know how much I can grow the business. The way in which that [quote] has been replayed in most of the reports is that I somehow said I would grow the Office business by that much. I want to grow the information worker business, where Office is a part of it. It's about a $10bn business today.
For the growth we can achieve this decade, about one-third will be from continuing to grow and enhance Office, while two-thirds will come from creating new categories of application value and services to support information work. I see you've brought your Tablet PC with you. Where you think the next generation of handwriting-recognition technology is heading?
There's going to be a big powwow in December with Bill [Gates] where we're going to go through all this. Is there much debate?
There is, yes. On the one hand, it seems obvious, right? But the problem is that this is basically just a big database of inking samples. If you put new samples [of handwriting] into the database, are you going to improve the database or are you going to degrade the recognition?
I think we're going to get there. I really believe that. But there also are very smart, very reasonable people who would say that's the wrong approach. Is it reasonable to believe you can improve upon the technology to get to 100 per cent recognition?
Recognition is very dictionary-based. At times, it works beautifully, and you say to yourself, "Gosh, how does it figure it out?" You'll be inking and make a mistake and say to yourself that there's no way it can recognize this - but it does. We need to figure out ways to get more words into the dictionary. We also need to improve the way we recognise context. What do you mean?
It would be much better if our recognition approach could be more easily sensitive to the context in which someone is inking. If I ink something, and it doesn't get correctly recognised, we should relax the dictionary relationship on my second attempt. I think a large percentage of the times that it fails is because you're inking something that wasn't in the dictionary. You carry around a tablet during the day. What do you use handwriting recognition for? I use it in the same way people that use a laptop or the way they use a piece of paper. You take extensive notes on it?
Sure. I believe you can type faster than you can ink but what you can't do is be as expressive. For example, while you're doing your notes, try and type this [underlining a word on his tablet computer]. You just can't do that. The value is not that people can ink notes faster than they can type notes. The value is that they can do notes the way they have always done notes - and more, like the ability to search on notes. What's the profile of the person who you expect will use this stuff?
Here's my theory. If you can get [a notebook] for about $1,800 with long battery life and high screen resolution, and for less than a 10 per cent increment have your tablet [PC], I think a significant part of that market will find that a compelling value proposition. How about languages? Which ones ship first?
English, French, German, Japanese and simplified and complex Chinese. I know our Italian and Spanish guys are not happy with me. We're pretty good at doing stuff in lots of languages. But the recognition technology actually takes years to develop in a given language, because you have to accumulate the database of all those samples of ink. It even turns out with US versus UK English that we've found some degradation of recognition accuracy. And that just means we have to beef up the database language. What sort of devices do you think will be the most suitable for handwriting recognition?
I'm a big fan of the pocket-sized form factor because it's something you can actually stick into your pocket. I also think you'll want a device for note taking. I think when it gets to 12.1-inch screens, then you'll feel: "Wow, I'm really writing on piece of paper."
The other one is something that we're working on at Microsoft Research that we call 'broad bench'. Think of it as three 20-inch LCD displays treated as one display. For the second part of this interview see: http://www.silicon.com/a55803 .