042. Clippy, The F*cking Clown

“Though nothing more than a thin metal wire, Clippit will help find what you need and keep it all together.” –Clippit being introduced in Office 97

As a company gains success and grows, taking risks becomes, well, riskier. The costs of failure come front and center, as the ability for a company to play out scenarios where something would not work overwhelms the naïve optimism that used to characterize efforts. It is like one day, suddenly, everything becomes more difficult and scarier. Clippy, née Clippit, needs no introduction as the failure, the evolution to kitsch, and the resurrection as a technology ahead of its time have been baked into even mainstream consciousness. If you would have asked me in 2000, three years after the debut, if I would still be talking about this failed feature, I would have LOLed. While I could probably fill a book with the story and the team that brought the feature, this is the story told in the context of the arc of the PC and Microsoft.

Also, a good time to note that success has many parents and failure has none. There’s no shortage of told-you-so around Clippy, until recently that is.

Note, this post is best read in a desktop browser for a complete experience.

Back to 041. Scaling the Office Infrastructure and Platform

Why not subscribe? 🙏


In early discussions, we attempted to explain how we had learned from the abysmal failure of Microsoft Bob and that we had a plan. At one point, the conversation turned from stepping through a complex task in Excel to BillG bringing us to tears in playing back what he heard. They were both tears of joy and tears of pain. It went something like this:

Demo: The Assistant will then appear and offer each step in sequence to create a chart, as the user interface does today. But it will be more friendly and approachable and have easy access to help content.
Bill: So, when I want to create a chart the clown will pop up and say, “I’m here to help” and . . .
Demo: Not clown, but assistant.
Bill: The clown pops up and then I’m like clicking on the clown saying clown next, next, clown next or something just to create a chart.
Demo: The Assistant is just a more approachable and helpful version of the same number of clicks and steps you always had.
Bill: Next . . . next . . .next, and pretty soon I just want the fucking clown to get out of the way.

Bill often had these routines or short skits that he would play out over and over—if you were the target, it was painful the first couple of times then it was for show for other attendees then you had to assert yourself. This was one of those. Through the entire rise and ultimate fall of the idea of an animated character or agent, which he referred to as a “clown”, this pattern complete with the escalating high-pitched frustrated BillG voice would make an appearance. I lost track of how many times he ridiculed the feature this way. Still, he doesn’t get the right to say told-you-so.

This started with the earliest products based on an animated helper that were developed in the early 1990s and released while I was working as Technical Assistant, so I was quite familiar with the above routine. Then Microsoft’s focus was on bringing software and PCs to children and like all products for children, the general theory of education told us that products needed to be fun, engaging, and immersive, and different from business-oriented or grownup products.

A pair of products were developed together for Windows 3, Creative Writer and Fine Artist, a kid-oriented word-processor and drawing program. While these products were nominally about the basics of productivity, they were part of an entire animated world called Imaginopolis hosted by the ever-present guide, McZee, a lanky purple humanoid. It is easy to be dismissive of the products, but in fact they featured enormous feature-depth for the time. McZee was more than just a helper, but essentially the full user interface for the products. All the actions were directed through McZee.

Measuring success in the new Microsoft Kids line was difficult because the unit sales weren’t spectacular and because everything was new and the company was determined to stick with it (remember the Microsoft reputation for taking three versions to get something right). The spiritual successor to these two products was an even greater product risk as it was not just for kids, but for the home. The problem that needed solving was that people were buying home computers, but lacked software to do home things, like keep lists, write letters, track to-do items, and calendars. While there were business packages that did that, the general theory was that software for the home needed to be more friendly and approachable, especially because those not skilled in business computers would use it.

The Media Equation by Byron Reeves and Clifford Nass
The book that defined key ideas in how humans interact with computers, tv, and new media as though they were real people and places. The authors were Byron Reeves and Clifford Nass (sadly, he passed away in 2013) of Stanford who contributed to Bob and Office, and collaborated with KarenFr and BarryL co-leaders of Bob. (Source: Personal Collection)

The Consumer Division where Microsoft Kids software came from was filled with people on a mission to bring software to a broader audience. One of those was Karen Fries (KarenFr) who was the lead advocate and pioneer for the use of what was widely known in academic circles as social interface. Karen was co-leading program management for these new products and was deeply immersed in the cutting-edge technology. She was a co-author of a 1994 paper “Seductive interfaces: satisfying a mass audience” with some of the early work in the area. Other authors on the paper were Stanford researchers, Clifford Nass and Byron Reeves. This was serious work with some depth. Nass and Reeves (who would later consult with our efforts in Office) developed the work into a book The Media Equation: How People Treat Computers, Television, and New Media Like Real People and Places. Their research developed and provided evidence of a core thesis that humans “treat computers, televisions, and new media as real people and places” and beyond that, humans develop models for interacting with technology and media based on how those works are designed. At the extreme, this explained frustration and fear of computers because of the general belief that computers are smarter than people and so interacting with them took on the traits of interacting with a much smarter and less tolerant human. This is what Karen, along with her co-leader (and designer) Barry Linnet (BarryL) set out to fix in developing Microsoft Bob, codename Utopia. No strangers to making easy to use software, Karen and Barry co-led the creation of Microsoft Publisher, a very successful and much-loved entry into what was known as desktop publishing (newsletters, certificates, menus, signs, etc.), aimed at home and small business users.

Writing a letter with Microsoft Bob
Some of the capabilities of writing a letter with Microsoft Bob. The product had significant feature depth compared to many products, all accessible through the social user interface. (Source: Personal Collection)

Like McZee, Microsoft Bob was an immersive environment. The experience, however, was less kid and more home. It was still animated, and it was still fun. Bob was the smiley face that occupied the middle letter O of the name, though within the software an ever-present puppy acted as the assistant and guide for using the many modules of the product. Each module was depicted as a place to click on in the home—click on a phone index for contacts, a pad of paper to write a letter, a checkbook for finances, a globe for a geography quiz (gosh, that was such a BillG thing), and so on. The software had even more depth than the previous products. As an example, a typical home letter-writing effort might be a complaint to an airline for lost luggage. Bob not only contained samples, but even maintained a list of airlines and addresses that it would use to pre-populate a complaint letter (and this is before the internet).

At the January 1995 Consumer Electronics Show, Bob was launched to immense fanfare and broad media coverage across print and even morning television. There was so much enthusiasm about home computers but before the internet people were just not sure what to do with them, at least broadly. That said, the product was unfortunately not well-received and ran into the buzzsaw of technologists who simply didn’t buy into the shell or veneer Bob created around Windows.

Why was Microsoft going through all this and making these risky, or even edgy, products? Many seemed puzzled by this at the time. In order to understand that today, one must recognize that using a PC in the early 1990s (and before) was not just difficult, but it was also confusing, frustrating, inscrutable, and by and large entirely inaccessible to most everyone unless you had to learn how to use one for work. In fact, using a computer usually meant signing up for an in-person class that would meet at night for a few hours over the course of several weeks—often buying a computer came not with an extended warranty upsell, but one of these classes. It was this era when businesses would put out job opportunities for people that had 1-2 years of experience using a PC, preferably Lotus 1-2-3 and WordPerfect. These products with their dizzying array of keystroke commands and chorded combinations of ALT, CTRL, and SHIFT keys were difficult if not bordering on impossible for most people to master.

Assortment of large books each more than 600 pages for using Microsoft Office 97
Using Office back then usually meant investing in after-market books such as some of these (including Microsoft’s own “At a Glance” series). These books often went to 700 or more pages and were as thick as phonebooks. (Source: Personal Collection)

If you’ve been reading along the past few months, then you know that Windows and the graphical user-interface was supposed to fix all this with its easy to use menus and direct manipulation with a mouse. Yet the exact opposite happened because while those made accessing commands easier, the number of possible commands was growing at a rapid pace. It wasn’t just that Word added bullets and numbering, but it added the myriad of options to stylize, format, and order paragraphs. And footnotes, endnotes, pagination, hanging indent, and on and on, then Excel and PowerPoint too. In order to mitigate the growing complexity of the products, Office developed an array of bolt-on utilities from massive printed and bound books, wizards (pioneered in Publisher), tutorials, getting started (like a tutorial but shorter), even a friendly tip-of-the-day that offered a quick refresher lesson when you launched a program. It got to the point where even these various forms of help needed an overview to explain them. Ironically, an after-market developed which packaged up all that information and the expertise of authors to create even more help. Typically, owners of Office (or even those considering owning the product) would invest in phonebook sized softcover books further explaining the use of the product. At first this seemed cool, then we started to realize the futility of our own product development efforts.

My college recruiting talk on developing the Assistant detailed the story of building the guru into Office in these several slides (animated). (Source: Personal Collection)

The one constant, as we studied the landscape of people using Office, was that getting anything done involved tracking down the nearby Office guru—the person that invested the time and effort to master the software more than the rest of the people in the office. Need to create a table, figure out a formula, draw an org chart then go down the hall and get help from the guru.

Chances were high that the product did what you thought you wanted to do, but the path through the maze of commands was not only difficult but fraught with the risk of destroying your work or getting the document into a state that would make further work even more difficult. We often would receive letters detailing specific features or outcomes a customer would like to achieve, only to learn that the feature was already in the product.

With Office96 we set out to build the guru into Office to solve this growing problem and dissatisfaction with the product. The early love of Office was turning into early signs of resentment as the customer-base grew. Early adopters loved the power of the product, but increasingly new customers felt overwhelmed by their lack of mastery. We had a genuine customer satisfaction problem on our hands.

As we knew from Nass and Reeves research, people had confidence in the tool to get things done but lacked a way to interact with it to find how out. If the right human guru was helping, then our challenge was to build a software equivalent.

That software equivalent would start with the clown, as BillG called it or Assistant as we called it. That is why the name of the internal implementation of the Assistant, tfc in our Hungarian notation, which stood for the f*cking clown. Even though Bill had ridiculed each social interface product, we were deep in the problem we needed to solve and optimistic we could figure out an approach. We needed to look no further than the computers on Star Trek, which enabled Captain Kirk and Spock to tap into the vast resources with vague questions and open-ended problems. Similarly, the industry was buzzing with the idea of agents that would be able to do work on your behalf such as find cheap airline flights or schedule meetings. Everywhere from Apple to the MIT Media Lab were talking about agents. There was ample evidence this was not simply a weird vision in our corner of the tech world. In fact, by some accounts we were in a race to have the first and best guru in the box.

The lesson from Bob was clearly that an entire immersive environment would not work, plus there was no way we would do that for Office. We also knew that rewiring the entire interface to do everything through the step-by-step interactions with the assistant would not work. Instead we wanted to combine the warmth and comfort of a social experience with the kind of help the guru provided in real life. While many would ultimately conclude that the paperclip was simply bolted on the side of Office to provide cuteness, we made three big technology investments, even bets, to bring Clippy to market.

Please consider subscribing

The first step in asking a Guru for help was to ask a question in your own language. The guru then maps that question to the typical answers or FAQ that they know about. You might ask “how to print sideways” and the guru knows to check the landscape option in the print dialog, or “how do I hide the elephants in Word” and the guru knows you are talking about the pilcrow symbol, ¶. A typical question might be even more abstract such as “how do I format alternating lines in a spreadsheet" which a guru might point to more sophisticated features of Excel rather than some of the direct formatting tools. This is precisely the technology we had developed and released for Office 95 as the Answer Wizard. In fact, we back-ported Answer Wizard to Office 95 because it was working pretty well and did not disturb the rest of the product. As mentioned previously, there was a collaboration with Microsoft Research (also a group from Stanford) that led this first pillar of the guru.

Next, one of the things a guru does, at least a good guru, is watch over your shoulder when you are struggling. Often diagnosing a problem, like trying to align two boxes in an org chart or get the columns of a table to be the right width, is not so much being told the right answer as much as being told which step was where you went astray. We knew many tasks in Office were composed of multiple steps that needed to be done in the right order, and often people would try something click undo and try something else. We posited that if we could track the activities as the product was used we could either proactively or on request (hitting the help key, F1) offer a suggestion from our library of help topics for how to get the right thing done. For example, if a user seemed to be clicking around paragraph formatting and indenting, the system might know enough to suggest a help topic on formatting headings or paragraph spacing. And if a user was stuck, simply hitting F1 was a way to summon the guru using the context that was accumulated over the most recent few minutes of use. This was another collaboration with Microsoft Research based on some of early 1990s work using Bayesian math to build a model for making these guesses based on contextual cues. This work came out of Stanford’s artificial intelligence lab and formed the early AI efforts in Microsoft Research. It too was all the rage at the time in academic tech circles.

It was this part of Clippy that proved to be the most challenging to deliver on the promise. Deciding when to fire off the assistant, finding that balance to being helpful versus annoying, is precisely what the human guru finds challenging when looking over your shoulder. Too little help and the product remains frustrating. Too much help and the user just wants to hand the keyboard over and say you do it. The artificial intelligence approach may or may not have been the right technology, eventually, but it proved inadequate at the time. The product had too many commands and entry points, or simply too many decisions to make at any given time to be truly helpful.

One mistake, well really the mistake, was firing Assistant on the most simple and obvious effort in Word. The sequence of starting a new document, typing Dear <name> and pressing return would cause the assistant to say, “Looks like you’re trying to write a letter.” And with that Clippy was forever sentenced to memedom (is that a word?)  In our heads we thought this was ok because we were already doing a tip (a yellow bar across the top of the screen) to alert the user to the letter writing feature. This was really a step too far. It did not help that if you clicked “Yes” the software would launch an incredibly complicated wizard offering all sorts of options for a letter, most of which went unused. A few years later, the Microsoft Researcher who contributed the Bayesian technology even turned on us in an interview with The Economist and said it was all because we didn’t use enough of the technology. That really hurt—it was as much their work as our work.

"Son of Paperclip" article in the Economist. It contains the paragraph "To be fair, that is not Dr Horvitz's fault." and goes on to explain how the product team did not really listen to him.
This is from The Economist, March 24, 2001 (more on this time in a future post). In this article you can see how at this point in the evolution of the feature, failure was indeed an orphan.

The third pillar of bringing the guru to Office was to offer the user the calming and comforting personality of a guru. Using a computer was difficult and frustrating, and we set out to bring some levity to the daily grind. Leaning heavily on the work of Nass and Reeves, we developed the actual character to represent the guru—to attach a personality to the source for answers and tips that would encourage help. We also went a step beyond that and decided that the Office Assistant would be where messages (or alerts) would come from. The ever-present “Do you want to save this file” or “The spell check is complete” would emanate from the assistant. This was the biggest and highest risk bet of the entire feature. It is also what separated the feature from the previous dozen attempts at providing help—it wasn’t yet another bolted-on tool, but it was in the flow of usage and there to help everyone. Internally we called this IntelliAssist.

The minute we had an animated Assistant it was obvious that any opinion or controversy about the feature would stem from the clown or character. Starting in early 1994 we began the most intensive usability research testing we had done to date on a feature. The number of tests, the number of locations and languages, and design ideas we iterated on was kind of mind-blowing. At one point people were flying to Japan and Europe to rerun tests to see how the results might differ. How big should the assistant be, how much noise should it make (if the user even had a sound card), how often should it appear, how animated should it be, and on and on. The iterations were seemingly endless, all with the goal of making it friendly and approachable while tapping into the fancy underlying artificial intelligence technology.

Choosing the actual character was incredibly controversial. It became immediately apparent everyone had an opinion, and importantly every major geography had their own view of what would work locally. It didn’t matter that many animated characters worked globally, there was a strong demand for input and oversight. The risk, after all, was very high. For example, Japan accounted for nearly one-third of Office profits because of the unique market there. The lead program manager, Sam Hobson (SamH), an experienced member of the Excel team who joined OPU (also a college hire like the rest of us) had the perfect demeanor for managing all the connections across the company.

Perhaps we were naive, but we never sat around contemplating the risk the business of doing this feature. In 1995, Platforms revenue was $2.36 billion, and Applications revenue was $3.58 billion—even a small hiccup in Office would be a huge deal. We weren’t comforted in the past sales of the product, but rather sought the comfort of believing we were on a mission to solve an acute customer problem—a problem left unchecked that could materially impact the business. How could a product remain successful if people increasingly dislike it?

Sam created huge boards of potential characters for everyone to look at and pick their favorite. He would lead tests at shopping malls and markets around the world understanding preferences. Meanwhile, Nass and Reeves reminded us these preferences were rather predictable and also not as crucial as maybe the sales people who saw this more as branding than utility believed. In one hilarious early use of these boards, Sam invited the spiritual leader (and then Microsoft board member) Mike Maples to pick his favorite character. Ever the rancher and Oklahoman, everyone thought Mike would pick the big dog or maybe the lion or something. Instead after browsing the dozens of choices, Mike went with. . . the pink bunny rabbit. He smiled and said it reminded him of the rabbits on the ranch. This kind of reaction is what led to the full gallery of choices. While the paper clip, Clippit aka Clippy, would be the default, we featured a dog, a cat, a happy smiling dot reminiscent of Bob, and several more including a really boring Office logo. The scale of Japan’s business required us to take their input and from that we ended up with the highly controversial Office Lady or Saeko Sensei, which to many at HQ was less than appropriate. Japan also came to love symbols of nature, and guided us to a Kairu, a dolphin and again there was irony in that choice that made us uncomfortable. We kept those characters to the Japanese version of the product. We would later add a small Macintosh-like computer called Max for Mac Office. Being Microsoft, we had an SDK and even a third-party partner that could (and would) create additional assistants.

The character, like the artificial intelligence behind the first two pillars, had a depth of capabilities that often go unappreciated and certainly did at the time. We were severely constrained in disk space and memory, not to mention graphics capabilities, yet wanted to provide a reasonable animation experience. This proved extremely difficult as the expectations for animation had been set by cartoons.

Join the discussion 🗣️

At one point we had a most memorable opportunity to meet with the legendary animators from Walt Disney, Frank Thomas and Ollie Johnston otherwise known as Frank and Ollie. Together they were involved with everything from Pinocchio to Fantasia to Bambi and more. An example of a constraint that frustrated us was the window the character was trapped in. We wanted to do a borderless Assistant like in Bob, but the platform constraints were too much when overlayed with regular Windows apps. Frank and Ollie not only relieved us of that but explained how we should use the window as their stage to allow for entrance and exit and directional animations. They also pushed us to add a sidekick (think Thumper) which was something they had pioneered in animation. They suggested Clippit have something like a little eraser friend. That was well beyond the two dozen or so animation sequences we could have but really brought us optimism for how the feature could evolve with more platform support.

Sound was still nascent in most PCs, constrained by the original MIDI sound capabilities. Windows 95 and multimedia was changing that. We also added a set of sounds that came along with animations which if a user had them on made a real difference in the experience.

These capabilities were coded throughout the product. The Assistant would occasionally just blink or smile or take note of work. If you stopped typing for a while it might perk up and notice you. Using a technical feature would come with a more substantial animation. The assistant was also programmed to get out of the way while you were typing or scrolling, which led to a fun game of chase-the-paperclip using mouse and the Excel grid as was commonly shown in demonstrations.

As we tested the character in various stages behind one-way glass or in focus groups, there was almost always surprise and more frequently than most would believe today praise and support for the feature. It is a cliché for a failed feature to say that it worked in early testing, but that was genuinely the case.

The assistant came with many options that could be enabled/disabled. This shows the full set including: respond to F1, help with Wizards, display alerts, search for both product and programming help, a variety of settings to show different tips, and then customize the frequency of showing tips.
As the product evolved, we expanded the customizability of the Assistant behavior. Our belief was that we should have limited the number of options, but the culture and spirit of Microsoft at the time tended to win out over that more narrowly held belief, resulting in a plethora of settings. (Source: Office 97, Personal Collection)

Still as the project progressed there were many that were nervous or outright hostile. As we showed the product to the hardcore technical audience, the reactions were often visceral and immediate. Either people wanted to immediately turn it off without much consideration or they would be thoughtful and suggest that it is not for them, but they could see others (read as less technical) people benefitting. As we would learn time and gain, when core technical users say that something isn’t for them but for others it too often means that the feature might be good, but it is going to need to get past these gatekeepers. We made a very difficult decision to provide an array of settings to control the various capabilities of the Assistant. In other words, we made it possible to turn it off. At the same time, we provided full programmability with Visual Basic for Applications (VBA) so that developers could create custom solutions with full control over the Assistant, including adding custom text in the balloons and choosing animations. Imagine how much fun that budget template in Excel could be with custom chatter from the Assistant!

The Assistant was one part of an enormous release of Office. The remainder of this chapter details some of the other challenges in building Office 97 on the platform and infrastructure described in the previous section.  The product reviews were ultimately mixed, but hardly universal, as we will see in the end of this chapter.

As we will see in future chapters, we stuck with and improved the Assistant in the next release of Office. By the second subsequent release we retired the feature, albeit in a humorous way. In parallel with Office 97 an effort began to bring the Assistant to Windows for use by third party developers. Microsoft Agent had much richer interactions using early speech recognition and voice but lacked deeper integration with applications unless coded by developers. Agent was used in Windows XP and remained available for some years.

The journey of Clippy (in spite of our best efforts that is what the feature came to be called) is one that parallels the PC for me in so many ways. It is not simply a failed feature, or that back-handed compliment of a feature that was simply too early like so many Microsoft features but represents that final attempt at trying to fix the desktop metaphor for typical or normal people so they could use a computer. What everyone came to realize was that the PC was a generational change and that for those growing up with a PC, it was just another arbitrary and random device in life that one just used. As we would learn, kids didn’t need different software. They just needed access to a PC. Once they had a PC they would make cooler, faster, and more fun documents with Office than we were. It was kids that loved WordArt and the new graphics in Word and PowerPoint, and they used them easily and more frequently than Boomers or Gen X trying to map typewriters to what a computer could do. It was not the complexity that was slowing people down, but the real concern that the wrong thing could undo hours of work. Kids did not have that fear (yet). We needed to worry less about dumbing the software down and more about how more complex things could get done in a way that had far less risk.

The other lesson from the Clippy experience is clearly how amazing it was that Microsoft even considered such a high-risk feature. Imagine doing a feature that you know at launch will have some people significantly annoyed with you but doing so also knowing that you could reach some other new customers or bring joy to customers that were otherwise worried. The whole business relies on upgrading existing customers and attracting new customers when all of them have an option of doing nothing or going to one of several competitors. The Microsoft that made Clippy is the risk-taking company that I admired so much. It was the failure of Clippy and the lack of repercussions that in a sense that cemented my own connection to the company. I got way more grief outside the company than inside.

And I needed that because for the next five years of college recruiting trips, I would have to answer snarky questions about Clippy from college students. The deepest pit in my stomach came when I was in New York on a trip at a low point in the Microsoft versus DOJ trial. I turned on the hotel television for some Late Night with Conan O'Brien and his opening monologue took a swipe at Microsoft, “Come on Bill, Microsoft got off easy compared to what the Government did to Clippy, that annoying paperclip icon that pops up in Microsoft Word” [emphasis added] followed by a gruesome violent act perpetrated against poor Clippy. That hurt. A lot. The cheers from the studio audience hurt even more. I was totally signed up for the risk and reviews but being mocked on my favorite late-night show. Ouch.

Then one campus season, perhaps in 2002 or 2003, those snarky comments turned into an expressed love of Clippy and comments like “I remember Clippy on my mom’s computer at work” or “I miss the Dog”. That was amazing. Only that was outdone when about a decade later, Clippy transitioned from nostalgia to a high-tech feature that was somehow ahead of its time. I wish I could say that was the case, but it was simply an idea. The implementation was decidedly 1997.


Enjoyed this? Please subscribe 🙏

On to 043. Dim Outlook

Join the discussion. 🗣️