Virtual Men: IM-bot for Keebler (part 2)
- Transfer
( beginning of article )
Ingredients
Murray went headlong into the project. “I'm the crazy scientist behind RecipeBuddie,” she says. “This is far from the same as a database wrapped in IM packaging. It's all about natural language. This is the territory of the writer, and writing scripts is much more fun than you can imagine. You must anticipate all the possible ways that people can ask you a question. For example, the expression "I hate onions." Someone might write, “I don't like onions,” or “I can't stand onions,” or “Just no onions.”
Then, you must master the different options for communication with the user. If someone says that he doesn’t like onions, you can answer “I understand. So no bow. Why don't you try one of these recipes? ”
This answer consists of three separate parts: confirmation, repetition of customer needs and recommendations. And each of the parts can be expressed in various ways. For example, an agent must recognize that “Hello” is the same as “Good day”, and “Hello”, and “Hayushki” and “Preved”.
Writing an agent using BuddyScript is like writing a play for two characters, an interactive agent and a user. The developer determines what words and phrases the agent will be able to understand and how he should respond.
“We make it so that RecipeBuddie can respond to things like“ I'm sad, ”she said in response,“ How would you feel about food that creates comfort? ” Or “I need to make dinner for the children,” and then she offers something suitable for the children. The database of recipes is not capable of this, since apple pie alone does not mean that children can like it. On top of the recipe database, we impose a layer that is a layer of humanity. ” This layer works with recipes based on associations with words that are not among the ingredients - for example, “kosher” or “picnic”.
To program RecipeBuddie, Emedia took 8 weeks of work on a tight schedule. About two thousand dialogue models were created. “The ideal combination for projects of this kind is a strong programmer and a strong writer. All you have is words. And your task is not just words, but a voice full of strength, ”says Murray. Her team was supposed to create a special voice for RecipeBuddie. They started with the hazy image of Emeril Lagasse, then switched to the prissier type Betty Crocker and only after that settled on Becky, who, as they thought, best fits the audience of Keebler - a woman, between 25 and 54. “She's provincial and she has a feeling humor, ”says Murray.
However, women of this age category are not the only users of RecipeBuddie. Apparently, some other groups cannot help asking the woman-bot a series of questions, from outrageously personal to completely offensive. “I deliberately tried to ask all the possible questions that I could think of,” says Murray, who also asked to test the bot and her friends. This prepared Becky for such inevitable questions as “Would you agree to meet with me?”, “How old are you?” And other more personal things, in response to which she should politely let the user in and translate the conversation on a culinary topic.
To taste
After technical difficulties were overcome, Murray faced political obstacles, starting with AOL, whose IM service is most widely used on the Internet. Before an external bot is launched on its network, AOL must give permission because “they must click the appropriate button that allows an unlimited flow of messages to be sent to one nickname,” Murray said. The catch is that AOL makes no money from ActiveBuddy bots. That is why, AOL requires that when launching bots, such as RecipeBuddie, they buy a media platform from her. Neither MSN nor Yahoo requires such purchases.
"AOL tried to develop its own software for creating bots," says Murray. “It's clear that they just want to get all the money on their own - when these bots were launched, RecipeBuddie had to pay.” Be that as it may, the three main IM services allow RecipeBuddie to work on their networks.
“We use different methods to measure the success of RecipeBuddie,” said Jeff Johansen, vice president of marketing at Keebler. “They include the number of users who added RecipeBuddie to their friends, the number of messages that RecipeBuddie exchanges, the number of recipes viewed, and the number of recipes printed. The number of users who have added RecipeBuddie to their friends reflects the intention to use RecipeBuddie more than once. In addition, we use the factor of printed recipes to determine intentions to make future purchases and expected offline sales. "We were shocked when it turned out that our plans for the quarter were exceeded, just a few weeks after the launch of RecipeBuddie."
RecipeBuddie was launched in September 2002. Keebler says they haven’t gotten a mind blowing break yet, but during the test that made Keebler get the idea of RecipeBuddie, out of 120,000 clicks, 6.5 percent clicked on the link to the online registration form for the web contest Cheez-It website.
ActiveBuddy also had high hopes for previous agents. When the Austin Powers agent was launched on June 26, 2002, it brought nearly 50 million messages in more than 2 million visits, in just one month. On average, consumers interacted with the Austin Powers bot for 8 minutes each, and this is very tempting news for marketers. The number of clicks on the movie website was 74 percent, the number of clicks on Austin Powers Doritos links was over 7 percent and more than 4 thousand users registered on the Doritos site for the first 4 days of the campaign.
“Judging by the number of clicks on the links, we can say that the program was successful,” Murray said, and added that during the process of creating RecipeBuddie I learned three important things. “Firstly, we just touch the surface of what we still have to do. Users have a desire to really deeply communicate with bots. Secondly, a very close-knit team should do this. This is a bit like writing a novel. Thirdly, there are many ways to allow people to contribute their own content. Right now we are working on a project that will allow us to compactly link frequently asked questions and answers to them. ”
As for RecipeBuddie, her Bombay sandwich turned out to be exactly what I was looking for and it was time to end our chat. “Thanks for everything,” I told her.
“You're welcome,” she replied. “May I find some other recipe for you today?” But before I could answer, she blurted out: “Bring some flavor to your life with the stunning taste of Toasted Crackers!”
Translated from English for:
blog.worldwebstudio.com
Ingredients
Murray went headlong into the project. “I'm the crazy scientist behind RecipeBuddie,” she says. “This is far from the same as a database wrapped in IM packaging. It's all about natural language. This is the territory of the writer, and writing scripts is much more fun than you can imagine. You must anticipate all the possible ways that people can ask you a question. For example, the expression "I hate onions." Someone might write, “I don't like onions,” or “I can't stand onions,” or “Just no onions.”
Then, you must master the different options for communication with the user. If someone says that he doesn’t like onions, you can answer “I understand. So no bow. Why don't you try one of these recipes? ”
This answer consists of three separate parts: confirmation, repetition of customer needs and recommendations. And each of the parts can be expressed in various ways. For example, an agent must recognize that “Hello” is the same as “Good day”, and “Hello”, and “Hayushki” and “Preved”.
Writing an agent using BuddyScript is like writing a play for two characters, an interactive agent and a user. The developer determines what words and phrases the agent will be able to understand and how he should respond.
“We make it so that RecipeBuddie can respond to things like“ I'm sad, ”she said in response,“ How would you feel about food that creates comfort? ” Or “I need to make dinner for the children,” and then she offers something suitable for the children. The database of recipes is not capable of this, since apple pie alone does not mean that children can like it. On top of the recipe database, we impose a layer that is a layer of humanity. ” This layer works with recipes based on associations with words that are not among the ingredients - for example, “kosher” or “picnic”.
To program RecipeBuddie, Emedia took 8 weeks of work on a tight schedule. About two thousand dialogue models were created. “The ideal combination for projects of this kind is a strong programmer and a strong writer. All you have is words. And your task is not just words, but a voice full of strength, ”says Murray. Her team was supposed to create a special voice for RecipeBuddie. They started with the hazy image of Emeril Lagasse, then switched to the prissier type Betty Crocker and only after that settled on Becky, who, as they thought, best fits the audience of Keebler - a woman, between 25 and 54. “She's provincial and she has a feeling humor, ”says Murray.
However, women of this age category are not the only users of RecipeBuddie. Apparently, some other groups cannot help asking the woman-bot a series of questions, from outrageously personal to completely offensive. “I deliberately tried to ask all the possible questions that I could think of,” says Murray, who also asked to test the bot and her friends. This prepared Becky for such inevitable questions as “Would you agree to meet with me?”, “How old are you?” And other more personal things, in response to which she should politely let the user in and translate the conversation on a culinary topic.
To taste
After technical difficulties were overcome, Murray faced political obstacles, starting with AOL, whose IM service is most widely used on the Internet. Before an external bot is launched on its network, AOL must give permission because “they must click the appropriate button that allows an unlimited flow of messages to be sent to one nickname,” Murray said. The catch is that AOL makes no money from ActiveBuddy bots. That is why, AOL requires that when launching bots, such as RecipeBuddie, they buy a media platform from her. Neither MSN nor Yahoo requires such purchases.
"AOL tried to develop its own software for creating bots," says Murray. “It's clear that they just want to get all the money on their own - when these bots were launched, RecipeBuddie had to pay.” Be that as it may, the three main IM services allow RecipeBuddie to work on their networks.
“We use different methods to measure the success of RecipeBuddie,” said Jeff Johansen, vice president of marketing at Keebler. “They include the number of users who added RecipeBuddie to their friends, the number of messages that RecipeBuddie exchanges, the number of recipes viewed, and the number of recipes printed. The number of users who have added RecipeBuddie to their friends reflects the intention to use RecipeBuddie more than once. In addition, we use the factor of printed recipes to determine intentions to make future purchases and expected offline sales. "We were shocked when it turned out that our plans for the quarter were exceeded, just a few weeks after the launch of RecipeBuddie."
RecipeBuddie was launched in September 2002. Keebler says they haven’t gotten a mind blowing break yet, but during the test that made Keebler get the idea of RecipeBuddie, out of 120,000 clicks, 6.5 percent clicked on the link to the online registration form for the web contest Cheez-It website.
ActiveBuddy also had high hopes for previous agents. When the Austin Powers agent was launched on June 26, 2002, it brought nearly 50 million messages in more than 2 million visits, in just one month. On average, consumers interacted with the Austin Powers bot for 8 minutes each, and this is very tempting news for marketers. The number of clicks on the movie website was 74 percent, the number of clicks on Austin Powers Doritos links was over 7 percent and more than 4 thousand users registered on the Doritos site for the first 4 days of the campaign.
“Judging by the number of clicks on the links, we can say that the program was successful,” Murray said, and added that during the process of creating RecipeBuddie I learned three important things. “Firstly, we just touch the surface of what we still have to do. Users have a desire to really deeply communicate with bots. Secondly, a very close-knit team should do this. This is a bit like writing a novel. Thirdly, there are many ways to allow people to contribute their own content. Right now we are working on a project that will allow us to compactly link frequently asked questions and answers to them. ”
As for RecipeBuddie, her Bombay sandwich turned out to be exactly what I was looking for and it was time to end our chat. “Thanks for everything,” I told her.
“You're welcome,” she replied. “May I find some other recipe for you today?” But before I could answer, she blurted out: “Bring some flavor to your life with the stunning taste of Toasted Crackers!”
Translated from English for:
blog.worldwebstudio.com
