If you've worked for a large organization, this should sound familiar: You're trying to follow the “proper channels” for procurement. What should only take a few minutes turns into a whirlwind of red tape the size of a kaiju - a Japanese movie monster.
Corporations aren't the only ones dealing with this red-tape monster. Until we confronted it, the City of San Francisco was tormented by it, too. Its procurement processes are complex, leading to a flood of questions that swept away the time of city workers. To throw them a lifeline, Yeti collaborated with them to design an online chatbot.
Our creation, PAIGE (Procurement Answers and Information Guided Experience), is a "limited memory" chatbot. By leveraging machine learning — a small and controlled subset of artificial intelligence - it takes into account the context of its current conversation. Rest assured, it's by no means about to eliminate the human race.
Still, even with its limited memory, PAIGE can answer about 80 percent of the questions asked about procurement procedures. Staffers who once spent hours fighting the monster now have time for the trickier, human-required aspects of their jobs.
Designing PAIGE wasn't exactly a snap. But at Yeti, we pride ourselves on finding creative solutions to complex problems, and we've alreadyexperimented with chatbots in the government space. Plus, procurement is a complex but ultimately rules-based process, meaning it's navigable by a robot that can translate these rules into human speech.
What We Learned Designing PAIGE
Now that PAIGE 1.0 is streamlining city procurement, we've tucked some valuable lessons under our belt:
- Talk to as many target users as possible.
Our first task was to better understand the procurement process. That meant talking to the city workers doing it every day.
City procurement, we discovered, is hair-raisingly complex. The more people we talked to, the more we realized we didn't know. By the time we started building, we'd spoken to staffers from almost every city department. It wasn't just fact-finding, either; we empathized with users, building mental models to better understand the context they were working in.
No matter what you're building, chatbot or otherwise, spend time talking to users. Really get to know them and their needs. Otherwise, you have no hope of solving their problem.
- Focus on workflows.
PAIGE shines in its role not just because of the AI technology behind it, but also because of the care we put into its workflows. Those workflows are what PAIGE uses to recognize "buckets" of questions and supply relevant answers.
For example, one workflow centers around commodity purchasing. Using a decision tree, PAIGE walks the user through how to purchase a commodity. If the user tells PAIGE that he wants to buy a commodity under $10,000, for example, PAIGE directs him to the Tech Marketplace and provides a handy navigational guide.
Of course, we weren't new to the idea of workflows. But this project underscored to us just how important robust workflows are to the value and functionality of a technology product.
- Look to the future.
Before we finished our first iteration, we were already thinking about ways we could add value to our blossoming civic chatbots platform. Some of the features we hope to add in the future include:
- Training portals. City interns spent more than 100 hours training the chatbot. We realized that if we could build a training portal, it could immensely reduce training time and provide additional value to users.
- Other departmental use cases. Procurement is just one way cities can leverage chatbots to engage citizens and cut red tape. We're already brainstorming other use cases with the city, including tax assistance, 311-call triage, zoning and building regulation, public utility inquiries, and more.
- Personas. Chatbots shouldn't sound like, well, robots. A few of the personalities we've considered for PAIGE include the "friendly sidekick" and "grumpy librarian". Personality adds flair to chatbots, which can improve adoption and acceptance. What’s fun gets used!
- Other languages. San Francisco is a diverse city with large Spanish- and Russian-speaking populations. To accommodate non-English speakers, we're thinking about using a tool like Google's Translate API to help everyone comfortably interact with PAIGE.
- Speech-to-text processing. Users input text into PAIGE, which then "speaks" an answer. But this can work the other way, too. Integrating Siri, Alexa, or Cortana would let users truly talk with PAIGE, creating a futuristic, conversational experience.
The City of San Francisco's PAIGE system is just a glimpse of what's to come. AI-based tools will become regular features of governments. They'll conduct citizen outreach, reduce staffing needs, and increase efficiency. With chatbots, the red-tape monster doesn't stand a chance.