Building a chatbot may be cheaper and easier than you think – but it needs careful planning to get it right.
Last month, Jisc’s London office was invaded by chatbots during a hackathon attended by developers from universities and colleges.
At the event on 24 September, three of the leading vendors of chatbot platforms - Amazon, Google and IBM - demonstrated in detail how their technology could tackle typical challenges facing colleges and universities.
It is part of a project Jisc is running with UCISA to help further and higher institutions to get started with chatbot technology. Using anonymised data from the Jisc service desk as a basis for their workshops, each platform demoed their technology and how it can be used to help create a virtual assistant to front conversations with website users.
What we learned at the hackathon
The next step is for developers to start the task of creating a chatbot to work in their own institution. They will no-doubt be reflecting on some of the take-away messages from the hackathon, which provide valuable lessons for anyone considering this type of technology.
1. It really is easy to get up and running
All of the platforms shown were able to ingest question and answer documents and conduct simple support calls using the information they contained.
2. Automating conversations with people exposes the gaps and inconsistencies in your information
Staff providing first line support can “paper over cracks” using their own knowledge and common sense. An automated agent probably won’t, so there may be work to do to complete the knowledge base for the chatbot.
3. Time needs to be invested in planning and architecting the solution to get it right
Once you get beyond the simple FAQs, you need to clear where information is coming from and how it fits together.
4. Training the system can be where the real investment is … but it pays off
The more time your staff, and sometimes willing volunteers (“guinea pigs”) spend with the system, refining the language it understands and uses and how it conducts conversations, the better the service will be.
5. People have no tolerance for a broken robot
Where we are prepared to forgive another person for a mistake or a pause, we are not with an automated system - our patience and trust disappears in an instant…
6. So… think carefully about where to start with chatbots
Choose a good service area to begin, where you have a sound source of information. Don’t be over-ambitious: if the chatbot can only reliably handle one in three calls, then that has to be the basis for the service, otherwise look elsewhere.
A detailed report of this phase of the project will be published in the coming weeks.
Come and meet a chatbot
The next step is for a more detailed look at each of the main vendor platforms in one-day events hosted by an institution that has implemented services using each of these platforms, which can share the lessons they have learned.
23 October at Lancaster University - focusing on the use of Amazon chabot technology, which started with a chatbot aimed at staff dealing with questions around their development and training procedures, and now focusing on development of bots to support parts of their student experience.
9 November at Bolton College, where a small team have developed a comprehensive set of services using IBM’s Watson toolset at the core, Amazon’s Echo devices, and Apple’s iOS mobile platform to provide voice interaction. The team is now looking at automated assessment and support for online tutorials.
To register for these free events, visit the events pages.