Let Them Eat Clean: Chatbots and Quality Data
Help your chatbot grow with a healthy data diet.
- By Renat Zubairov
- October 15, 2019
Whatever your business, the chances are you are at some stage of entering the Age of the Chatbot. Whether your enterprise has just started to explore artificial intelligence (AI) or is fully deploying an AI strategy -- or somewhere in between -- getting the most out of your chatbot means thinking carefully about the data it will be fed.
Gartner predicts that by 2020, organizations will be capturing 85 percent of their customer contact using AI. Furthermore, we at elastic.io also anticipate a trend towards chatbot technology being developed as bespoke solutions tailored to specific workplaces and integrating multiple bots across business functions.
We are already seeing AI incorporated from customer service, sales, and marketing through to logistics and inventory management, as well as branching out into such areas as cybersecurity and financial management. The more that related areas of a business can seamlessly interact, the more powerful the chatbot function can be for customers, users, and management alike.
Often the challenges enterprises face deploying effective chatbot solutions follow common themes, with many boiling down to one key ingredient: data. The following will equip you to avoid (or address) common pitfalls faced in deploying chatbot technologies.
Chatbots Require Context -- Lots of It
Companies want a chatbot to handle as many questions and scenarios as quickly and accurately as possible from Day One. However, AI is a learning technology, not a plug-and-play solution, so the better the chatbot interoperates with other applications, the more prepared it can be when it meets a customer -- and the more effective and intelligent will be its response.
Humans don't interact within set parameters. We tend to have several words with similar meanings and pepper sentences with colloquialisms and expressions. Every query is different, even when the answer may be the same, so breadth of context is critical.
Providing context from multiple sources across the enterprise requires organizations to link AI to data sources that may be hosted in the cloud or on premises (or a mix of both) and provide a broad foundation on which to predict, extrapolate, and learn. Working as an integrated part of the overall system structure, instead of in isolation, provides the bot with access to contextual data that creates stronger algorithms and establishes known pathways it can follow to resolve queries.
Teaching Begins in the Testbed
The expectation of chatbots being ready, willing, and able from the day they go live puts enormous pressure on the testing phase. Integrating with several data sources during learning can be costly -- though critical -- and tends to be a commonly cut corner. This may be by using a third-party data set or by limiting integration with other enterprise applications until a proof-of-concept is complete.
Ideally, testing should use the same live data sources on which the chatbot is expected to eventually run, but customers don't want to be exposed to the learning phase and organizations don't want to risk their business information being compromised in any way. Options such as an integration-platform-as-a-service (iPaaS) allow companies to integrate enterprise sources in real time within a separate but secure environment, enabling historic and current data to feed the chatbot without risk to the business or customer.
If companies choose to use third-party data sources, the iPaaS can provide integration across related fields in as many sources as required, although this approach should be treated with caution.
Identifying and Eliminating Errors Across Multiple Sources
In 2017, Facebook chatbots hit a 70 percent failure rate. Similarly, industry estimates suggest that one in seven conversations with a chatbot eventually gets routed to a person and one in eight is abandoned altogether.
One of the problems behind this lack of quality interaction is the data fed to the chatbot when it references multiple data sources, where errors can easily throw it off course. Enterprise IT structures are notorious for carrying discrepancies between similar fields in different software. The problem increases with software-as-a-service -- typically a departmental purchase rather than part of a cohesive, enterprise-wide IT strategy. If the online sales interface receives an update, such as a new delivery address or order amendment, is it set up to communicate with corresponding fields in the logistics software or inventory platform to feed the right information to a chatbot?
In most instances, probably not. Integration across the full enterprise architecture helps organizations quickly identify and eliminate errors that can easily occur within multiple data silos, thereby enhancing the quality and scope of data on which the chatbot functions.
Setting Up Your Chatbot for Success
Integrating multiple data sources that are not necessarily compatible can create a major headache for organizations, but an iPaaS can help.
Whether the solution is on premises, in the cloud, or outsourced, connecting data sets within the enterprise enables the chatbot to create patterns and predictions based on real and current intelligence. From this foundation the chatbot has plenty of good, clean food on which to grow.
They are what they eat, so when inviting chatbots to join your organization, let them eat clean!
About the Author
Renat Zubairov is CEO and co-founder of elastic.io, a Germany-based born-in-the-cloud innovator and expert in integration solutions that supports organizations of all sizes in their digital strategy by helping them spend less time integrating and monitoring data sources across the business and more time using this data to improve business operations. You can contact the author via email.