An op/ed in TechCrunch caught our eye over Memorial Day weekend. We were drawn in by the author’s perspective on broad versus narrow chatbots. At Next IT, we’ve been preaching the benefits of domain-specific language models for 15 years, and it is gratifying to see the concept pick up steam with others.
There was also a passage that didn’t sit right with us -- one that we think mischaracterizes the enterprise IVA’s and chatbots, such as those used by a travel company. The author argued:
“For example, if someone built a chatbot to help you book flights, you can imagine about 100-200 things a human could possibly say to the chatbot. So, it’s quite easy to build a brain for this need. I can do it in a few weeks and be pretty sure that I’ll get 99 percent accuracy in responding to the human.”
We have, in fact, built IVA’s that help consumers books flights, so we’re speaking from experience when we say: it’s just not that easy.
Complexity is a necessary condition of the modern enterprise. Our typical customer has a lot of innovation happening in their IT environments, but there’s also a lot of legacy software, there are integration gaps and there’s almost always a resource shortage. Toss in a little hardware maintenance, data management, governance and security risks, and even the best CIO’s get a little dizzy.
Also consider the complexity in serving thousands of customers simultaneously in an enterprise where each customer has a unique context and journey. For example, a repeat customer may want to book a trip using rewards points, is flexible about arrival and departure dates so long as they get the whole trip on points, but only wants to book a window seat. Could a chatbot trained on just 200 queries and access to just one API get the job done? Would the customer enjoy the experience? Would it scale to serve 10,000 customers?
Most of our deployments involve a good dose of integration with existing enterprise systems. These integrations make the difference between a functioning chatbot and an intelligent assistant. An IVA that can tap into data in CRM systems is able to piece together the customer's context before they’ve finished asking their first question.
We became proficient at connecting into those systems over the past 15 years because we learned that the added integration complexity resulted in better ROI for our customers. In other words: the perceived complexity had upside. Furthermore, we developed tools that allow the business to train and refine their IVA’s over time, ensuring the technology changes as the business changes. It would be great if you could “set it and forget it,” but the enterprise isn’t that simple. 99% accuracy today does not guarantee 99% accuracy tomorrow, so we plan for constant change.
Complexity is ubiquitous
The author later goes on to say:
“For something like travel bookings via a chatbot, the body of knowledge you need is not immense. I imagine you could literally plug your chatbot into a TripAdvisor API and deliver a pretty robust product… But for anything in financial services, which is the world in which I operate, the barriers and the time it takes to compile this body of knowledge is immense.”
Speaking specifically to the travel industry, we’re amazed each day by how many concurrent and complex travel processes IT departments support. Based on our research, just the most common customer support processes in travel represent a massive body of knowledge covering:
- New reservations and booking
- Existing reservations (check and change)
- Loyalty and rewards
- Day-of service logistics
- Deals packages and discounts
- In-travel services
- Log-in issues
- Seating and accommodation options
- Infant, pet and special needs requests
While those process are unique to travel, it does not mean travel is any more or less complex than other industries. It simply means a domain-specific solution is going to get better, more sustainable results for a given travel company.
Much like the author’s company, Next IT has also delivered solutions to financial services customers. We see complexity in both industries, every day.
When you’re solving real-world business problems with this technology, it’s never just a matter of building a brain and plugging into an API. It’s a matter of integration with systems of record, continuous improvement of language models, designing and deploying to deliver measurable ROI to the business, and so much more that you can read about on this blog.
If you’re considering deploying a chatbot or IVA in your business, make sure to rigorously document the intricacies of your business challenges and goals. Once you’ve done that, you can look for technologies that are capable of doing the whole job. If you have a truly complex business challenge, it’s unlikely that a simple, cobbled-together chatbot is going to solve your problem.
Or as conventional wisdom tells us: what’s easy isn’t always what’s right.