Inbenta is Changing How the Enterprise Handles Customer Service

Microsoft came up with a brilliant idea.

They created a smart chatbot called ‘Tay’ that would interact with the Twitter community with zero human intervention. Not only would it interact with the community, but it would learn from it as well.

As Tay learned more, it would speak like the community did to blend in. It was Microsoft’s attempt to show their progress in artificial intelligence.

See any potential problems with this situation?

Well, let’s just say Tay got along just fine with the Twitter community – and also learned how to become a racist within hours. Microsoft pulled the plug on the chatbot and published a swift apology.

Since then, the global push for artificial intelligence that works and is able to grow and learn with you has become paramount to botmakers.

More Advanced Chatbots

Last week I sat down with the Chief Marketing Officer for Inbenta, John Forrester, and we talked about what makes a chatbot successful (as well as how they handle creating intelligent bots).

Inbenta is recognized in the market as one of the top 10 chatbots for Enterprise Customer Service according to Forrester Research. There’s a lot that goes into programming chatbots to be, well...chatty.

Here are 5 things that chatbots need to to do to succeed:


At Inbenta, the developers have worked hard to make the translation between what the user says and what the chatbot interprets as natural as possible.

The end goal for every user is that communication will be smooth, deliberate, and that the bot will pick up on the nuances of what is being said. Inbenta has developed their own lexicons in over 25 languages, all supported by their global offices and computational linguists.

This isn’t a digital game of Go Fish because there’s no guessing involved. The developers wanted to make it simple for users to interact with bots, but in order to make that plausible, there’s a lot of forethought that must go into the programming stages.

“One of the biggest things that our chatbots do is look for slang. People don’t always use correct grammar, they’re not actually typing out full sentences. They are using a lot more abbreviations and jargon terms,” says Forrester.


So why do most chatbots fail?

The answer lies in this question. No, literally – chatbots can’t determine differences between questions and input – which explains why Siri gets confused so easily.

Inbenta developers have caught on and now incorporate verifying questions into the programming.

For example, if you enter “Tuesday,” the chatbot has to understand that it isn’t a concrete date. Instead, it refers to the nearest open Tuesday on the calendar. This is a tough distinction for something without a frontal lobe.

“Decision trees are the ability to construct a tree branching of conversations, which may start out with chit-chat and end up as a transaction later on in the day,” says Forrester.

“We not only use classic yes/no decision trees, we can also do linguistic branches in our decision trees. We can ask intuitive follow up questions, and the chatbot will be able to interpret the user’s answers, which makes for a better customer experience.”


Microsoft embraced the “just ship it” motto, which is needed for their brand as a technology provider. But when it comes to customer service, mistakes like Tay can be catastrophic for a company.

“These are real conversations with real people who need help with their situation. We designed our chatbots to not only give the right answer, but to not steer off in a wild direction in case the conversation escalates,” says Forrester.

I used to work for both Accenture and Deloitte, and I know firsthand just how hard it is to implement any new technology. I was impressed when Inbenta told me it typically takes three months to build and implement an entire chatbot.

"It also doesn’t require an army of people, typically a dedicated botmaster and some content experts to make high-level changes in the way the content reads and how the decision-trees are logically structured. Our clients’ technology changes so quickly that we have to be able to work really fast to stay relevant in this market,” says Forrester.


Another key aspect of a chatbot platform’s success is the ability to customize.

Enterprises have jargon – especially in the customer service sector – and this jargon is integral to the customer’s experience.

Being able to customize these conversations and have the chatbot keep up is paramount to botmakers and their ability to scale. This process can be very costly for some companies.

Forrester explained that they don’t need to use coding to facilitate this customization. That means it’s easy to maintain and implement, and the cost of change is minimal.

Similar to the language programming, Forrester says they don’t need hundreds of programmers to make these customizations occur seamlessly.

“Everything is based on the lexicon, which is able to interpret data and understand what the customer is asking about. It can even pick up on one-liners and understand what customers mean,” says Forrester.


Obviously, the end goal of chatbots isn’t to understand human language—it’s to facilitate smoother transactions online and interact with consumers in place of an actual person.

Customer support during Black Friday is a great example of a case when chatbots need the ability to scale up or down.

As the holidays quickly approach, customer service representatives are upped in anticipation of high call volumes. Even humans get asked the question, “are you a bot?”

However, Forrester informed me that not every company needs human representatives. Many don’t have any people on their workforce, opting instead for an army of chatbots to interact with the customers.

There are a lot of customers in the enterprise sector. Which, by definition, means there are also a lot of moving parts and problems to deal with. Because of this, any company attempting to do what Inbenta is working on will need to learn how to scale efficiently.

“That’s the biggest thing: speed of market, and being able to innovate for our clients,” says Forrester.


Understanding unstructured data is the new kid on the block for a lot of big companies.

“We’re excited about technologies out there for machine-reading,” says Forrester. “It’s a technology with the ability to structure data through neural networks. It can “read” pages of information, structure the data out of unstructured data and infer what it means. Language is a new focus in the industry for neural networks, and the challenge with unstructured data is huge.”

This post was published on the now-closed HuffPost Contributor platform. Contributors control their own work and posted freely to our site. If you need to flag this entry as abusive, send us an email.