I have been doing a lot of thinking and reading about Artificial Intelligence (AI) and trying to understand how businesses can get started with Practical AI. One way for companies to start on this journey is with Chatbots.

Chatbots allow learning, training of staff and spark thinking of how AI can be applied in other areas of the business. Like with any new technology, the more you use it the better you get with it.

In 2016 I was at the Microsoft Partner Conference and saw the CEO of Microsoft, Satya Nadella, presenting on a project they were working on with McDonald’s (maker of my favorite Quarter Pounder with cheese).

Burgers and Coke

In this short video, you can see how a trained algorithm can accurately process a verbal order from a drive-through customer. The complexity of the order and accuracy of the output is incredible; especially considering this is an example from 2016. Once this solution is commercialized, I can see many drive-through order-takers becoming chatbot led.

Some things I noted from this example:

1.  McDonald’s has a lot of data to train the chatbot with. Think about how many drive thru’s they have around the world.
2. There is instant feedback on the correctness of the order as soon as the customer reaches the pick-up window. This data can be fed back into the algorithm to make it more accurate, since it can understand the before and after results of successful or unsuccessful orders.
3. Think of every phone-based order system out there, before long those call center operators could be replaced with a chatbot.

For this to happen, I think that we need to make sure that the chatbots we create are as human-like as possible. That the flow of the conversation should seem natural and not computer generated.

A great chatbot should be to pass the Turing test. We will look at an example shortly that illustrates this point.

How do you create Data Conversations?

We need to create chatbots to have data-driven conversations that rely on collected data to feed a conversation. How many times have you gone to a website that has a “chat” function that starts with a low-tech chatbot? It may start with something like, “How can I help?” so you either talk or key-in what you need help with, and it searches for keywords or phrases from your request and searches a knowledge base for potential answers. It responds with “Have you seen this link?” or “Read this document” … “Did that solve your problem?” “NO, IT DID NOT!!!” is what I’m sure a lot of us want to answer with in that scenario.

Part of the problem here is that this exchange is not how people talk or communicate with each other in real life. I can see a programmer seeing how this solves the problem, but for the person engaging with the chatbot it is far from ideal.

We need to consider the person who is talking with the chatbot, and make sure the user experience (UX) is on par with a human conversation. Yes, it will take a little more work for the developer to facilitate this engagement, but it will move us to a more person-to-person engagement experience.

How can I get Started with AI today?

Improve the quality and quantity of your data today, so that you are ready for the AI-driven future. To train an effective chatbot you are going to need huge amounts of data and the more relevant that data is the better.

Everything from talking about the weather in the customer’s location, or the latest sports news, to what the Kardashians are up to could be part of a natural conversation that is led by the algorithm.

Companies will move to gather micro snippets of information on a customer to allow for a personally tailored conversation. Added to this information are the details you already know about the customer.

For example, past ordering behaviors, conversation insights from previous interactions or even industry trends will allow you to tailor the conversation for each specific customer. Also, consider the public data that is available for training your chatbot: news, current events, location-based information, weather, and other government-provided or publicly available data.

You do not need to wait until you have a Data Scientist to get started. It is more important that you do get started using whatever resources and data you have. More than ever, location-based data will help tailor the conversation – the commute time to work, or shops in that area could all be used to create a more personal experience for the individual. Rather than the person feeling like they are just a number engaging in with the business.

Contact Us Form

How about starting with the Contact Us form on your website? You could replace it with a chatbot that is integrated into your back office systems. Then when your customer provides information, the Chatbot is able to use your customer database to personalize the experience. Trust is built as answers that are more relevant are provided by the chatbot in a human-like conversation, meaning the customer is more willing to provide more of their personal information in order to get a better outcome.

Give your Chatbot Personality – “It’s Alive!”

Create your chatbot to work in a lifelike way, as I stated above the goal should be to past the Turing test of the 1950s. Google gave a great example of this with Duplex in 2018 at their I/O conference. We do not have the budgets of Google yet, but there are steps you can take now to get started.

Do not wait for perfect research on how you can start data conversations today with your customers so that your staff learn what data to collect, what conversations to make data-driven, and how developers can be engaged to create the conversations.

Chatbots Can Go Wrong

I know some people have concerns about how this technology will affect our lives. Should humans be informed that they are talking to an algorithm rather than a real person, even though the experience will become much better for the customer?

We are evolving from organic (human) algorithms to inorganic algorithms. The inorganic will end up creating better experiences for customers, as they won’t have bad hair days, get distracted or be limited to only their own knowledge. It is not there yet, but we are all learning.

There is a lot to learn here. Microsoft research ran a project called Tay on Twitter in 2016. It ran into some problems and had to shut down.

The key takeaway from this for me is that we need to keep researching and improving our knowledge of AI and chatbots.

How do you Train a Chatbot?

With lots of data and machine learning. The chatbot needs to:

• Recognize questions
• Learn from its mistakes
• Consume external data sources
• Have access to additional content not just about the product or services offered but about the person who it is communicating with and their context

Let me know what you think in the comments below. I’d love to know if you’ve had any brilliant experiences with chatbots.