As Artificial Intelligence (AI) continues to creep into our daily personal and business lives, brands are starting to feel the pressure to incorporate AI into their marketing and advertising strategies. They are right to feel the pressure – when executed correctly, AI is the key to connecting brands with the right audience and giving them the competitive edge they need in today’s market. Unfortunately what I’m seeing more and more of is people jumping on the AI bandwagon without understanding what it is and the best way to utilize it.

Probably the biggest AI hype right now is around the use of a chatbot, or a service powered by rules and artificial intelligence that allows the user to interact via a chat interface. Instead of clicking buttons, you type (or say) a command and expect the chatbot to execute it. The service could be anything from functional to fun – think Facebook Messenger, Apple’s Siri, Amazon’s Alexa, Slack, Telegram, etc. We’ve seen a surge in chatbots as businesses are using them as a way to communicate with their customers or organizations to communicate with their clients, or organizations to communicate with employees. Millennials have jumped on the ‘bot bandwagon’ in particular, as a recent poll indicates that 86 percent of millennials agree that brands should promote deals, products and services via chatbots. Also about 58 percent of millennials respondents in the same survey said that they had a positive experience when interacting with a chatbot.

So every brand should start using chatbots, right? Not so fast. Like all AI, chatbots are still in their infancy stage and the technology needs to be refined. A recent report claims that 70% of Facebook’s chatbots “failed to fulfill”, or basically that chatbots couldn’t understand what users were saying, and in some cases humans had to step in and intervene. 70% is a large failure rate – think if 70% of airplane flights “failed,” people would never fly! The issue is looking at AI and chatbots as a ‘one size fits all’ when in reality, brands need customized intelligence and data to successfully deploy and utilize AI.

So how do you get started in developing and executing a chatbot program that will be successful? We need to take it back to basics like you would if developing a product. First you have to identify the business problem you’re trying to solve, the audience it speaks to, and then you worry about the data to fill the AI brain with so the chat bot can learn and get better. Just like when big data became all the rage, companies quickly learned that if they inputted wonky data, they received wonky results. Chatbots are very similar if it starts to learn something inaccurately or is fed wonky data to start it will produce false results and humans will need to reset and intervene.

You also have to take into account what vernacular and vocabulary you want the bot to be able to understand and what you want its knowledge base to be. Most organizations starting out using bots do not even think about these very basic steps before deploying something, so without you teaching the toddler wrong from right the bot will not produce the desired results and whats more will grow in the wrong direction.

Many banks are starting to use chatbots in the right way, for example In Sweden, Swedbank’s Nina Web assistant achieved an average of 30,000 conversations per month and first-contact resolution of 78% in its first three months. Nina can handle over 350 different customer questions and answers. Several other banks in the UK and internationally have similar systems in place or are trialling them.

Aeromexico debuted Aerobot last month at Facebook’s F8 conference, which in the first six months took the airline’s automation from 0-96%. The customer service bot even understands various Spanish dialects because the airline and partners fed it historical travel customer data to give it an education from the start.

As we move into the World of Everything (WoE) or a 365° view of everything tangible and intangible that surrounds us and how to use it to our benefit (think how casinos pump scents into their facilities to change players moods), brands need to customize AI into their business strategy and not the other way around. We’ll replace artificial intelligence with cognitive intelligence by coupling biometric data with AI data to create the most robust form of humanized intelligence. We will see a shift from statistics-based AI to the idea of an exactitude-based AI that looks at data and solves it, creating a customized model.