Everyone wants an easy answer. For the digital marketer, that can be hard to get…and expensive.
Online retailers are investing in AI to glean that last sliver of insight from their data. Or they can entrust an intelligent agent (IA) with that mission.
IA? AI?
“The agent approach focuses on the objective and the action, as opposed to data collection and prediction,” said Matt Gershoff, CEO of Conductrics, which provides an intelligent-agent based platform for A/B testing and optimization.
AI relies on machine learning, which has to plow through massive amounts of data to acquire expertise. Even then, the data has to be “cleansed”. Duplicated entries have to be removed. Scale must be consistent. And a human has to revisit the AI to make sure it does not go off course. No one wants to replicate the Microsoft Tay disaster.
“IA is really just reframing the optimization problem. One still needs to deal with any and all issues around automated systems.” Gershoff said. IA faces the data, instructed to “take the best action in a given situation.”
One way to think of an intelligent agent is “weak AI”, explained Gershoff. “It is not machine learning. It’s more a way of framing a problem.” It’s an interrelated three-step approach: 1) data collection; 2) prediction; and 3) action.
“Data collection” for an intelligent agent is how it perceives the environment it operates in. The agent can predict the best action as it makes the optimal choice, but it also needs to take the action to see if the choice was the best one. If it wasn’t, it learns from that experience to make a better choice the next go round.
Applied to marketing, an intelligent agent can examine in incoming stream on online visitors. Based on user attributes, the agent can randomly serve one of three offers. The intelligent agent keeps score of which offers scores the most conversions, and will rate that offer as best and serve it up more frequently, Gershoff explained.
Down to earth
Firms are already applying intelligent agents to solving practical problems. NeuraFlash is a Salesforce partner, crafting bot solutions that rely on Salesforce’s Einstein AI platform. The firm is using the intelligent agent approach to craft chatbots for customer service.
“A lot of companies have a lot of data,” noted T. Brett Chisholm, CEO of NeuraFlash. “They have millions of conversations saved in CRM every day.”
“We, as a company, focus on how to leverage existing data to create the initial model,” Chisholm continued. For NeuraFlash, the approach is to examine and analyze that company’s customer service data, looking for patterns of frequently asked questions, like order shipping status. One could craft a list of 10 to 20 FAQs that the chatbot can answer, relying on the intelligent agent to pluck the relevant information from the company’s data. Such a solution can be up and running in 3-4 months, yielding a quick return on investment, Chisholm added.
In practice, the customer should be able to ask a routine question in plain language about an order status, price of a good, or its shipping cost, and get an answer. Do this successfully and the customer is delighted, Chisholm noted. But get it wrong, and the customer will be frustrated.
“A lot of automated systems are not integrated, so they get caught in a loop.” Chisholm said. “If the bot says the wrong thing, it can be very problematic.” When the agent-driven system senses the customer’s frustration, or knows the right answer was not delivered more than once, then it automatically switches the query to a human for direct intervention.
“The nice thing about an intelligent agent is that it brings you to your destination, Chisholm said. But the solution using the agent has to be focused, and right now that means taking care of routine “high runner” problems. “If you build too many things, you are looking at an overlap and it makes the model worse,” he said. After all, the goal for the intelligent agent is to handle routine “high runner” problems so that a human is freed up to resolve more complex issues for the customer.
Answers are not decisions
Globant is another company pursuing the intelligent agent/chatbot approach. The enabling tool, however, has been a recent ramp-up in natural language processing as the input for the agent, noted, J.J. Lopez Murphy, technology director for AI at Globant.
That combination of natural language processing and IA technology is yielding a “chatbot with brains” that can interact with the customer, Lopez Murphy said.
One example Lopez Murphy gave was for an intelligent agent acting as a legal assistant. A user can ask the agent to look up a price in a contract. The agent-driven bot would search the contract and retrieve the answer. As contracts can run tens or hundreds of pages, being able to search the document using a voice input can be a time-saver. The intelligent agent “can extract the answer without previous knowledge,” Lopez Murphy explained. The task is as simple as 1) select content, and 2) extract answer.
The agent provides information, but cannot make a decision for you.
