Artificial Intelligence will redefine the nature of BPO employees

The romance of AI of course is to create a digital customer interaction interface that appears human.

What can AI do? 

Software solution Blue.ai can, for instance, run an interactive platform for holiday bookings. A customer could speak to a bot about preferred locations, budgets, and other preferences, and the bot would run searches and retrieve information as if it were a booking agent.

Ahmad Daamoush, chairman and chief executive of Antwerp Technologies, which owns Cyprus-based Blue.ai, explains that the system at full-scale deployment can tell the user: “There are no flights to Paris for the date you want to travel and the next day’s flight doesn’t fit in your budget. How about traveling to Amsterdam first and coming to Paris from there?”

For a financial services client, a similar system by a competitor, Veritran, can authenticate transactions on a video call with an AI chatbot the way some banks offer premium customers to connect with branch managers.

Most of the use cases of customer-facing AI tend to hinge on language and the complexity in which it is interpreted and reproduced.

Orange particularly relishes its achievements in building language models. “We tune the existing models to recognise combinations of words and to understand the interactions between concepts. And that, in turn dramatically improves the accuracy of the model,” said Jarrett.

The company has even taught its bot to speak Wolof, a Senegalese language spoken merely by an estimated 10 million people.

The crux of that lies in data that needs to be sanitised and interpreted by the machine learning engine.

“We support integrations with a ChatGPT because we already have a set of customers who don’t bring their own data or want us to support them in their data,” said Daamoush. However, the more relevant data that will differentiate consumer experience would likely be proprietary.

What AI is already doing? 

Arguably, the most advanced consumer-facing AI bot in use would be Amazon’s Alexa. Even that, while greatly improved, is still on a learning curve. Anyone who has interacted with a telecom company chatbot knows the experience can be quite frustrating. The companies recognise this.

Vodafone’s chat bot, for instance, promises to connect a user to a human the moment it can’t find an answer.

It is perhaps for this reason that most companies use AI for inward facing applications. “So we see AI as a way to give superpowers to our internal employees,” said Jarrett.

At International Business Machines (IBM), which started the AI program with Watson a long time ago, AI is being used to reduce the work of network engineers. Barely a decade ago, companies deployed large data centres. Now, much of that has been moved to a fluid cloud which moves content across locations based on requirement.

Ciaran Ryan, Vice President of Software Networking at IBM, has been deploying AI solutions that can deal with network issues. Earlier, he explains, when there was an error or a traffic buildup on a company network, the system would flag an error by generating a ticket. Each ticket, irrespective of severity, had to be looked at by a systems engineer. That frequency has been reduced to a small fraction with AI becoming capable of dealing with many of these tickets.

At Infosys, AI is being optimised to become an ‘avatar’ of a programmer that can help with mundane coding.

Will AI obliterate jobs? 

In their current forms, yes. The ability of machines to take decisions like humans will render high-volume, repetitive jobs redundant. But that day is still far off.

Well-interfaced systems, like ChatGPT, still have a lot to learn and can hallucinate, or create incorrect results that appear to be correct. Sometimes, “the model will tell you something that's very convincing and it's wrong,” said Jarrett. In these cases, it takes an expert to call the bluff. So the oversight of a person is still necessary.

IBM’s Ryan explains there is a confidence issue in AI’s ability to take the right decision, and so far the engine prompts a solution that has to be authenticated by a person. “That trust is increasing, to allow the AI operations decisions to go ahead,” he said.

At Orange, this means the composition of solution teams is going to undergo change. Where earlier technology was heavily dependent on engineers, the teams now undergo a change and focus on training and correcting AI-led decisions because a bias can creep into a model because of a data skew.

With live inputs that will self-correct, explains Jarrett, the teaching teams would evolve into a ‘champion to the customer,’ a machine learning software engineer, and a data scientist.

More importantly, the problem individuals are solving has to be of a higher order than bodies in the long run. “We're not automating all these people out of a job. AI is giving them powers to become more efficient at their job, but that requires them to learn new skills,” explains Jarrett.

Good AI is costly

“When you pay everything can be done,” said Daamoush, but the cost of an AI solution can range from half a million dollars to $177 million, in his experience. For large companies like PepsiCo or Unilever, investing is an easy decision and worth the risk.

As you trickle into the medium segment of businesses, while the technology risks disadvantage them against competitors, it may not be viable to invest in the solution given the cost and effort required to train the system.

The language models that Orange prides itself in are themselves not always practical. “We’re interested in but not committed to using large language models. They're extremely power-consumptive at creation. They're fairly expensive to run. They can hallucinate,” said Jarrett. “Our teams are working hard to see if we can mitigate these concerns and risks.”

The ubiquitousness of AI will therefore force a rethink for business process outsourcing companies on their staff composition. That opens the recruitment floor to a new, non-engineering, creative employee in the near future.