Ryan Falkenberg. Image: Supplied.
JOHANNESBURG - The potential of AI is clear. We all get it. We read how Elon Musk, Uber and Airbnb are transforming entire industries with AI. We see how established technology giants such as Microsoft, Apple and Google are investing billions into AI-powered features and start-ups to ensure they maintain their dominant market positions.

And yet within most traditional businesses, AI remains something confined to boardroom conversations, innovation labs and the odd sponsored pilot project. Few have managed to take an AI-driven solution into full scale production. Why?

A question of perceived impact 

For many companies, their existing business model has been designed to enable a wide staff complement to deliver a consistent service by following a pre-defined business formula. This formula is captured within policy and procedural documents and training materials, or coded into various operational technologies and machinery. And while some decision-making is automated, the majority is still performed by people. 

To move to a model where most decision-making, as well as decision execution, is performed autonomously via technology is a big jump. It requires major changes in how people, processes and systems work within the organisation. It also requires significant work in capturing the decision logic in people’s heads, and automating it into intelligent systems. For many, this effort is daunting, and so they opt to kick the AI can a little further down the road, until ‘our competitors adopt it’ or ‘the technology matures’. Unfortunately, in many cases, when this happens it probably will be too late.

Pressing problems

Cognitive computers are amazing at predictive logic. They can crunch through huge sets of data, and pick up patterns that few humans could detect. Based on these patterns, they can predict what is most likely going to happen next, given a similar pattern emerging. That’s how Amazon and Netflix are able to suggest books and films that we are most likely to enjoy, based on who we are, our past choices, and the choices of other people similar to us. 

Predictive logic, enhanced by machine learning off huge data sets, allows companies to optimise things like next best product recommendations and error identification and prevention. It thrives off massive data volumes and improves as it learns.

In many SMEs, however, transactional volumes are not at the level where this form of optimisation is possible, let alone viable. Their most pressing scaling challenge is not improving decision accuracy through improved prediction. It’s improving decision consistency across all staff and customer engagements. This relates to scaling the logic that staff are required to apply when they make decisions and take actions defined by company regulatory, policy and procedure rules. Its prescribed to them, and they have no creative space to change it. And in most cases, they are required to prove they applied it correctly, for compliance or regulatory reasons. 

Prescriptive logic has traditionally been captured in two ways. The first is via knowledge objects, where the decision-making logic is described for the person. The second method is to map the logic, typically using decision trees or process flow diagrams. Neither method allows us to reflect the true multi-dimensional nature of the required logic, given all possible contexts. This is where most knowledge bases, scripting tools, iBPMs and expert systems struggle  – when context matters. For many companies, it is here where the most pressing pain point resides – ensuring a more consistent, compliant application of their prescribed decision-making formula across their sales, service and operations areas. 

And in many cases, it is here where data is limited, and the need to show that the right factors and rules were considered in making the decision makes many AI solutions unviable. 

Dependence on IT to make it work

Another reason for slow AI adoption is that most AI solutions still require highly skilled coders with experience in programming languages such as Python to get them working. This limits businesses’ ability to accelerate adoption, as it immediately requires a sponsored IT project that must be approved via the typical change approval processes. There is no room for business teams to ‘test our thinking, and give this a try’. As a result, only the large projects get approval, limiting the adoption of AI across all parts of the business.

A platform that business can use 

What is needed to allow more businesses to actively begin their AI journey is a platform that business people can access and work with that allows them to capture the logic they want to replicate, and to build Virtual Advisor apps that they can then deploy and test, without too much technical complexity. 

This would allow an internal contact centre team to build their own call expert, applying all the logic their internal experts would apply to every known call, for example. They can simply assign a small team to tackle this within normal operational hours, and after a few weeks they can deploy and test their own solution within their own environment, and fine tune their logic as their agents use it. Within a matter of weeks, they can bring down average handling times, improve first call resolutions and reduce their training requirements.

The same applies to the sales, HR and technical support teams. Or the operations team. Or the compliance team. Just as Word Processors made capturing knowledge something every department and team could do themselves, so Virtual Advisors offer a platform that allows departments to capture the logic they want everyone to consistently apply and make it available to teams as they perform.

And then, once they have proven the impact of their logic and want to optimise its impact, they can then engage with IT to integrate their apps within operating systems and chatbot front ends, as well as enhance predictability by leveraging machine learning off their data patterns. 

Without a platform that can give business the ability to get started on their digital journey, AI will continue to remain an expensive, exclusive capability that resides only within the IT and digital teams.

Fortunately these platforms now exist. Called intelligence augmentation (IA) systems they are available today and can be used by companies as a first step on the road to fully-fledged AI.  

Ryan Falkenberg is co-founder and chief executive of Clevva, a technology that allows non-coders to capture expert logic into Navigation Apps.


The views expressed here are not necessarily those of Independent Media.

- BUSINESS REPORT