To master AI as a revenue catalyst requires first understanding its costs

Almost immediately after it became apparent that the world was changing under our feet, artificial intelligence (AI) emerged as an enabler across the region. Suddenly, these technologies were not just being advocated by their vendors, but by analysts who pointed out that AI could help not only with battling the Covid pandemic, but in fomenting slick, robust economic recovery. Each industry could do something concrete to reinvent itself in the wake of seismic shifts in customer needs.

Analysis of economies across the Middle East and North Africa (MENA) predicts the region will respond dramatically to these arguments. The benefits of machine learning and advanced data analytics are now well known, but not everyone is enjoying success with these tools, because not everyone has the optimal migration methodology in place. If ingenuity does not flow quickly from AI investment, costs can outweigh gains before the technology has had a chance to add value. To prevent this, it is advisable to analyse each use case to see which will add the greatest benefits in terms of productivity, operational efficiency, customer retention, and so on, and then price each of them to arrive at a cost-benefit ratio.

Lower cost use cases can be leveraged to build trust in AI, by demonstrating value for each business stakeholder. But management of projects must happen at an umbrella level. Allowing unit heads to run their own projects in isolation can quickly lead to failure. Of course, this enterprise-wide AI approach hinges on being able to accurately analyse costs.

So, what are the costs of AI? How can they be separated and examined in isolation?

Getting data ready

Data is food for AI. If not properly formatted, then your AI will choke. Value generation can only come from effective collation and preparation, which takes time and represents a cost. Another term for this process is data wrangling — further indicating just how difficult and time-consuming this phase of the digital transformation project can be.

This is where the umbrella strategy will reap dividends. By taking an enterprise-wide approach to use cases, data can be warehoused on the same basis, meaning the organisation’s data-cleaning phase is implemented once, bringing siloed stores together in a uniform fashion rather than having separate data-homogenising projects for each
use case.

Deploying solutions

Moving to production brings several headaches, especially when each project may involve a variety of workflows and stakeholders. Operationalisation, therefore, can contribute greatly to costs, as development cycles spill over from weeks into months. Here, costs emerge not only from labour but also from lost revenue because the solution is not in place in a production setting. Solutions to these costs can be found in process optimisation. Consistency in the development lifecycle can serve all use cases. Creating these directives can be thought of like the data-cleaning phase. If it is done properly in advance, it no longer arises as a concern in subsequent use cases. When development teams already know how to assemble and package code, and are well-versed in the formalities of releasing it, then business value can be added more quickly.

Acquiring and retaining talent

By getting the first two costs under control, the third becomes a more straightforward issue. The region is amid a skills crisis within the types of roles that typically deliver digital transformation. Data scientists and AI specialists are rare and expensive to attract. And while the cost of acquisition may be an externality that organisations cannot control, the retention or loss of such skills is entirely of their own making.

By ensuring repetitive tasks like data-cleaning are reduced to one-shot projects, enterprises create roles that are more about solving problems than about chore-like grinds. In this regard, proper resourcing will be a vital component of ensuring that an organisation needs only acquire a skillset once, rather than having to spend money and time on repetitive recruitment.

Maintaining models and technologies

When data changes, the results from existing models may not tally with reality, leading to a potential cost, and such a cost applies to each use case. So-called MLOps can be a means of controlling maintenance costs, unifying the task across use cases. At the same time, AI technologies themselves are evolving and attracting different stakeholders to new capabilities. Again, if an organisation has kept its enterprise-wide strategy in place, it can more readily evaluate new business cases.

From cost centre to revenue source

The region’s innovators will struggle to realise the potential of AI if costs are poorly understood. As smart technologies take more prominent position among the economic activity of nations, enterprises must look for efficiencies in their implementation. By scaling with due diligence, winners will accrue benefits more reliably and ensure that AI becomes a source of revenue, rather than a drain on investment.

Sid Bhatia is the regional director – Middle East and Turkey at Dataiku