Enterprises Can Leverage Data to Build Human-Centric AI-Platforms

“Before embarking on an AI project, companies must develop a clear strategy to ensure the quality of their data, which also means testing for bias and removing or accounting for it as much as possible,” says Kurt Muehmel, Chief Customer Officer, Dataiku, in an exclusive interview with EnterpriseTalk.

 ET Bureau: What challenges do enterprises encounter while making data the beacon of their digitization journey?

 Kurt Muehmel: There are thousands of challenges that enterprises can run into when implementing AI and advanced analytics. Two of the biggest ones that I’ve seen are going into an AI effort with the wrong mindset, and not investing enough in data quality.

Companies can easily fall into a trap when it comes to AI: they assume that they can slap an algorithm onto any business process and see results, and don’t realize that AI implementation is the opposite of a quick fix.

The reality is that AI and advanced analytics can help organizations in every industry, but these organizations need to be intentional in their usage and take the time to fully integrate AI into their business. AI isn’t a switch you can flip, it’s an incredibly powerful tool that requires constant maintenance and monitoring to yield real value.

Many of the maintenance tasks that tend to be overlooked are related to data quality, which must be considered at every point in the data pipeline. Even the smartest, most well-designed model will ultimately become obsolete if the datasets in use are not centralized, comprehensive, and up-to-date.

Also Read: How do CIOs implement effective AI governance

Before embarking on an AI project, companies must develop a clear strategy to ensure the quality of their data, which also means testing for bias and removing or accounting for it as much as possible.

ET Bureau: What steps can enterprises take to build effective human-centric AI platforms?

Kurt Muehmel: The first step that any company should take to building a human-centric AI practice is to truly incorporate their AI projects into their organizational structure. If an enterprise’s data scientists are siloed away from all other units, the insights that they’re able to surface might not make any meaningful difference in business processes, and they’re missing out on valuable perspectives from other teams.

As brilliant as data scientists are, it’s unfair to expect them to become instant experts in every department, industry or organization they serve. Subject matter experts need to be involved in all stages of development and operationalization. Additionally, business teams should be encouraged to provide input on their needs, interests and pain points. When operationalization is done in a vacuum, it is doomed to fail.

In order for AI implementation to be both effective and human-centric, all levels of the enterprise need to be brought to the table, with the right tools that allow them to be a part of the process. MLOps — the consistent, clear communication between data scientists and the operations team — is key in creating useful AI platforms.

Executives and business analysts should also carefully consider the KPIs that they hold their AI projects to. If the KPIs are not thoughtfully aligned with the problem the project was designed to address, it will be nigh on impossible to determine how much the project is helping the teams it was designed to empower, or how to improve it.

ET Bureau: How can enterprises leverage the power of AI for making informed decisions?

Kurt Muehmel: Data is only as good as the questions enterprises ask, and AI can only help companies make informed decisions when they have a clear idea of what processes they want to enhance. The first step in any AI effort should be a careful consideration of the problems that the team wants to address and the results they want to see.

Any AI project needs to be rigorously and specifically planned before it can kick off – a clearly defined vision will streamline the process and help the entire organization align on methodology, costs and deadlines.

Once the goals of an AI project are clearly articulated and agreed upon, the entire enterprise can be more clear-eyed and focused on figuring out what needs to be developed from scratch and where they can leverage existing assets. Training and developing models is an expensive, time-consuming process that requires a lot of trial and error. Unfortunately, these cost and timeline considerations can scare organizations away from AI projects that could have a real, positive impact on their bottom line.

That’s why I recommend that enterprises take a more open source-inspired view of data projects. It seems silly to write software from scratch when there’s code out there that will perfectly serve the needs of an enterprise, and the same thinking applies to advanced analytics — it’s very rare that an AI project needs to start from zero. Using preexisting, proven assets like cleaned and prepped datasets, proven algorithms, or pre-trained models can make development faster and more reliable, not to mention more cost-effective, and deliver data-driven decision-making at the speed businesses need.

ET Bureau: According to you, what does the future hold for enterprise AI given today’s ever-evolving data?

Kurt Muehmel: I think that the enterprise landscape will witness an increased emphasis placed on data governance across the board. As AI becomes ubiquitous, governments across the globe are slowly catching up, creating data privacy laws that match the prevalence of data in the modern world. Global regulations are becoming more specific and varied, and companies will need to create comprehensive and sustainable governance frameworks in order to comply and make data governance more streamlined and effective.

Also Read: Data Security vs. Data Governance – Building the Base of a Trusted Analytics Framework

In 2019, Singapore became the first country to adopt an overarching framework for AI governance. I expect to see more companies (and countries) follow in their footsteps and prioritize an ongoing governance strategy. Enterprises will continue to prioritize governance in any AI initiative and work to create frameworks that all levels of their organization can follow.

The applications of enterprise AI will, without a doubt, continue to grow in scope and scale, with more companies and organizations adopting it every year. But as all of these enterprises jump into the world of advanced analytics, they will inevitably bump into the roadblocks and sticking points mentioned above. As these technologies become an everyday part of countless industries, the world as a whole will get a better idea of how to avoid these traps and create practical and genuinely valuable AI solutions.