AI needs the rigour of routine to be successful
- Posted on April 20, 2023
- Estimated reading time 5 minutes
ChatGPT has reignited the hype around artificial intelligence (AI). The fact is that, since 2017, the narrative has painted a rosy picture of enterprise AI adoption. It’s heavily implied that AI has matured, and all enterprises have an AI capability. The reality, though, is different.
In my conversations with organizations, it’s clear that we are nowhere close to getting significant value from AI implementations as they exist today. In a study by MIT Sloan, 70% of companies reported minimal or no impact from AI. Even if they made significant investments in AI, 40% of organizations did not report business gains. Moreover, 87% of these projects never even make it to production. As a practitioner in this field, I am quite disappointed and partly want to take blame as well.
While primarily data-based companies like Nvidia and Facebook have made progress on the AI front, traditional organizations haven’t seen significant progress in the last four years. Yes, some AI-based automation comes into play for demand forecasting or routing the call to the right person in contact centres. But the bigger impact of AI is missing. It’s still being used mainly for insights, and honestly, it should be doing much more by now. From an infrastructure point of view, the market has arrived – superior computing power, data storage, ease of deployment are no longer challenges. But still, success is elusive.
It makes you wonder why, doesn’t it?
Why, despite the hype, is AI failing enterprises?
One of the main reasons is that AI projects are still considered a technology implementation and not embedded into the business processes.
Secondly, AI implementations have become like pet projects – interested people are deploying them as point solutions to solve a specific problem without looking at the larger impact on business. So, for example, if you have a fraudulent check problem, deploy some vision-based AI and read the check to detect and validate the signature. This band-aid approach to solve quickly for what’s broken, to add a little bit of intelligence to sort out recurring issues, works in the short run. But it affects the business’ ability to get larger outcomes.
And finally, there is resistance from the organization itself; people are unwilling to adopt a new way of doing things to work alongside the machines. Add to that there is a genuine concern on ethics and fairness and unfounded fear of job loss.
How do we ensure sustained success for AI?
A successful AI journey at speed and scale requires four things – Platform, Governance, People, and a Business-led Approach.
- Platforms – Though it sounds technical and obvious, but Platform is an important step towards achieving AI at scale. We need two kinds of environments/platforms to ensure AI success: Deployment environments (Dev, SIT, UAT) and Exploration environment. The first one – Deployment environment – we are all familiar with. The latter, Exploration environment, is required to allow AI users to continuously hypothesize and experiment with the data – based on broad themes like revenue growth, profitability growth, customer experience, compliance, and more. The successful AI/ML (machine learning) experimental models need to be moved to the Deployment environment and eventually to production.
- Governance – It’s important to look at AI governance from two angles. One from an Ethical standpoint to ensure all experiments are conducted within regulatory guidelines of each country/organization around data privacy and security. Asking questions about the purpose of AI, whether it harms anyone in any way, violates any privacy, is a good starting point to establish guidelines. The second aspect is the Technical governance that enables continuous adjustment. Technical governance puts in place guidelines on how the code is deployed, how model performance will be monitored, how will data drift and feature changes will be accounted for, and so on. Lastly, more than any technology, the AI has to be human centric.
- People and skills – We need to think carefully about the AI skills you need in the organization. Does everyone need to be a citizen data scientist or developer, or is it enough that they have data literacy? Do you need to outsource all AI/ML skills or keep some core competencies in-house? Is AI the job for IT alone, or does the business have equal, if not more, ownership? My personal view is that citizen data scientist or citizen developer remains a buzzword and not necessarily required to scale AI in an organization. However, understanding AI implications and data literacy is a must for everyone in the organization.
- Business-led approach – One of the core reasons why AI is unable to scale is the lack of trust and collaboration between business and IT. This is why business teams appreciate Power BI dashboards but don’t trust the predictions made based on AI models. This lack of trust stems from underappreciation of the technology or having been disappointed in the past. Unless AI decisions are explained to business, unless these teams are involved from the get-go, we won’t see success.
I believe the first step towards improvement is to accept reality. And the reality is that a pet project approach and heroic flash-in-a-pan success won’t result in sustained success from AI initiatives. For long-term success, we need to make AI routine, governed, and almost boring! Because at the end of the day, it is not the excitement but the routine that will give organizations and businesses the sustained success at scale and at speed.
As a starting point, we invite you to register for a complimentary two-hour workshop to explore opportunities to drive business value from AI in your organization.