Pavan Bedadala, Senior Director, Product Management at Commvault, outlines how to avoid pitfalls and common problems by taking a considered approach in the first 60 days of an AI journey.
The pressure to be ahead of the AI curve is immense, with organisations and shareholders eager to see tangible results quickly. Resisting the temptation to rush headlong into ill-conceived projects and satisfy this demand requires fortitude, and a pragmatic approach.
Having a measured plan in place to evaluate AI, and articulating what will happen during the first 60 days, will reassure stakeholders that AI initiatives are under control. It will remind all those involved of the dangers of a reactive response to new technologies, and the importance of laying down properly defined, solid objectives for its adoption.
This is the point, at the beginning of the AI journey, when business leaders should endorse the processes and regimes necessary to help ensure that AI systems will be successful. Priorities should include clear specifications for project outcomes, and rigorous pre-production testing schedules. Leadership buy-in at this early stage is critical to set the modus operandi for all AI initiatives to follow.
AI is not the primary goal
It’s vital to confer closely with senior managers to pinpoint key business problems and requirements, and ascertain why they believe AI could provide the answer. From day one, focus on specific business needs, don’t get diverted by the latest AI buzzwords or trends.
Keep in mind that the overall goal is not necessarily to deploy AI-powered applications. It’s about understanding the issue that needs to be solved and finding the most appropriate solution, whether that incorporates AI or not. Encourage reviews of existing procedures and decide whether positive changes can be made in the short term without the need for immediate investment in AI.
As with any new technology, there may be misconceptions about where AI would be a good fit. Explaining its value for specific tasks and outcomes will promote better understanding. Highlight that it is typically data-driven areas characterised by mundane number crunching, laborious repetitive rules-based tasks, and painstaking analysis of data, that benefit most. Provide an overview of how AI tools can bring significant efficiency gains to these functions as they excel at tasks like data entry, reporting, in depth data analysis, and automating workflows and processes.
Then, clarify which areas are likely to gain the most; it is these areas where AI should be applied.
Set realistic expectations
Having identified the right project areas for AI, it might sound like plain sailing from here-on, but AI models must be fed reliable data, then trained and tested to ensure they are providing accurate, unbiased results. Getting to this stage requires diligent preparation.
Different data sources need to be identified and audited for quality and compliance. Otherwise, feeding poor data into the AI models will deliver equally poor and, potentially, harmful results that risk scaling as the system grows. It is important business leaders recognise that it’s not a case of pouring in any quality of data and, overnight, expecting the AI model to deliver accurate results and reliable insights.
During this early planning stage, set expectations about the readiness of data for AI projects. Many organisations have vast stores of data collected over decades and are recognising its untapped potential. Whether that’s for creating more personalised buying experiences, improving customer service, predicting future trends or developing new products, there’s all manner of insights waiting to be uncovered. But the initial data enablement takes time.
Ensure that business leaders understand that data often sits in disparate silos and must be consolidated and cleaned before it can be utilised in AI models. This is an area well-suited to AI tools devised for efficiently removing duplicates and formatting data ready for further processing. Emphasise to stakeholders that it’s imperative that this process is carried out before ingesting data into models.
When the data is ready, the next step is rigorous testing of the AI model to ensure that it is running properly.
Define success metrics
Clearly define what success looks like for every AI model, including accuracy, error rates, and specific performance benchmarks. Use curated test data sets reflecting typical real-world scenarios and uncommon ‘edge’ cases that fall outside the expected range of inputs or conditions. This will help identify weaknesses and minimise nonsensical outputs. Ensure that development processes are transparent, and how the model arrives at decisions is documented and explainable.
After this stage, AI models can start to analyse massive datasets revealing trends and insights that would have taken days or weeks of human effort to uncover.
However, never lose sight of unplanned consequences that might arise from such initiatives, such as inadvertently giving staff access to confidential data, or their impact on cybersecurity and compliance responsibilities. And, with AI evolving so quickly, models must stay under constant review and be capable of incorporating enhancements to maintain optimal results.
By taking a considered approach in the first 60 days, organisations can ensure their AI initiatives start out on a firm footing, avoiding the pitfalls and expense of embarking on poorly thought-out projects that ultimately fail to deliver.
