In our last three articles on AI and AI consulting, we shared our thoughts about customer centric AI transformation and specific steps for implementing AI technology. Here, we are zooming in on data strategy to address some of the common data-related obstacles, in hope to help organizations progress on their AI journey.
To leverage AI without a data strategy is like expecting a race car to win without fuel. “Data has been called ‘the new oil.’ Granted, the world is in the process of transitioning its way to other energy sources. Even so, the idea is clear: information is power.” https://aisera.com/blog/what-is-aicx/
Before coming up with a customized data strategy for your organization’s specific needs – be it obtaining high quality and large quantities of data, identifying applicable business use cases, building data infrastructure – first ensure that your data foundation is safe and sound: cybersecurity, privacy, and responsible use.
Protecting customer and partner data is top priority. Protecting data privacy and security is an ongoing process, as a company’s internal data lake expands, the amount of data accumulated, integrated, and applied to AI will multiply. Whether the desired business outcomes are cost savings, operational efficiency, reducing risks, enhancing compliance, streamlining marketing and sales, increasing customer satisfaction, improving customer journey … Underpinning them all is a strong cybersecurity foundation that needs to be constantly reinforced.
“To ensure robust data protection and seamless access for authorized personnel, using AI-powered security products to fortify end-user devices and services while simplifying information security and compliance,” a 2024 Microsoft blog suggested that “utilizing Microsoft data security solutions to safeguard their environments and confidently prepare for their continued adoption and deployment of AI”.
Is AI the best solution for protecting data security? The fact is, not even Microsoft is immune from data breaches and cyber attacks. Logic and common sense suggest that keeping humans in the loop, in a human-AI partnership, is safer than relying heavily or solely on AI solutions for cybersecurity. Technology alone does not solve technology failures perpetrated by human hackers. (Understandably Microsoft is in the business of selling their AI solutions.)
Once you have an infrastructure of data security in place by using humans, technology, procedures, systems, checks and balances, AI, and whatever else, The next step is to develop a data strategy to guide data collection for specific use cases.
But hold off: In order for your AI to be scalable in the future, leaders need to first tie their data strategy to the needs and wants of their customers or the end users. The truth is, what corporate leaders THINK their customers want is not always the same as what their customers really want. To avoid wasting resources, time, and investment in building data and AI technology in the wrong direction, first start by investigating the true needs and wants of the end-users, before formulating your data strategy.
Envision how the creation and collection of more and/or higher quality data will result in more and deeper customer relationships, and accelerate your business outcomes that lead to a higher level of customer satisfaction. “Work backwards from your opportunities towards data requirements,” per an Amazon whitepaper.
Only after you have aligned your priorities with key stakeholders and identified opportunities for AI adoption – all in line with your growth objectives – can you then identify applicable business use cases to develop your data strategy.
At this stage, you need to analyze your business processes and identify the most important areas where AI can improve, such as decision-making for better products and services, enhanced customer experience, and optimized efficiency. Once you have identified applicable business use cases, you are on the way to your comprehensive AI strategy roadmap.
Again, the business use cases must be in alignment with the organization’s long term mission and goals, not just for a myopic, temporary patchwork of fixes.
After identifying the areas where AI can create value, you can then determine the data required to train AI models, prioritize data source opportunities, and establish a roadmap for AI implementation with measurable business outcomes.
Next, find out the exact data assets and sources the above initiatives and opportunities depend upon.
The value is in the quality of useful data. Large amounts of useless data won’t do you any good. Determining the value of data needs to be in line with the organization’s AI goals and data strategies. Deciding what types of data to be collected and developed is often not the responsibility of AI engineers, but that of the leadership of an organization and/or their delegates.
Like a moat outside a castle, a “Model of Adversarial Training” defends machine learning models from adversarial attacks by training them on adversarially perturbed examples. Unique and valuable data specific for your verticals can enable you to acquire customers faster than your competition, therefore must be protected with a “moat”, especially when your platforms have network effects, you have the advantages for scaling faster than your competitors.
As for the exclusive data that is fed to bespoke GenAI for enterprise vertical use, “your data is incredibly valuable, that should be encrypted with your keys. First, build a data set. Second, build an LLM. Third, build applications”, as noted in an article by Joanne Z. Tan, quoting Dr. Muddu Sudhakar, Co-Founder & CEO of Aisera. Dr. Sudhakar also advised enterprises to “bring AI to data, leave data where it is.”
What is “high quality data?” – valid, auditable, timely, relevant, and valuable information obtained from trusted and private sources in a responsible and secure manner. It is often exclusive to an organization.
Why is high quality data important? Towards the end goals of using AI to increase efficiency and customer satisfaction, solve related challenges and predict customers needs, your data strategy needs to focus on developing high quality data to train AI algorithms. Only with high quality data can AI accurately predict and deliver desired business outcomes. “With high quality data, AI can generate better insights and predictions, leading to better, deeper, and more customer relationships and CSAT (customer satisfaction).” https://blogs.microsoft.com/blog/2024/04/24/leading-in-the-era-of-ai-how-microsofts-platform-differentiation-and-copilot-empowerment-are-driving-ai-transformation/
Data storage and management: When it is safe, it is preferred to have data stored in an internal central location, or in fewer internal data warehouses, in order to enable convenient access for AI engineers.
External data acquisition: Depending on the specific needs of an organization, a data acquisition strategy may be needed to strategically and systemically purchase relevant and useful data from external sources.
Monetize synthetic data: Conversely, if you have synthetic data that is highly valued by those who need it, turn it into a revenue source.
“Land grab exclusive data. Monetize synthetic data.” – Jeremiah Owyang, Blitzscaling Ventures General Partner, as quoted in an article by Joanne Z. Tan, 10 Plus Brand.
The principles in the above article and our previous articles don’t just apply to AI. They apply to any digital transformation initiative.
©Stephen Wullschleger all rights reserved.
If you need to leverage the highly experienced AI consultants and seasoned management advisors at Wull to help you formulate sophisticated data strategy, please contact us for further discussion. Thank you.
Resources:
https://docs.aws.amazon.com/whitepapers/latest/aws-caf-for-ai/your-ai-transformation-journey.html
https://aisera.com/blog/what-is-aicx/
https://10plusbrand.com/2024/06/05/gen-ai-jobs-data-joanne-tan-ai-experience-designer/
AI Leadership: Decision Making Steps for AI Assessment (Pt 1 of 4)
Leave a Reply