Fine-tuning has been the sole method by which a model could be adapted to accomplish specific tasks. Today, the current large language model can be prompt-engineered to achieve similar results. An AI task that would have taken 6 months in the past can now be accomplished in a matter of minutes or hours.
This development opens up numerous opportunities. At the same time, it’s important for product and engineering teams to remember that AI is not a strategy; it’s a tool that helps you achieve your strategy. If you’re building AI just for the sake of AI, you’ll waste time and resources rushing products and features to market that users will ignore or quickly abandon.
In order to build product capabilities that harness the true power of AI, product and engineering leaders must apply the tried-and-true strategy of customer-centric product building to the promising potential of integrating AI features. Delivering customer-centric AI means offering AI product experiences that are highly targeted to individual users, protect customer data, and empower users to choose how much or how little they want AI to show up in their product journey.
This sounds obvious, but it’s easier said than done – look at all the AI features in the market today that look like afterthoughts and add-ons. In fact, I believe there are three key pillars of product development that teams should lean into to build meaningful, customer-centric AI product experiences: data privacy, data governance, and user choice.
Privacy and protection are king
If users are going to try a product, let alone commit to it, they need to trust the company that made it. At the same time, companies have to collect user data to create great AI experiences. These two things are naturally at odds.
Assuming that selling customer data is not a fundamental part of how your company conducts business and generates revenue, customers need to understand the checks and balances you have in place to ensure the security and non-sale of their data. It begins with adopting a privacy-first mindset and ensuring that your business model aligns with this principle. By embracing a privacy-first model, you not only become a responsible corporate entity but also earn your customers’ trust, which in turn will result in business results.
Examine the data that exits your environment and assess whether it raises privacy concerns. For instance, it may be acceptable to send metadata to an AI provider like OpenAI, but sending personally identifiable information (PII) should be avoided. Once you have the right protocols and tools in place, regularly conduct audits to confirm that your company’s privacy measures are compliant and that your technology has privacy and security controls directly integrated within it. Maintaining the highest level of trust with customers when it comes to their data is completely essential for any AI product to be successful.
Become a master in data governance
In a recent survey of Chief Data Officers, 45% of CDO’s ranked clear and effective data governance policies as a top priority. It makes sense – without data governance, there’s no guarantee that the data being used within an AI model is accurate and or even reliable. Even with proper governance, data can become chaotic. Making data governance a top priority at the onset of product building helps to ensure responsible stewardship of customer data throughout the AI development lifecycle. A well-oiled data governance machine enables companies to train the most accurate AI models, which in turn builds customer trust.
While there are many aspects of data governance, one key element that I find many companies struggle with is data discoverability – understanding who needs access to which elements of the data, and then making that data available to the right internal teams. If engineers aren’t able to find or access the data they need to build and fine-tune models, the product will never improve. A lot of factors can impact data discoverability – different naming conventions across teams, unrecorded data transformations, copying data, and so on. My advice is to enforce a set of data standards across the entire organization that lays out a clear process for naming, moving, transforming, and storing data. Nevertheless, it’s essential to accept that data can become disorganized over time, and data governance is a continuous, iterative process. AI tools and models can also be harnessed to enhance data discoverability.
Provide customers with transparency and choice
Privacy and data governance are non-negotiable, but there is a third, perhaps less obviously “table stakes” pillar of customer-centric AI: user choice and transparency about what aspects of your product use AI. Call out where AI is showing up in user experiences throughout the entire product journey and provide users with the choice to opt in or out at every step.
This doesn’t have to be an all-or-nothing decision for your customers. Whenever possible, present customers with options in the form of a sliding scale, or easy ways to opt out if needed. That way, users can feel in control of their own AI usage and dictate their desired experience, and companies don’t risk losing a subset of their users entirely. Of course, the more data you can collect, the more you can optimize a user experience, so it comes down to striking the right balance. If users decide to opt in, they can enjoy the advantages of a fine-tuned model that harnesses the collective data of all participants.
Customer-centric AI is the key to success
As engineers and product builders, we want to build, iterate, and ship as fast as possible to improve product experiences. At the same time, we cannot lose sight of end users who are the heart of the products we deliver. Privacy and governance are paramount, but in order to have a truly customer-centric AI strategy, you need to put the decision-making power in the hands of your customers. As engineering leaders, we should all foster collaborative partnerships with users throughout the development process. Giving customers a voice and a seat at the table will ensure your company is at the helm of the next wave of AI innovation.
