Credibly’s strategy for incorporating Artificial Intelligence (AI) into its underwriting system provides a blueprint for companies wishing to smoothly adopt transformative technologies. It’s a deliberate, well-thought-out plan, dating back several years.
Chief revenue officer MJ Jiang said she immediately saw AI’s potential to change the customer digital experience back in 2023 with the advent of ChatGPT. She approached the C-suite and urged them to take a proactive stance.
With AI models constantly improving, Jiang said at the outset that they’d be at their worst. That was the time to develop internal competencies so Credibly could be a leader as AI became commonplace. Failing to do that would leave a company at the whim of third-party vendors pushing solutions that are hard to evaluate.
Start with the basics
Jiang began with the basics, working with staff to learn how to work with AI. It’s still so new that nobody’s an expert, so push boundaries and learn from the experience.
“I conducted company-wide training, starting with what LLMs are,” Jiang recalled. “How are they trained? What’s the training data? How do models evolve? Why are they important?”
That training, offered company-wide, removed much of the fear of the unknown. From there, Jiang brought together people from different areas to document use cases. Credibly has roughly 100, which are reviewed and shared across departments.
Those ideas are grouped into three layers of innovation. The first is Distributed Innovation, which runs across departments and can see employees stepping on each other’s toes.
Enterprise Innovation sees code and API development. What concepts are shippable? Benchmarks are higher here.
The final level was a top-down strategy that began with executives taking multi-day training that discussed the future of work in an AI world and how it should look at Credibly. Any successful strategy needs C-suite buy-in, and this helped to achieve it.
Jiang also knew it was important to define the human role alongside every AI strategy. Does a human take the lead with AI serving a complementary role, or is AI in the lead with a human in the loop?
“There’s a level of complexity that I think I find very beautiful,” Jiang said. “I think future leaders working in this space need to understand that they’re managing the complexity of expertise around AI. Learning how to work with that is going to take years.”
There are no experts (yet)
Beware of companies claiming expertise this early in the game. If they haven’t identified a true value proposition, the technology will quickly evolve and wipe them out. For similar reasons, don’t hire a chief AI officer to take the lead. Build a corporate competency instead.
Now is the time to exercise uninhibited imagination, because that’s where the winners will come from. It begins with data extraction, summary and cleaning to prompt evolution, bias removal, and information sorting. That creates consistency, which fosters reliability.
This allows underwriters to create complete business profiles and training systems. That’s especially important as Credibly may only have a few months of banking information to work with. The user experience improves because irrelevant questions are removed from previously boilerplate questionnaires. Now you have AI with guardrails.
“Because it’s part of the training data and these large language models, you can actually get really far with creating a good, robust picture of what this business should look like,” Jiang said. “That’s where the human expertise comes in, when you notice something out of the ordinary. You can now also train these models to surface these discrepancies and to have pretty good confidence; scores saying this business is a legitimate business based on its presence.”
That work frees underwriters to focus on more stimulating tasks. Prepare for that in advance by anticipating the new work and training staff.
Guardrails, regulation, and fraud: Issues to consider
As the company becomes more familiar with AI, it should know where and where not to deploy it, and when to include guardrails. One certainty is that AI will greatly aid in the presentation of consistent, unified information in seconds.
AI also raises legal and ethical issues. Criminals are gaining competency just as fast as companies; Jiang said it is a constant battle to stay ahead. Deep fake quality keeps improving, as does the ability to fake documents. Add them together, and impersonations are harder to spot.
Governments must contend with balancing innovation and privacy. International firms are challenged by a lack of uniformity in regulatory approaches across jurisdictions. Even those operating inside the United States must prepare for differences in state regulations that threaten to stifle regulation.
“How much do we get a federal view, and how much is going to be just a patchwork of regulations that everybody has to figure out?” Jiang asked. “(That’s) very painful and time-consuming and does not create a good customer experience. So I would hope that we could stay away from that.”
Future applications show promise
As she looks to AI’s upcoming applications, Jiang is excited about a few areas. The first is native agents built into LLMs. It’s already started, but just like how we have native apps on our phones, Jiang sees them in LLMs, too.
“There’s no reason why, if I were an LLM, if I were one of the tech labs, I would give that up to third-party developers,” she reasoned.
AI will also change search and marketing as we know it.
“E-commerce, the whole experience of being able to have long, engaging conversations with the product, and then be able to make decisions with it without ever leaving a chat window, and eventually, probably voice,” Jiang concluded. “These are opportunities that we’re going to see very quickly.”