James Taylor, CEO of Decision Management Solutions and a leading expert in helping businesses adopt digital decision-making.
Recent months have seen a dramatic increase in enthusiasm around large language models (LLMs) and generative AI. Although this is not new, recent performance improvements and the immense increase in model size and complexity mean that they can handle more complex problems. The ability of these models to summarize complex documents, bring together multiple threads, and generate readable text, of any length and in any style, is quite remarkable. Every company is rightly studying how to get the most out of it.
Many companies are particularly interested in allowing their clients to interact with an LLM who “represents” the company. Such high-powered chatbots would make decisions on behalf of the business, giving customers answers 24/7. They would relieve call centers overloaded with calls, allowing human staff to focus on higher value-added activities. This way, they could reduce costs while improving customer service and satisfaction.
However, this powerful and compelling use case is at risk because of the way LLMs make their decisions. Trained to predict what the next word, sentence or paragraph might be, LLMs struggle to provide transparency about how they make their decisions. They may give answers that do not comply with published regulations or company policies, potentially creating legal exposure for the company. They have the potential tohallucinate“, supporting incorrect answers with seemingly valid references and explanations, which could lead to significant problems for the company promoting their use.
How, then, can you benefit from an LLM without these risks? One option is to combine it with digital decision systems. Based on my years of experience consulting with companies building digital decisioning systems, here’s what I’ve learned about how these two technologies can complement each other.
Understanding digital decision-making
Largest organizations around the world have been grappling with the need to automate business decision-making for years. When transactions or customer interactions require rapid response times or the ability to handle very high volumes, businesses automate. When a need for automation is accompanied by decision complexity, especially when it is unclear what the best answer should be, traditional systems development methods may be insufficient. That’s why I recommend decision management, or digital decisioning, models when the complexity of decision making, such as ensuring regulatory compliance, enforcement of multiple corporate policies, or in-depth knowledge of best practices , is required.
Digital decision systems utilize expert software knowledge through the use of decision models, business rules, and decision tables. These business-oriented representations have the advantage of allowing domain and subject experts within a company to directly specify, update and manage decision-making. This generally eliminates the need to create separate IT requirements documents for implementation by another team. Decision platforms and business rules management systems can help ensure that this logic can be executed at scale. This combination provides transparency because the design is generally clear, even to non-technical people, and because the execution is recorded.
Advanced analytics complements these expert-based approaches. Predictive analytics and machine learning approaches extract business insights from historical data. The resulting analytical models can be deployed to replace rules of thumb. They can be combined with expert guidance and business understanding to improve decision accuracy.
However, these systems are generally inaccessible to customers. They are most often deployed as application programming interfaces (APIs) that can be accessed by enterprise applications and back-office systems. Powerful loan application processing, claims processing, fraud identification and more are possible, but these systems are rarely accessible to customers or the company representatives they deal with.
Combining digital decision-making with LLMs
With the advent of LLMs and generative AI, clients are no longer willing to settle for middlemen. They want to be able to carry out their own transactions, even the most complex ones. Yet, as mentioned above, LLMs cannot yet be fully trusted to make repeatable, transparent, and compliant decisions. However, when LLMs are effectively combined with expert-based decision making, it can facilitate an interactive AI-driven experience for customers, combined with the transparency, repeatability and compliance of digital decisioning.
To see how this might work in practice, think about complaints management. Suppose an insurer must apply federal and state mandates, company policies and individual plan designs to see if a claim is covered under a policy and calculate the amount that will be paid. It should use historical data to determine whether there are patterns of fraud or waste that should be reported. It needs to do this at scale for millions of claims. Decision making can be designed using a decision model so that it matches the way the business operates. This model can then orchestrate the business rules and analytics needed for decision-making. The system can record how this decision was made. It is consistent, transparent, precise and reliable.
However, it is only accessible by submitting a complaint to the complaints system. A patient eventually gets an independent explanation of the benefits, but only after a claim has been submitted. Providers must also submit claims and schedule payment based on their understanding of the contracts and terms they have signed.
Add an LLM, however, and everything changes. An LLM can power a provider or patient chatbot. It can find relevant documents and summarize them, helping people understand the basis of the decision. It can interact with members and staff of the provider to identify the patient and collect the information required by the digital decision system. Once he has the information he needs or gets it from an internal system, he can call the digital decision system to get the official answer: What would the insurer say if this claim was submitted, to at that moment, by this patient, with this medical file? history? The result will be the same as if you submitted it.
Additionally, because the LLM manages the conversation, it can take the technical explanation or log provided by the decision-making system and transform it into something the patient or member understands. The member benefits from easy-to-use, chatty, and human-centered interaction. And a precise, transparent, compliant and reliable decision.
LLMs are a powerful, game-changing technology, but they can complement, rather than replace, expert-based decision-making solutions.
Forbes Business Advice is the leading growth and networking organization for business owners and leaders. Am I eligible?