Balancing Simplicity and Flexibility in Proposal Tools

Today, every business software startup is an “AI Company.”  Regardless of market segment or technology space, a key challenge is in balancing how you build modern, AI-native business applications with a mix of simplified, purpose-built functions and more complex capabilities that more technically proficient users require with the flexibility to leverage the extensive power of the LLMs powering these tools.

The Toaster Analogy 

Consider this scenario that occurs in kitchens every day: 

You have a toaster with buttons, one of which says “Bread” and another “Bagel.” You know that every morning when you need your breakfast in a hurry, you can load your bagel, press that button, and you end up with a perfectly toasted everything bagel ready for cream cheese to start your day.

In that scenario, the “Bagel” button produces a deterministic outcome. When pressed, it has a predefined temperature and time which will perfectly toast your bagel every morning. This is a great system for a quick, repeatable result that homemakers and novice cooks love, as it consistently produces the same output (a nicely toasted bagel).

But what if you have a more complex breakfast you want to make, and are a trained chef who’s an expert in the science of food and baking? Perhaps you want a soufflé, with numerous steps in the cooking process that can all have variability.  

In that scenario, having a button that says “Souffle’” would be amazing, but to build that into a toaster where it could consistently produce the desired output would be difficult. Souffles are notoriously challenging for even experienced chefs. You need to achieve the right consistency of the egg whites, and properly temper the merging of ingredients. There are numerous variables that affect the process, requiring a level of experience to properly execute a well-crafted soufflé. Factors like humidity and barometric pressure can affect how and when things need to be applied in the process. 

Chefs, experts in the domain of food, have the experience to know when and how much egg white to add to the mixture, how much water to add to the water bath, and when to adjust the oven temperature to produce the desired result. This requires the chef managing the cooking process along the way as it is probabilistic in nature, requiring a different set of controls to manage the inputs for making a soufflé, as it’s a more complex process than simply toasting a bagel. 

As experts, they require the flexibility to control the process with different tools and machines in their kitchen; whereas a novice cook wants a toaster that has a button with predefined inputs to run an identical toasting process every time. If the system has too much flexibility (controls for the inputs) it can make the process too complicated for him.

The Parallels for AI

Similar to the above analogy, AI-based business software powered by major LLMs across the market can do some amazing things these days. They are transforming systems of record into “systems of action” and the AI-based business software that companies are working to build are not “all the same” just because they are in a specific trade space.

For context, Rohirrim provides best-of-breed software that helps our customers expedite growth by automating RFP responses, leveraging our patented organization-specific architecture. Our software is all purpose-built with specific features designed to meet the needs of users requiring different types of functionality. For some of RohanRFP’s features, they are “toasteresque,” where you click a button and a function is executed. Behind the scenes, predefined inputs (prompts) orchestrate the workflow to perform tasks such as summarizing a content section or converting a set of text into a table. Those features are built in a way that produces more deterministic outputs while leveraging the LLM. They are designed to be simple and well-defined on purpose.

However, there are other features that enable the accomplishment of more complex workflows, which are super powerful yet, by design, require some knowledge from the user to guide the system to the desired outcome. For example, our Shred-to-Comply assistant is designed to allow proposal managers to rapidly shred RFPs and build compliance matrices. Shred-to-Comply was designed with a keen focus on accuracy in identifying the key requirements—often buried across dozens of pages—while also considering the realities of dealing with complex documents (such as Federal Government RFPs). 

These types of RFPs can have hundreds of pages that need to be reviewed and shredded. If you want an AI-based system to be effective, anything less than 100% accuracy in the shred to identify requirements is unacceptable to users working against tight deadlines on multi-million-dollar proposals that must be both compelling and complete, yet compliant, to be considered. This is an example of where we worked hard to balance how far the AI would go and “where” our customers would want to put a “human in the loop” to finalize and approve the shred, ensuring a high level of accuracy in the system-created compliance matrix. 

At Rohirrim, our founding team lived through the pain of dealing with complex RFPs, and we recognized the importance of building a feature that drives efficiency while producing effective output. This drove us to build a feature that is not “toasteresque,” yet massively reduces operational overhead for users by integrating AI into the workflow, while not relying entirely on it if complete accuracy is required, as no existing LLM can guarantee 100% accuracy by itself.

What does this mean for those looking for AI-based solutions?

There’s a good lesson here for anyone considering AI-based business software to solve challenges they are facing. Regardless of the functionality you are looking for promised by the power of LLMs, it’s important to look at the nuances of the vendor offerings in the trade space, and compare the “real world” potential of different solutions, while also considering how they fit into your specific workflows. 

Many people think that they want a “toaster” to solve their business needs, but in reality, they need a solution that balances flexibility with user experience to ensure they are getting the proper outcomes from their investment. In many cases, it’s hard to identify what your vendor’s solution will actually provide without test driving it. For those buying an AI-based solution, it’s not all about the quantity of features the vendor offers, but really about the quality and value of the outcome provided.

Brian Shealey

EVP GTM