83% of Companies Prioritize AI—What About You?

Angelie Janssen 

No problem, below are some examples, we had over 200 different ways people would record a holiday as "Bed & Breakfast" Some of these are fairly obvious and indeed around 50% of those bookings were recorded with fairly obvious descriptions, however the remaining 50% of those bookings fell in to categories that were harder to pragmatically identify, particularly since "Breakfast" is an entirely different booking type.

Self catering is another example of how a free text field can make reporting a challenge without lengthy clean up exercises.

Using the LLM in the model has taken care of spelling mistakes and "Obvious" matches.

With all things AI, we do need to periodically check for errors but these are now the exception, not the rule.
 

Chris Banks the most common use of LLM, for us, is to ask questions and have the system answer them.  For example an employee may ask, in a Discovery, "what are the top five most profitable counties in Florida?"  Pyramid uses the LLM and returns the results (in this case usually in a map).  Then the user may ask additional questions like "With these counties show the products being sold and explain the results." This use of LLM is commonly used by management as they want answers to questions but don't necessarily want to learn how to create a Discovery to get them.  On the developer side they often kick start their dashboard development by starting a presentation and asking the LLM to "Create a dashboard for an underwriting manager, include trend lines, profitability, forecasts and explain the results."  This creates the basis of the dashboard they are looking for and then they can go through tweaking it, rather than having to build from scratch.

The second area we commonly see the LLM being used is when the user want's supplemental data.  Using the example above where they listed the top five counties by profitability, they may want to see what the average number of household members are to see if there is a marketing opportunity there.  To do this they simply right click on County (in the Discovery), choose "AI Driven Value" and enter something like "average number of household members."  The LLM then returns the values and Pyramid adds a value with the average number of household members for each of the five counties. * A nice feature to add would be an explain option to have the LLM explain where it sourced the data from

Lastly, we have also employed AI with Python in our data models.  There is a lot in this area to unpack, like the examples so nicely presented by Darren Tuffs, but there are other areas we use it.  We often use it to jump start our Python code.  Even you are a seasoned Python developer, it's much easier to tell the LLM what you want Python to do and let it write the initial code for you.  It saves a lot of key strokes and often times may come up with a slicker way to code the solution.  We have been pleasantly surprised at how sophisticated the Python code the LLM can return for complex issues.  Then of course if it coded something we don't understand, the "Explain" button is great as we get to see the explanation of what the code is doing.

Chris Banks I hope this helps in the different ways we have been using LLMs.