
They say you can’t teach an old dog new tricks, so in the world of software, an entirely new dog was introduced.
Traditional software is classic plug-and-chug, where every input results in the same output. Acting as a reliable and predictable mechanism, it is classified as deterministic. No surprises here. Straightforward it may be, and straightforward are the code and rules. In a series of if-else statements, each reliant on the one before it, the workflow must be soundproof. The software won’t give you a surprise, but in turn, it can’t handle one either.
Take a CRM for example. When a user submits a form in a CRM system, traditional software validates it against pre-defined logic. But what if the CRM isn’t told that “Chief Marketing Officer” means “CMO”? What if your sales team suddenly wants to prioritize leads coming from a new social platform? Traditional CRM isn’t cut-out to “learn” – it needs to be taught explicitly with structured data. Meanwhile, high-value leads slip through the cracks, and the system is lodged in the past.
This is where Language Learning Models (LLMs) and modern AI come into play. Since the dawn of ChatGPT circa Nov 2022, the software paradigm shifted into a non-deterministic domain – flexible, probabilistic, and for some, even intimidating. As its name suggests, LLMs don’t glean from structured, hard-coded inputs; instead, they observe patterns and predict. Extrapolating from massive pools of data, LLMs infer context to generate responses based on statistical patterns – not rules.
Here converges accuracy and approximation – what are the tradeoffs between the two, and once we know them, what do we do with it?
Occam’s Razor states the simplest choice is the best answer. But simplicity does not signal certainty or completeness.
From relationships to career choices, rarely ever is an output or decision absolutely foolproof – such is the nature with generative AI. So why is it that we value limited, accurate outputs over expanded, approximate ones? Why fix the last 10 percent when the first 90 gives you the signal needed to act?
Your CRM is built to answer clear-cut questions:
Who owns the account?
What’s the deal size?
But relationships rarely fit into these neat fields. The signals of a relationship often live outside those lines – through call recaps, scattered emails, and Slack chains. Here, LLMs add value, not only processing, but interpreting. Instead of relying on defined inputs, they’re able to glean patterns from unstructured, human elements, such as shifts in tone or phrasing. This is not about disregarding structured data; it’s about giving it context and a story.
To relegate business to a traditional CRM is to relegate your client relationships — so why not put a little faith in an LLM? Sentiment cannot be answered solely on the basis of numbers and names, which is why some degree of prediction – whether from a user or LLM – is necessary. Humans estimate just as much as an LLM, gathering cues, forming inferences, and trusting our judgement. Our own approximations don’t deem us reliable or untrustworthy, but somewhere in the messy spectrum. That being said, in the way we never blindly trust others, we shouldn’t blindly trust an LLM – but that’s not a reason to cross it off the list.
Quitting the chase for “perfect” is easier said than done. But to give balm to all the perfectionists in the world, “perfect” can’t always be validated. There’s no fortune cookie to crack or a Magic Eight ball to shake. Perfection isn’t concrete – after all, perfection is the enemy of good. It’s sometimes a round-about journey, maybe even an exercise in futility.
This is what makes Kaboom possible. Instead of prolonging this chase, Kaboom leans into approximation as a feature, not a flaw. Turning unstructured data from emails and call notes into relationship intelligence, action now lies at the end of the path – not perfection.
You don’t need a dashboard of numbers and signs. You need a system that can do the guesswork for you, even when nothing is explicitly said.
So stop waiting for the perfect data. Act on the signals you already have.