This post was originally published on LinkedIn.

2010-2024: Everything was rules-based and predictable:

• Account does X → hit this threshold → send to sales
• Contact does X → Score reaches Y → hand to sales
• Personas we like are X → Pull list of Job title = "X"
• Company size > 5000 → route to enterprise team

We built these rigid, if/then systems because we needed consistency and scale. And it worked.

BUT:

Now with AI, we're all experimenting with non-deterministic approaches to these same processes.

> Instead of "enrich contacts with VP or Director titles", it's "AI, go find the contacts you think are best for this account."

> Instead of "prioritize accounts over $50M revenue", it's "AI, tell me which accounts we should go after this quarter."

It's cool to see this play out, but it opens up things we've never had to worry about before.

LLMs can get things wrong because they're very smart but they are making it up in real time. Maybe leaning in directions you didn't expect.

You try to guide it with prompt engineering, breaking prompts into discrete steps, adding guardrails, context. But it's still not always going to make the decision an if/then statement would. But maybe that's a good thing?

HERE'S WHAT'S REALLY INTERESTING:

Over the next couple years, we're going to learn three things:

1. What NEEDS to stay deterministic (probably compliance, data security, certain handoffs)
2. What we're OK letting AI figure out (maybe contact selection, timing optimization, ICP, personas, segments)
3. What's actually BETTER when AI figures it out vs. pre-set rules

That last one is where it gets exciting.

AI might see patterns humans miss.

We're seeing this with some of our clients already. AI is surfacing accounts that don't fit the traditional profile. The trick is being able to report on if these convert better over the long run (time will tell).

I don't have the answer yet on where this settles.

But watching this deterministic vs non-deterministic shift play out is very interesting.

Anyone else wrestling with this? What are you comfortable letting AI decide vs keeping locked down in rules?