AI and the $1.3 million Letter

A few years ago I was working with an employer when he received a contract audit. Not just any contract audit. THE contract audit.

Every purchase order for the last five years from his largest single account was forensically examined and the resulting report he was sent included thousands of contract breaches and requests for retroactive payments per the terms of their signed standing agreement.

In total, the amount of liabilities for alleged breach of contract exceeded $3 million. Immediately the occurrence was escalated and attorneys from both sides began negotiating. After over a month, they reached a stalemate, settling at just over $1.3 million.

I wandered into my boss' office one afternoon to see him scrolling through a massive excel document. He was highlighting cells, taking notes on a pad, and visibly frustrated.

"We're going to have to pay this damn thing." He said, not looking away from his screen.

"How much?" I asked.

"Almost a million and a half."

"Do we owe them that?" I pressed. This was serious money.

"Hell no. Absolutely not." He replied.

"So how are you responding to them?" Thankfully, my relationship with my boss was incredibly strong and built on mutual respect.

"I'm grouping them and responding with why they aren't right by their claims of contract breach."

I decided to pull out all the stops and use my built-up relationship capital in that moment.

"Can I read the letter?"

He swung his head and looked directly at me. He took a moment to what I assume was gauge my intent.

"Yes."

I closed his office door and stepped over closer to the large screen facing outward above his desk. He dragged the letter over to the screen and I took a few moments to read it.

"That's good. Really good." I said.

"Thank you." He replied.

"I've heard you explain how this happened several times, and it makes sense to me, but I don't see it in the letter." I was getting in to deep water at this moment.

"They don't care about that. They want their money." He answered.

"You're right." I agreed. "However, it seems to me like can win this thing and not pay them anything." I suggested.

"What."

"What if, instead of directly responding to their accusations, you turned it back around on them and asked them 'how can I' questions?" He stared at me silently.

An incredible thing about working with someone you respect that is also highly intelligent is the opportunity to communicate in executive short-hand. Rather than dull him with my explanation, by asking a question he could fill in far more quickly on his own, we could advance through a high-level conversation at an uniquely rapid pace. Instead of mouthing on and on about my idea, he took a moment to process what that would look like and I shut up to let him.

"Help me write it."

So for the next 90 minutes, we pored over contract language, where the breakdown in expectation from the account management team was compared to our performance, the communication, and failure rate. It was an exhausting exercise in humility to face over and over again how this could have happened, and reframe it in a way that accepted responsibility without acknowledging sole liability.

In conclusion, later that evening my boss clicked send on an 1,100 word email that embodied much of what Chris Voss and Tahl Raz with the The Black Swan Group cover in their book "Never Split the Difference."

Here's my question:

Could ChatGPT 4 have written that letter?

There's an argument that says "Yes, absolutely." And if you were to take the technical aspects of what the letter actually contained, that would be the winning position. But that isn't what ultimately won and absolved the company of all liabilities.

It was the context.

ChatGPT, Dall-E, MidJourney and all other artificial intelligences are built using empirical data. That is to say, for the developers who write the code, when they're giving the machine learning guidance in how to infer meaning from the information presented, they are rarely presented with two options and pick one. It's nearly always fifty, or one hundred, or a thousand different inferences. Rather than look at the consequences of each decision tree, they are forced to pick an iteration and move forward, rarely, if ever, looking back. Corrections are made if the result is less-than-accurate, but that's how you develop a technology by orders of magnitude.

A more simple way to say this is: it's far easier to correct a mistake from an AI than it is to get it programmed correctly 100% of the time -- the opposite of nearly every program on the planet. That's why it's called machine learning.

When my boss and I were writing the letter, over and over again -- probably dozens of times in an hour and a half -- I simply asked him the question: "is that how they will understand this?" and "what do you think they'll conclude if we say it that way?"

I had nearly no subject matter expertise to provide insight for my boss as to how his decades-long customer group would respond. To assume any of that would've been complete folly. Instead, he could remain laser-focused on delivering the information at hand and I merely provided the role of being his red team.

Artificial intelligence performs at an unbelievably high level at providing answers when given the correct context.

And as with every industry on the planet, the best individuals at providing context for artificial intelligence, customers, executive leadership teams, and everyone else are the highest-performing technicians in their field. So it's no surprise that the most successful users of AI will likely be those same people. Early adopters will come and go with the tide, but those at the top of their field willing to embrace the integration of AI into their workplaces and careers will be providing the best context for the tool to be used as aggressively as possible.

In conclusion, while AI eventually could in theory write that letter, at it's current infant stage it does not possess the ability to gain the necessary context outside high-level executive guidance like what my boss deployed.

That letter did not receive a response of any kind for over 90 days and was never directly replied to, resulting in the absolution of liability for our organization to any damages, and instead became the catalyst for implementing changes on both parties to avoid repeat failures in the future.