A Customer Success Scoring Model That Would Have Taken Months Was Built in 5 Hours With AI

Last week, SmarterX and Marketing AI Institute founder and CEO Paul Roetzer sat down on a hotel patio with his laptop, opened three AI models, and built something in a few hours that previously would have taken 50 to 100 hours of work.

By the time he boarded his flight the next morning, he had a complete customer success scoring model with weighted variables, health tiers, a HubSpot implementation guide, and a strategy brief ready to share with his team.

Total time: three to five hours.

On Episode 197 of The Artificial Intelligence Show, Roetzer walked through exactly how he did it, step by step, prompt by prompt. It's one of the clearest demonstrations of what AI-assisted strategic work looks like in practice.

The Problem: 150 Accounts and No Way to Measure Success

SmarterX launched AI Academy in 2020 as a basic online education platform. In August 2025, the company soft-launched business accounts, an enterprise version where companies buy licenses for their employees to learn AI. A new AI-powered learning management system went live in November 2025.

Four months later, more than 150 business accounts were onboarded. Growth was strong. But there was a gap: no defined scoring model to monitor account health, predict renewals, or flag churn risk.

The AI-generated problem statement Roetzer developed captured it precisely: "AI Academy has successfully onboarded 150 plus business accounts since the official launch in November, 2025, but we do not yet have a defined success score or operating model to measure, manage, and predict enterprise adoption across highly varied AI maturity levels."

The reality is that in any given enterprise account of 100 employees, you're dealing with wildly different AI maturity levels. Maybe 25% are power users who can't live without AI. Another 25% are curious but haven't integrated it into daily workflows. A quarter use it passively without even knowing it. And the last 25% want nothing to do with it.

Without a scoring model, there's no way to manage that complexity or predict which accounts are thriving and which are about to leave.

The Multi-Model Process That Made It Possible

Roetzer's approach wasn't a single chat conversation. It was a deliberate, multi-model workflow where each AI played a distinct role.

Step 1: Problem statement. He used ProblemsGPT, a free custom GPT he built, to turn a rough narrative into a sharp problem definition.

Step 2: Variable generation. He prompted both ChatGPT and Google Gemini to recommend variables for a success score, using custom versions of each: a Co-CEO GPT trained on SmarterX's company history and revenue model, and an AI teaching assistant trained on AI Academy's instructional design principles.

Step 3: Curation. He pulled both sets of recommendations into a Google Doc and started editing. This is where domain expertise matters. Roetzer has built lead scoring systems before, for his own agency and for clients. He knew the general workflow but AI accelerated the execution.

"I immediately was like, ‘Wow, I might actually be able to do this tonight,’" he says.

Step 4: The objective outsider. This is where it got interesting. He went to Claude but intentionally did not give it his existing draft. Instead, he pointed Claude to the AI Academy webpage and said: “Learn about this offering, then I'll tell you what to do next.” He wanted a fresh, unbiased perspective.

“Claude Crushed It”

After Roetzer fed his revised model back into Claude, something unexpected happened. Without being prompted, Claude produced a comprehensive workbook that included a scoring model with weights and criteria, health tiers with recommended actions and outreach cadence, a HubSpot implementation guide, a score calculator for manual testing, and lifecycle weighting considerations based on adoption phases.

“Claude crushed it,” he says. "It was honest-to-God, top-level, senior-level strategist work, as good as anything I've ever gotten from a senior strategist in our company or in an agency."

He then asked Claude to build strategy briefs for each tab of the workbook, spent another hour at the airport editing the output, and sent the whole thing to his team.

The Human Insight AI Couldn't Replace

The team met, reviewed the AI-assisted model, added comments, and worked toward consensus on the variables and weights. The meeting was critical.

"We actually came up with an MVP approach that Claude, ChatGPT, and Gemini hadn't thought of,” Roetzer says. “A faster way to actually get this in use within two weeks."

AI did 90% of the heavy lifting, compressing weeks of solo work into hours. But the final solution, the practical shortcut, the thing that made it deployable faster, came from humans who had context and experience..

What Would Have Happened Without AI

"If I did not have these models, this success score would've taken three more months to do,” he says. “My schedule between now and April is booked solid. I would not have had time to build this. And instead it's built, it will be activated, and in theory, it will be worth millions of dollars to the company over the next couple years."

That's a key value of AI: What becomes possible when the time barrier drops. Projects that would have sat for months get done in a single evening. Strategic work that requires expensive consultants gets drafted by a leader with domain expertise and the right AI workflow.

The Playbook Anyone Can Use

What’s even more valuable is how replicable it is. There's nothing in this process that requires technical skills. The workflow is clear: define the problem with AI, generate options across multiple models, curate with your own expertise, get an objective outside perspective from a model that hasn't seen your draft, then bring humans in to refine and find what AI missed.

The key distinction is between basic AI usage and a true multi-model workflow. This isn't using AI as a search engine or going back and forth a couple times. It's using custom GPTs and gems customized to different use cases.

It's having multiple models play off each other, check each other's work, and give different perspectives. The synthesis of all those inputs creates something exponentially more valuable than any single tool alone.

"This is practical stuff that anyone can do," Roetzer says. "Any leader in a company who has domain expertise can just work with the models to do it better and faster."