How to Solve Any Business Problem with AI

Ten years ago I built a consulting firm from nothing. Grew it to a team of eight. We were good enough to get acqui-hired into a larger brand, where I spent the next five years.

Now I'm back out on my own, rebuilding.

I know what needs to happen. Lead generation. Client onboarding. Prospecting systems. Internal operations. I've done all of this before. And I know the pain that comes with it. Scaling has always meant people. Contractors you have to find, onboard, and train. Your budget limits the quality of who you can afford. And even when you find good people, turning what's in your head into something they execute consistently is its own full-time job.

I see the same problem in my client work. I train sales teams for a living, and whether it's a contractor following an SOP or a sales rep following a playbook, the core challenge is identical: consistency.

Everyone talks about AI hallucinations, but nobody talks about the human version. The inconsistency that quietly kills deals, training programs, and operational processes every day.

You build the playbook. You train the team. And then you hope they actually follow it. SOPs you write at midnight knowing that maybe half the steps get executed the way you intended. Reps skip discovery questions. Managers wing their coaching. People get tired, distracted, overconfident.

The biggest risk in any process isn't a bad strategy. It's execution risk. The gap between what the playbook says and what the human actually does.

I've always been wired to fight that with systems. Partly discipline, partly necessity. I have ADHD, so if something isn't a documented process, it doesn't get done. That orientation has shaped how I build everything: how I delegate, how I onboard, how I deliver client work.

Something shifted in Q4 2025. The SOPs I've always written don't just train people anymore.

They run.

They run with best practices embedded, at full effort, every time. They don't skip steps because they're tired or think they know better. The execution risk that's followed me through every business I've built, every sales team I've trained, drops close to zero.

In Q1 2026, working alone, I accomplished more than I would have expected to achieve in an entire year. No engineering team. No technical background.

I built 30+ working automations. Not prototypes, not demos. Systems that run my business every single day.

And the thing that made it possible isn't what most people think.

Goldman Sachs estimates that two-thirds of jobs in the US and Europe involve tasks exposed to AI automation. Most people read that as a threat. I read it as a building permit.

If most of your work involves research, communication, analysis, writing, or moving information between systems, then the question isn't whether AI can help with your problem. It's whether you can describe your problem as a process.

Every system I described above was built using the same five steps.

Step 1: Start with the business problem. Identify the processes that contribute to it. Map the steps, the decisions, the criteria for moving from one stage to the next.

Step 2: Research the best practices. Use deep research to find the proven frameworks. AI out of the box gives average answers. Best practices in gives excellent answers out.

Step 3: Let AI interview you. Give it your process and your research. Then let it ask you every question it needs to build something personalized to your business.

Step 4: Let AI build it. Agents can now plan, execute, and connect to your tools. You bring the vision. They bridge the technical gaps.

Step 5: Iterate. Nothing works perfectly the first time. But the cycle is hours, not weeks.

That's the framework. Here's what it produced.

Here's a sample of what one person built in a single quarter.

I built a lead generation engine. Six automated workflows run every Monday, searching for newly hired CROs, AI transformation leaders, and GTM engineers. They research each person, score them, and write personalized outreach sequences.

By the time I sit down with coffee, there are ready-to-send drafts in my inbox. I had my first sales meeting two weeks after turning the system on, from about 40 contacts.

Separately, I built a 259-contact campaign targeting private equity firms. Seven audience segments, 70 custom emails, tenure-based message variants. The whole system is documented well enough that my team executes it without me.

I built AI voice agents that conduct structured 25-30 minute interviews with real humans. Three separate systems: one for change management diagnostics, one for mapping a client's outreach process, one for content workflow analysis. Each one extracts insights from the transcript, generates a formatted report, logs everything to a tracking sheet, and sends a notification. All within 60 seconds of the call ending. These aren't chatbots. They're doing work I used to do myself or pay specialists to do.

Client onboarding triggers automatically when someone gives a verbal commitment. Contract, communications, invoice contact collection, QuickBooks. One trigger, full sequence.

A task tracker scans my meeting transcripts three times a day, extracts action items, assigns them to the right person, and flags priorities. It even overrides me when I volunteer for tasks I shouldn't be doing. (It knows my habits.)

I built a library of 30+ custom AI skills that power my consulting work. Call scoring using MEDDPICC and SPICED frameworks. Seller performance analysis across eight dimensions. Buyer journey classification. Sales process diagnostics. Proposal generation in my voice. Multi-touch outreach sequences matched to case studies and offers.

I built an ROI model for AI voice agents designed to survive a CFO review. A diagnostic sales deck. A revenue calculator. Tools that do the selling before I ever get on a call.

Every one of these was a goal I had for the year. I expected to get to maybe half of them by December, working weekends to push through. Instead, I built all of them in a quarter. Following the same five steps, every time.

Here's the part that matters: I don't have an engineering background. I have a law degree. Six months ago, I had never built an automation, worked with an API, or configured a webhook.

The framework works. Let me show you why.

This is where most people go wrong. They start with the tool. "I should use AI for something." That's backwards. Start with the problem, then work backwards to the processes that surround it.

Michael Gerber built an entire philosophy around this in The E-Myth Revisited: design the business so it's system-dependent, not founder-dependent. If you can't describe how something works to someone who's never done it before, you can't scale it.

Rachel Woods built a practice around it with what she calls AI playbooking. Her core principle: own the playbook, rent the tech. Document your process so thoroughly that AI can execute it, and you can migrate to better tools without starting over.

The discipline is the same. You need the sequential steps, the if-then decision points, the criteria for moving from one stage to another, and the technology stack where the information lives. That's your SOP. That's the raw material.

Your SOP doesn't need to be perfect. You need the goals, the general steps, and the key decision points. Claude can help you identify gaps and fill them as you move toward implementation. But you need at least an outline before AI can act on it.

If you're not someone who naturally documents everything, tools like Tango, Scribe, and Loom let you record your screen while you walk through a workflow and turn it into a step-by-step SOP automatically. I'm obsessive about process documentation by nature. It's what gives me confidence that the output will be consistent, whether it's me executing or an agent. But the starting point is the same for everyone: get the process out of your head and into a format AI can work with.

This is the step most people skip, and it's the difference between a mediocre build and one that actually works.

Before I built voice interview agents, I ran deep research on ElevenLabs architecture, common failure modes, and best practices for configuring conversational AI. Before I built call scoring systems, I combined my own consulting frameworks with published research on what separates high-performing sales calls from average ones.

I use Gemini Deep Research, ChatGPT, Perplexity, and Claude for this. Multiple tools, multiple passes. I don't read 40-page papers. I throw them into Claude and ask: "What are we learning here? What does the research say about the right way to build this?" If you're not running out of deep research queries every month, you're leaving opportunity on the table.

This is the step nobody talks about. Once you have your SOP and your research, you give both to Claude along with context about your business. Then you tell it to ask you as many questions as it needs to clarify your intent and personalize the solution.

Claude has a built-in feature called Ask User Input that presents structured questions you can click through, including multiple choice, ranking, and open-ended prompts. It turns what would be a messy back-and-forth into a focused interview. But any LLM can do a version of this through text-based Q&A. The key is the same: instead of trying to write the perfect prompt, let the AI surface what it doesn't know. It will find gaps you didn't realize existed.

Steps 1 through 3 are the thinking work. Step 4 is where the work gets done.

Six months ago, we were still in the chatbot era. You could ask AI questions and get drafts and summaries. But you couldn't hand it a plan and say "go execute this." That changed. Coding agents like Claude Code, GPT Codex, and Manus can now write software, build automations, configure integrations, and deploy systems. Browser-based agents can navigate websites, fill forms, pull data, and take actions in any software you can access through a screen. MCP connectors are becoming the standard way agents plug into your existing tools: your CRM, your email, your spreadsheets, your project management platform.

In practice, a single build often uses all three. When I built my voice interview agents, Claude configured the post-call analysis pipeline through API connectors, navigated the ElevenLabs dashboard in a browser to set webhook URLs and test the configuration, and then flagged the pieces that needed me: setting up the Google Drive folder structure and granting the right permissions. Connector, browser, human. Each one handled the part it was best suited for.

Nothing works perfectly the first time. But the iteration cycle is hours, not weeks. Every week, the ecosystem expands. New connectors, new plugins, new capabilities. Skills you create become reusable. Each build compounds into the next one.

The ability to think in processes is quickly becoming the dividing line between people who use AI and people who build with it. And that's a skill, not a talent.

I almost called this article "How to Do Almost Anything with AI." Here's why.

My wife and I are building a grocery and meal planning system. Same five steps.

The problem: meal planning and ordering eat hours every week. The process: we mapped months of ordering patterns, recurring staples, dietary preferences, and the approval workflow we actually follow.

The research: how our delivery platform handles automation, what can be triggered by API versus what needs a browser. The AI interview: Claude asked us dozens of questions about preferences and edge cases we hadn't thought through.

The build: an automation that pre-populates our cart, schedules meals, tracks recurring items, and texts us both on Thursday night with the proposed order. If we approve, it goes into the cart for delivery.

The framework doesn't care whether the problem is business or personal. The Wall Street Journal recently reported on this exact trend: people using AI agents to compare insurance plans, order groceries, build running coaches, and track household chores.

But here's the finding that stuck with me. Researchers from UCLA, Stanford, and USC studied what people actually do with time AI gives them back. Most of them just game, scroll social media, and stream video.

That's the gap this framework closes. Without a process for applying AI, you get marginal convenience. With one, you get leverage.

A single person, with curiosity, persistence, and the ability to think in processes, can achieve dramatically more than they could have imagined six months ago. I'm living it.

What's the process in your business, or your life, that eats your time? That you've done a hundred times? That you could describe in steps?

That's your starting point.

In Part 2, I'll share the actual prompts, deep research workflows, and build walkthroughs behind several of these systems.