The Gap Between the Pitch and the Work
Every few weeks I see another LinkedIn post from someone selling a “$47 AI consulting course” that promises you can charge $10,000 per client by next month. The pitch is always the same: AI is magic, clients are desperate, and you just need to position yourself correctly to unlock unlimited income.
I am an AI consultant. I have built AI systems for my own businesses and implemented them for clients. Here is what nobody in that LinkedIn post tells you: the work is operational, not magical, and the clients who get real results are the ones who understand that before we start.
This is what it actually looks like.
What My Clients Actually Need
The businesses that call me are not looking for a chatbot. They are looking for relief. Specifically, they want to stop doing repetitive work that is eating their time without producing growth.
The most common scenarios I see:
- A solo attorney spending 3 hours per day on intake calls and follow-up emails that should take 30 minutes
- A service business owner manually creating social media content every morning instead of doing billable work
- A small agency that has grown to the point where their manual reporting process takes two full days every month
- An e-commerce business where someone manually copies order data between systems because the integrations were never built
None of these clients say “I need AI.” They say “I need to stop doing this thing that is killing me.” My job is to figure out whether AI is the right solution, build it if it is, and tell them honestly if it is not.
That last part matters more than most consultants admit. I have turned down engagements where the client wanted AI but the actual problem was a broken process that no amount of automation would fix. Building AI on top of a broken workflow produces automated broken outputs, just faster. That is not a result anyone wants to pay for.
What a Real Engagement Looks Like
Before I build anything, I spend time understanding exactly how the business operates right now. I call this the discovery phase, and it usually takes two to three sessions.
I ask questions like:
- Walk me through exactly what you do when a new lead comes in
- What data do you have, where does it live, and who has access to it
- Which tasks do you do every day that feel like copy-paste work
- What would you do with 10 extra hours per week
By the end of discovery, I have a documented workflow map and a ranked list of automation candidates. I rank them by two factors: how much time they waste, and how straightforward they are to automate.
Then comes the build phase. This is where I design and implement the actual systems. Depending on the scope, this takes anywhere from two weeks to two months. I always build in a supervised mode first, meaning the automation runs but a human reviews the output before anything goes live or gets sent to a client.
After the supervised phase, I review performance with the client. We look at accuracy rates, time saved, and any edge cases the system missed. We refine the prompts, adjust the workflows, and then gradually increase autonomy on the tasks where accuracy has been consistently high.
The final phase is handoff and maintenance. I document everything the client needs to know to manage the system day to day. I set up monitoring so they know when something breaks. And I offer monthly retainers for clients who want ongoing optimization because the systems always need tuning.
The Tools I Actually Use
I am not going to give you the generic “ChatGPT and Zapier” answer. Here is what is actually running in my client systems and my own operations:
Claude Code is my primary AI interface. It is what I use to build, debug, and iterate on AI agents. It writes the automation scripts, manages the workflows, and does the heavy reasoning work. For clients who need content generation or analysis, Claude Sonnet handles the output. For strategic synthesis and architecture decisions, I use Claude Opus.
Ollama running on local hardware is where I route simple tasks that do not need Claude-level intelligence. For things like status checks, data classification, and heartbeat monitoring, I use Gemma 4 or Qwen3 running on a Mac Mini connected over a Thunderbolt bridge. The latency is under a millisecond and the inference cost is zero. I tell clients about this because it dramatically reduces their ongoing AI operating costs.
Supabase handles structured data. Lead records, intake forms, campaign data, CRM-adjacent storage. It is Postgres with a clean API and real-time capabilities, and it costs almost nothing at the volumes small businesses need.
Instantly manages automated email outreach when clients have a cold email component. I build personalization layers on top of it using AI-generated variables per prospect.
n8n handles workflow orchestration for clients who need integrations between multiple tools without custom code at every step.
Python scripts do the glue work. Data sync between systems, API calls, scheduled tasks, alert monitoring. Nothing sophisticated, just reliable.
The truth is that for 90% of small business AI use cases, you do not need expensive enterprise tools. You need good prompts, clean data, and reliable infrastructure. Most of that can be built on tools that cost under $300 per month total.
What a Real Timeline Looks Like
I will give you actual numbers from a recent engagement with a service business that wanted to automate their content and lead follow-up:
- Week 1 to 2: Discovery. Two 90-minute sessions, workflow documentation, ranked automation list.
- Week 3 to 5: Build phase. Set up the content pipeline (LinkedIn posts, blog articles), configured the lead intake workflow, connected their CRM.
- Week 6: Supervised testing. Client reviewed every output. We caught and fixed three prompt issues during this phase.
- Week 7 to 8: Handoff. Documentation, monitoring setup, team training (45 minutes total, one Loom video).
- Month 2 onward: Monthly retainer for optimization and monitoring.
That is eight weeks from kickoff to a running system. Not two weeks, which is what most gurus promise. Not six months, which is what enterprise vendors charge for. Eight weeks of focused work with a client who shows up and answers questions quickly.
When clients disappear for weeks between sessions, the timeline doubles. That is not a technology problem. It is a project management reality.
What I Refuse to Promise
This is the section that separates real consulting from the LinkedIn course version.
I do not promise specific revenue outcomes from AI implementation. I can tell you what similar systems have produced in my own business. I cannot tell you what they will produce in yours, because I do not control your sales process, your offer, or your market.
I do not promise “set it and forget it.” Every AI system I have ever built needed ongoing tuning. Prompts drift. Data sources change. New edge cases appear. The clients who expect zero maintenance get surprised. The clients I prepare honestly do not.
I do not promise speed over quality. I can move fast, but speed and rigor in supervised mode are not the same thing. Rushing past the supervised phase to get to “full automation” is how clients end up with AI systems that publish hallucinated claims on their website.
I do not take on clients whose core process is broken and who are not willing to fix it first. AI amplifies whatever exists underneath it. If the underlying workflow is chaotic, the automation will produce chaos at scale.
What the Real Results Look Like
In my own operation, my 15-agent AI system handles tasks that used to take me four to six hours per day:
- Five LinkedIn posts written and scheduled daily
- Two blog articles researched, written, and published daily
- Daily briefings on ad performance, lead pipeline status, and revenue metrics
- Automated email outreach sequences personalized per prospect
- Revenue reports compiled every week without me touching a spreadsheet
The system runs on two Mac Minis and costs roughly $200 to $250 per month in SaaS tools on top of my Claude subscription. It produces what would cost $3,000 to $5,000 per month if I outsourced the same volume of work to human contractors.
For client systems, the results are more specific to their context. The law firm with the intake problem went from a 40-minute average response time on new leads to under 3 minutes for the first automated touchpoint. The content business eliminated 12 hours of weekly manual work. The agency cut their reporting time from two days to two hours.
None of those numbers happen in week one. They happen after the supervised phase, after the refinement, and after the client’s team gets comfortable trusting the output. Realistic timeline to meaningful results: 60 to 90 days from kickoff.
What to Do Next
- Audit your own time first. Before any conversation about AI consulting, spend one week tracking exactly where your hours go. The automation opportunities will be obvious. You are looking for anything you do more than three times per week that follows a predictable pattern.
- Document one workflow end to end. Pick the most painful repetitive task in your business. Write out every step from trigger to completion. If you cannot document it, you cannot automate it. The documentation process often reveals that the problem is simpler (or messier) than you thought.
- Calculate what the problem costs you. Hours per week multiplied by what your time is worth. If the number is over $500 per month, it is worth a conversation about automation. If it is under $200 per month, a VA might cost less than a system build.
- Start with the lowest-risk automation. Do not automate customer-facing communication on your first project. Automate something internal and invisible first. Internal reports, data entry, research summaries. Get comfortable with the supervised mode workflow before anything touches a client or prospect.
- If you want to talk through your specific situation, book a call. I do free 30-minute discovery calls. No pitch, no close. Just a conversation about what you are trying to solve and whether AI is the right answer. If it is not, I will tell you that too.
