How to Sell Managed AI Employees to Small Businesses (zero code)

This business model is hiding in plain sight.

In the next few minutes, I'll show you how one non-technical operator is installing AI "employees" inside small businesses and getting paid a monthly retainer to manage them, with zero coding background and almost no audience.

I recently interviewed Phil Goodwin, a product designer who's completed about 21 agent setups and turned them into recurring revenue in roughly a month of running the managed model.

I've spent the last year selling AI services to small businesses, and this is one of the cleanest early playbooks I've seen.

In this article, I'm giving you the entire playbook for selling managed AI employees:

• The offer structure (and why most people package agents backwards)
• How to pick the first role that creates the "light-bulb moment"
• The dead-simple fulfillment setup (one Telegram group)
• The weekly value ledger that proves ROI to your client
• The exact pricing evolution from $500 setups to $1,000/month retainers
• The 12-step paid-pilot sequence to land your first client

Almost nobody is talking about the managed layer of this business. That's the entire opportunity.

Let's get into it.

Sell a Role, Not an Agent

The foundation of this entire model is a positioning shift. Without it, nothing else matters.

Most people build an agent, install it, hand the client a login, and hope the business owner figures out the rest.

Bad idea.

A small-business owner doesn't wake up wanting a VPS, an LLM, or another AI tool. They want someone (or something) to own work they're tired of doing.

So Phil positions the offer like hiring a digital employee. One clearly defined role, usually admin or executive-assistant work.

What the AI employee comes with:

• A clear job description
• Company knowledge and memory
• Access to the specific tools the role requires
• A dedicated computer or server environment
• Approval gates for consequential actions
• Monitoring and repairs
• Ongoing workflow coaching for the owner

How it operates:

1. A client request or scheduled job triggers the work
2. The agent uses company knowledge and connected tools
3. A human approves sensitive actions when required
4. The agent completes the work
5. The operator monitors the system and improves the workflow

The agent is only one part of the offer. The actual product is the managed business result.

One Thing to Note Before We Continue

A working agent doesn't automatically create adoption.

Phil learned this the expensive way. His first offer was one-time infrastructure setup, and he climbed from $500 to $1,000, $1,500, and $2,000 per project. He'd configure everything, give the client 14 days of support, and move on.

Some clients stopped using what he built.

The technology worked. The clients didn't. They had no idea what to delegate, how to build workflows, or how to talk to an agent. They treated it like Google: ask a question, get an answer, leave.

That gap is exactly why the management layer exists, and it's a large part of what the client is actually paying for. You're selling adoption much more than you're selling the install.

Picking the First Role (+ a Real Client Example)

Don't promise a company unlimited agents. Most small businesses need one painful category of work taken off the owner's plate.

Good starting roles:

• Central-office administration
• Executive-assistant work
• Research and prospecting
• Recurring reports
• File and document management
• Content support

Here's what this looks like in the wild.

One of Phil's early managed clients runs a family directional-drilling business. The owner bounces between contractor hours, field logs, QuickBooks, PDFs, contracts, spreadsheets, and job folders all day.

In one request, the owner asked the agent to:

1. Search his inbox for an email from a specific person
2. Find an invoice for a particular job
3. Create a new job folder
4. Save the invoice PDF inside it
5. Extract the relevant info into a spreadsheet
6. Save the spreadsheet in Dropbox
7. Return the link

The agent came back with the completed sequence roughly 6 to 10 minutes later.

That's the light-bulb moment you're engineering. Not a flashy demo. A real piece of work the owner already understands and already hates doing.

Once that first workflow lands, the client starts asking the most important question in this entire business:

"What else can it do?"

The Fulfillment Setup (One Telegram Group)

Phil's smartest fulfillment decision costs nothing: he creates a Telegram group with three members. The client, the agent, and himself.

The client watches how Phil communicates with the agent. Phil watches what the client requests, spots unclear instructions, and identifies new workflows worth building.

It's observable coaching instead of forcing the client to learn alone.

The full management layer can include:

• The shared client-agent-operator channel
• A monthly workflow review
• Monitoring and failure alerts
• Repairs and maintenance
• Human approval gates
• A shared spreadsheet or other editable source of truth
• Ongoing refinement of the agent's responsibilities

One rule Phil recommends: make the agent write important outputs into a human-readable system like Excel. Research shouldn't disappear into the agent's private memory. Payments shouldn't live only inside a chat. The client and the agent need a shared source of truth they can both inspect and update.

The Weekly Value Ledger (Proving ROI)

This was my biggest takeaway from the conversation.

Phil gives the agent a value ledger. Every task gets logged with what was completed, how long it would normally take a person, the estimated dollar value of that time, and which tasks still need a baseline from the client.

I'd expand it into a weekly report the agent sends the owner every Friday:

WEEKLY VALUE REPORT — [Client Name] — Week of [Date]

TASKS COMPLETED: [list]
AGENT COMPLETION TIME: [total]
ESTIMATED HUMAN TIME RETURNED: [hours]
HOURLY-VALUE ASSUMPTION: [$X/hr — client confirmed Y/N]
BASELINES STILL NEEDED: [tasks awaiting client's time estimate]
ISSUES RESOLVED THIS WEEK: [list]
NEXT WORKFLOW TO ADD: [proposal]

If the agent doesn't know how long a task normally takes, it asks.

That last part matters. The goal isn't to invent impressive ROI numbers. It's to make the estimate visible, let the client correct it, and improve the ledger over time.

Phil's drilling client's ledger estimated roughly 63 hours returned in the first week, and he separately described an estimated $6,300 week. Those are ledger estimates, not audited savings, which is exactly why you agree on the time baseline and hourly value with the client up front.

Most AI operators report what they built. Clients care about what the system did. A weekly value report turns an invisible retainer into a visible operating record.

Getting Your First Client (No Audience Required)

Phil had around 1,500 followers on X when he started. His early clients came from friends, local relationships, and referrals. Not a content funnel.

The sequence was simple. He built agents for himself first. Then he invited people he already knew to test the service with a fully transparent pitch:

"I'm testing a managed-agent service. I'll install the agent, help you build workflows, and charge you a lower early-client price while I improve the offer."

That's a stronger position than pretending to be an established expert, because the client knows exactly what they're getting:

• A lower early-adopter price
• More attention from the operator
• A chance to shape the service
• Honest expectations about where the offer is today

In return, you get experience, feedback, proof, and potentially a referral. Phil's early friend projects introduced him to paying clients in completely different markets.

Pricing (Match Your Price to Your Proof)

Phil's pricing evolved with the maturity of the offer. Here's the full progression:

Early infrastructure-only setups: $500 to $2,000, one-time.

First managed pilots: No setup fee, roughly $250/month to cover infrastructure while he learned the model in two industries.

Recent managed engagement: Roughly $1,500 setup + $500/month, including a monthly call, the shared Telegram channel, and ongoing support.

Stated target: Roughly $2,000 setup + $1,000/month per agent. That's his target, not pricing he's proven repeatedly yet.

The lesson isn't to copy his numbers. It's to match your price to the proof you currently have:

• Don't absorb the infrastructure and model costs yourself
• Charge for at least some of your time
• Explain why an early client is getting a lower price
• Document the work and results
• Raise prices as fulfillment gets repeatable and value gets provable

You don't need to start at $5,000/month. You also don't need to work for free. You can get paid to learn, as long as the client knows that's exactly what's happening.

Optimizing the System & Pro Tips

1. Run the 12-step paid-pilot sequence

If I were starting from zero:

1. Build an agent for yourself first
2. Pick one deployment environment and learn it well
3. Find one business owner in your existing network
4. Identify one painful, repeatable role
5. Sell a transparent paid pilot
6. Define the agent's permissions and approval gates
7. Create a shared client-agent-operator channel
8. Write important outputs into a shared source of truth
9. Track every task in a weekly value ledger
10. Validate the estimates with the client
11. Turn the first clear win into a testimonial or referral
12. Raise pricing only when proof and fulfillment improve

2. Expand the role, not the headcount

Don't rush to sell agent number two. Deepen the first role until the client is asking for more.

3. Respect the risk

If an agent can access email, financial files, company records, or client data, you're responsible for permissions, security, approval boundaries, monitoring, and recovery. Non-technical doesn't mean unaccountable. Stay inside what you can fulfill safely and bring in technical help when the risk exceeds your competence.

Final Thoughts

The agent build gets you through the door. The management, coaching, reliability, and proof are what make the service worth renewing.

You don't need to know everything before you start, you just need to be one useful step ahead of the client, honest about where you are, and capable of staying with the system after launch.

If you want a clear roadmap for building your company knowledge base + managed agents on top of it, grab my free AI Workforce Roadmap.

The full conversation with Phil Goodwin is on the Build With AI podcast, available on YouTube, Spotify, and Apple.

I hope you found this valuable.

If you did, follow me @CoreyGanim for more real AI implementations for non-technical operators, and repost this article so other operators can see it.

- Corey