How Real Business Owners Use AI (Finance, Marketing & Customer Service)Panel Discussion

Last month at Roseville Venture Lab, we tried something different.

Instead of another slide deck about “what AI could do,” we put business owners on stage to talk about what AI is doing right now—in the messy middle of running a real company. As I said in the opening, the goal of this series is straightforward: help business owners use AI to stay competitive.

Our panel featured:

• Jason Wang (CaterAI) — voice AI that answers calls and takes orders for restaurants

• Michelle & Jose Luis Martinez (The Ludlow Cleaning Company) — a “non-tech” service business using AI for ops, HR-adjacent work, proposals, and profitability

• David Webb (LowPropTax) — property tax appeals automated end-to-end, driven by data + AI workflows

Below are the biggest takeaways—written for the operator who just wants to know: what’s worth copying this week?

5 Practical Lessons from the AI for Business Panel

1) Start with one idea: “Let AI do what you don’t love.”

Jason’s framing was refreshingly honest: use AI to offload the work you don’t enjoy, so you can focus on the parts of the business only you can do.

Concrete example: If you hate prospecting, have AI do the list-building and prep. If you hate writing emails, have AI draft them—then you edit for tone and accuracy.

My suggestion: Don’t try to “AI your whole business” in a weekend. Pick one weekly pain and standardize it.

2) Service businesses can win big—because AI is a “cheap cofounder”

Michelle dropped a line that got a lot of nods: she uses AI for “probably 75%” of her workday.

Her use cases weren’t futuristic. They were painfully practical:

• Drafting and replying to emails

• Capturing notes during walkthroughs (voice)

• Turning photos + notes into proposals

• Creating SOPs and improving workflows

• Profitability analysis by job and by employee (productivity rates)

Concrete example: She described walking a 22,000 sq ft facility and using AI to help turn that into a polished proposal workflow (including experimenting with turning proposals into a simple hosted webpage).

My suggestion: If you’re in a “sweaty business,” AI is leverage. It doesn’t clean the building—but it does clean up the admin pile.

3) Marketing isn’t “make 100 posts.” It’s “figure out what works for your business.”

One of the more mature moments in the panel was the pushback on social posting-for-the-sake-of-posting.

The takeaway wasn’t “AI makes content easy.” It was: AI makes it easier to think strategically about what marketing channels actually matter for your business type. (Because yes—some businesses won’t win on Instagram, and that’s okay.)

Concrete example: David mentioned using tools that can generate branded social content at scale, and also using AI to analyze analytics data to find what’s working.

My suggestion: Use AI to answer:

• “Where do my customers actually come from?”

• “What should I stop doing?”

• “What’s the one channel I should double down on?”

4) Customer service automation works… if you build guardrails

The best Q&A question of the night was essentially: How do you trust what the AI is saying?

David’s answer was the right one: guardrails (clear rules + known sources + human escalation for edge cases).

Jason backed it up with the real-world version: early mistakes happen, you treat it like training a new employee, and you continuously correct and improve.

Concrete example: Jason shared a restaurant deployment where the voice agent handles dozens of calls per day and drives meaningful phone-order revenue—helping restaurants capture orders during rush periods when humans can’t pick up every call.

My suggestion: Automate first responses and routine questions, but keep:

• billing disputes

• compliance-heavy topics

• anything “high emotion”
  in human hands.

5) The real moat isn’t the model—it’s your process + your data

This was the most “startup” moment of the night.

David’s point: models are getting commoditized, but proprietary workflow + proprietary data stays defensible. He described aggressively gathering case data (including public-record requests) to build an advantage that’s hard to copy quickly.

Concrete example: If you’re a tax appeal company, your moat might be thousands of past cases and outcomes. If you’re a cleaning company, your moat might be job-level productivity benchmarks and repeatable SOPs. If you’re a restaurant, your moat might be menu logic + voice workflows that match how you actually operate.

My suggestion: Don’t obsess over “which model.” Obsess over:

• your repeatable workflow

• the data you already have (or can legally collect)

• the templates your team can reuse every week

The question everyone asked (quietly): “How much does this cost?”

Two useful soundbites from the panel:

• You can start on free plans and still get meaningful results; heavy business usage can justify paid tiers.

• Spend varies wildly by how you use AI: direct API usage can get expensive fast, while normal subscription usage is often enough for most small businesses.

Takeaway

AI doesn’t replace your expertise—it amplifies it.
Start with one workflow you hate, add guardrails, and build a small “library” of prompts + templates your team can repeat weekly.

Actionable recommendation: This week, pick one: (1) proposal creation, (2) customer email replies, or (3) a monthly KPI review—and turn it into a reusable AI workflow with a simple checklist.