In this week’s episode of the Niche Pursuits podcast, Corey Ganim and I discuss how to think about AI automation in a way that actually leads to ROI. This topic has a particular focus on AI agents, reusable “skills”, and why tools like Claude Cowork are changing what automation looks like day to day.
Corey runs Return My Time, and he comes at this topic from a builder’s perspective, not a buzzword perspective. He also knows our audience is already past the “what is ChatGPT?” stage, so we spent the bulk of the conversation in the weeds on frameworks, execution, and practical examples.
Watch the Full Episode
Why Most AI Experiments Never Turn Into ROI
Corey started by calling out something most of us have felt: there are too many tools, too much noise, and too many “cool demos” that never translate into better business outcomes. His fix is to stop judging AI by how impressive it looks and start judging it by what it improves in your business.
That leads into his “three levers of ROI” framework, which is a simple filter you can apply before you spend a minute setting anything up. If a tool, agent, or workflow does not pull at least one of these levers, Corey’s view is that it is not worth your time.
- Effectiveness: Does this help you make more money?
- Efficiency: Does this save meaningful time?
- Quality: Does this improve the product or customer experience?
A Concrete Example of “Effectiveness” ROI
A lot of business owners start with efficiency, because saving time is the most obvious win. Corey agrees that basic chatbot use can help, but he thinks the bigger upside shows up when AI helps you capture revenue you would otherwise miss.
He shared an example of an AI “speed to lead” agent that responds to inbound leads quickly with a personalized response. In the interview, he referenced a stat that you are “21 times more likely” to land a customer if you respond within 60 seconds versus an hour or two. This is exactly the kind of leverage that can impact revenue, not just workload.
- Local service businesses are especially sensitive to response time because shoppers often contact multiple providers.
- “Speed to lead” workflows can be implemented without replacing your full sales process; they just remove the delay.
- If you already buy leads or run paid ads, a faster response improves the economics of what you are already spending.
Where to Start: Audit, Optimize, Automate
After you decide what outcomes matter, the next problem is figuring out what to automate first. This is where Corey introduced his AOA framework: audit, optimize, automate, in that order.
His point was that most people jump straight to automation because they are frustrated with a task, but that can lock in a bloated process. If you automate a 12-step workflow that could be reduced to 6 steps, you have not really improved the system; you have just buried inefficiency under automation.
- Audit: Identify what you are actually spending time on (calendar + to-do list is a good starting source of truth).
- Optimize: Reduce steps, remove waste, simplify handoffs, clarify what “done” looks like.
- Automate: Only after the process is cleaner should you hand it to an agent or automation tool.
Using AI to Optimize Before You Automate
One of the more useful parts of this conversation was the idea that AI can help with the optimization step, not just the automation step. Many of us assume we need to be the expert who redesigns a workflow, but Corey’s view is that AI is often more objective than we are because it has no attachment to “how we’ve always done it”.
He described a practical way to do it: brain dump your current process in plain language, let the AI propose a slimmer version, and then take that optimized version forward into automation. Corey used his own weekly Costco grocery ordering as an example, where optimization cut out steps before he handed the workflow to an agent.
- A small reduction in steps can matter more than it sounds, because recurring tasks compound over weeks.
- Voice-to-text can make “process capture” faster than typing when you are documenting a workflow.
- Optimization is also where you clarify constraints, like what inputs are required and what the output format should be.
Claude Cowork: What It Is and Why It Feels Different
Claude Cowork represents a shift from “AI answers questions” to “AI does work”. Corey framed it in a memorable way: a normal chatbot is reactive, while an agent is more like a chatbot with hands that can interact with your tools.
He explained that Cowork can connect to tools like email, Google Drive, CRMs, and task managers, then perform actions inside them. The interface still looks like a chat, but the capabilities change because it can execute real tasks and run scheduled workflows.
- Tool connections are where a lot of value appears, because context no longer lives only inside the chat window.
- Scheduling matters because it turns AI from “something you remember to use” into “something that shows up on time”.
- Execution plans are useful because they force structure before an agent starts clicking around in your systems.
A Practical Cowork Use Case: Better Sales Calls With Less Prep
One example Corey gave was for a friend running a real estate coaching business with a high-ticket offer (he mentioned a $13K ticket size). That friend was taking 5-7 sales calls per day and spending 10-15 minutes before each call scrolling through CRM history to remember context.
Corey’s recommendation was to connect Cowork to the CRM and have it automatically send a summary a few minutes before each call, including context from prior conversations and conversation starters. The outcome is that instead of preparing 10-15 minutes ahead, you can prepare in seconds with a glance at a briefing message.
- This kind of workflow scales well to a team because it standardizes preparation across multiple reps.
- The improvement is not only time savings, but it can also reduce “cold start” awkwardness at the beginning of calls.
- Summaries can include key constraints like budget, timeline, and objections that you want top-of-mind.
Agents vs Skills: The Missing Link for People Who Feel Stuck
A big theme of the episode was why many savvy AI users still feel like they’re not getting the promised time savings. You can be good at prompting and still get trapped in a loop: paste an email, rewrite it, shorten it, change tone, paste again, and eventually it feels like you could have written it yourself.
Corey answers that skills are what make AI reusable and consistent. In his framing, the agent is the doer, while the skill is the recipe the agent follows for a specific task, step by step, in the same way every time.
- Skills reduce “prompt drift” because you are not reinventing the instructions on every run.
- Skills can be shared across tools because the concept is spreading across the AI ecosystem.
- The skill approach is especially useful for any work you do weekly or daily, because repetition is where the ROI shows up.
The Foundational Skill Corey Recommends Everyone Build First
If Corey could give only one next step, it was this: build your brand voice skill. He argued that it is the foundation beneath many other skills, because it makes outputs consistently sound like you without repeated back-and-forth editing.
He also said it’s not complicated to set up, and that he built his in about 20 minutes by giving examples of his writing, plus transcripts that show how he speaks. From there, he described using Claude’s skill-building capability to create and install the skill, so it becomes the default “voice layer” across future work.
- Inputs that help a brand voice skill: writing samples, email examples, podcast transcripts, and a short list of “words I avoid”.
- A brand voice skill is not limited to marketing. It also improves internal communication, like team updates and SOP drafts.
- Once you have this skill, you can pair it with other skills so the entire system stays consistent.
High-Impact Skills Most Business Owners Can Use Immediately
After we dug into the concept, I asked Corey for examples of skills that are widely applicable. He mentioned brand voice first, then pointed to common places where business owners spend time and attention: email, daily planning, and sales-call context.
What stood out to me is that none of these are “flashy automations”. They are the kind of recurring, slightly annoying tasks that quietly eat hours each week, which makes them perfect targets for skill-based automation.
- Email drafting skill: A default format and tone so routine messages come out usable immediately.
- Daily brief skill: A scheduled morning summary of calendar items and key messages to handle.
- Sales call prep skill: A pre-call “battle card” pulled from CRM history and external context like a LinkedIn summary.
Quality Assurance: How to Trust Outputs Without Babysitting Them
One pushback people have is, “How do I make sure this doesn’t produce garbage?” Corey’s approach was to treat skills like software: version them, test them, and refine until you trust them.
He said V1 is rarely perfect, but it is often 80% of the way there, and then V2, V3, or V4 is where it becomes dependable. The key is to bake QA into the skill creation process by running it, reviewing the output, and adjusting the instructions until the output format and quality match what you need.
- Examples of QA tweaks include changing output format (Word doc vs Excel), tightening structure, or adding required sections.
- Testing is fastest when you use realistic inputs, not idealized examples.
- Once a skill is stable, you typically only update it when requirements change.
What “At Scale” Can Actually Look Like in a Week
Corey shared a personal metric that helps make this real. He described using Todoist to identify recurring tasks, then challenging himself to turn those tasks into skills so he would not have to do them manually again.
After about two and a half months of doing this, he said he had equipped his AI agent with around 42 skills. The important part was not just the number, but the idea that each new skill represents a task he can now offload fully, rather than partially.
- This is a compounding system, as each new skill reduces future workload and increases future capacity.
- Recurring tasks are the sweet spot because they repay the setup time over and over.
- A “skills inventory” can become a playbook you can hand to future hires or contractors.
Content Automation That Still Performs: Corey’s X Article Example
Corey also gave a performance example that caught my attention, because it connects automation to real outcomes. He described building an “X article writer” skill that generates an 800-1200-word article in his voice, plus multiple headline and thumbnail options, turning a two-hour process into about 10-15 minutes.
Even more interesting, he said an article he posted the day before the interview was fully AI-generated with no edits and reached 3.6 million views about 36 hours after posting. That is a strong reminder that “automated” doesn’t automatically mean “low quality”, especially when skills and voice constraints are handled well.
- If a platform is rewarding a format, like long-form articles, automation can help you publish consistently enough to benefit.
- Multiple headline variations matter because titles are often the highest leverage variable in content performance.
- The best use cases usually combine research, structure, and brand voice, rather than relying on raw generation alone.
Corey’s Offer and an ROI Snapshot
Corey briefly outlined what ReturnMyTime.com does today, including an “AI tools assessment” where he interviews a business owner, identifies bottlenecks, and prescribes off-the-shelf tools that can save time. He described a target of buying back 5-10 hours per week, with a refund guarantee if they cannot find at least five hours per week in savings.
He also shared an average cost snapshot: about $42 per month in tools paired with roughly six hours saved per week, which he framed as a simple “what is your time worth?” calculation. Whether someone buys help or does it solo, the takeaway is that small monthly tool costs can be trivial compared to the value of reclaimed hours.
- He pointed listeners to a short “AI literacy” quiz at returnmytime.com/quiz that recommends resources based on your level.
- The assessment model is interesting because it prioritizes off-the-shelf adoption, not custom engineering from day one.
- Time-saving estimates become more believable when they are tied to specific workflows, not vague “AI will help”.
Final Thoughts
The big win from this episode is the shift from “using AI” to building an AI operating system for your business. Corey’s frameworks keep you focused on outcomes first, then guide you through a sensible build order so you do not automate chaos.
If you’re overwhelmed, the simplest move is the one Corey kept repeating: build the brand voice skill first, then start converting your recurring tasks into skills one by one. When you approach AI this way, you are not chasing tools; you are steadily removing work from your plate and replacing it with reusable systems.
Links & Resources
