Automate the task that repeats on a trigger or a schedule, has a clear input and a clear output, and costs you money when someone forgets it. For most small businesses that task is lead intake. Publishing comes second and reporting comes third.
That is my short answer. The rest of this post shows what we actually automate for clients, using only work that is on our live case studies, and then gives you the five checks I run before I say yes to a first project. If you want the wiring behind all of it, I wrote that up separately in how we actually run automation at Tuscan. This post covers what to automate. That one covers how it runs and how it breaks.
What we actually automate for small business clients
Our AI and business automation work in Indianapolis falls into five buckets: lead intake, publishing, reporting, follow-up and data clean-up. Here is what that looks like on three real client builds.
Lead intake: the Roots Realty CRM
Roots Realty is an Indianapolis real estate team. Their leads arrive from several places. Website forms, showing requests, Calendly bookings, BiggerPockets inquiries and Eventbrite registrations all land in the Roots CRM as leads. We built that CRM at admin.rootsrealty.co, and agents work their contacts and deals from one place, on desktop or phone.
Without one list, every source is its own inbox. Someone has to remember to check five places, and a lead that lands in the fifth place on a Friday can sit until Monday. One list removes the remembering. You can read the full build on the Roots Realty case study.
Data from another system: the MLS sync
Listings on the Roots site come straight from the MLS. MIBOR data syncs through MLS Grid every three hours, and every active home in the search gets its own page. The 27 neighborhood guides carry monthly market numbers computed from the same MLS data, and those numbers refresh on the 8th of each month.
The sync runs on a schedule, so nobody has to copy a price into a page. The data already lives in one system, so we moved it to the other. If your business has numbers that sit in one tool and get retyped into another, this is the same job. Real estate teams with this problem can start with our page on real estate websites.
Publishing: one post, every place it belongs
Publishing comes next, and three of our clients run it.
- Illinois Street Food Emporium. Every newsletter and blog post the restaurant publishes has also posted to its Google Business Profile since June 22, 2026. One publish gives them a page on their own domain and a fresh update on the profile, with no extra work from the team. Their Instagram also feeds the site. See the ISFE case study, or our page on restaurant websites.
- Roots Realty. When a newsletter goes out, a matching post goes to the Google Business Profile. The blog, a newsletter archive covering three publications, the podcast and events all publish as pages on rootsrealty.co. The archive held 121 published newsletter issues as of September 1, 2026.
- Get Indiana. When a new podcast episode goes up, it lands in the CMS as a draft. Someone on the team reviews it and hits publish, and the episode page is live on their domain. As of September 2026 the CMS held 371 published episodes and 459 newsletter issues.
Nate Spangle, Get Indiana's founder, describes it this way: "We publish an episode or newsletter and it's on the site and updating our Google Business Profile instantly." Google explains what those profile updates are in its own help page on creating and managing posts on your Business Profile.
Finding what to cover: the trending-news scraper
Get Indiana also has a scraper we built that flags trending Indiana news. Nate's words: "They built us a scraper that flags trending Indiana news so we know what to cover next." Reading a lot of sources and sorting them is slow for a person and fast for software, which makes it a good candidate for this kind of automation. You can see the whole system on the Get Indiana case study, where the stack includes the Claude API.
Where AI fits and where it stays out
Most of the examples above need no AI at all. Moving a lead from a form to a CRM is plain automation. Syncing MLS data is plain automation. AI belongs on the reading and writing steps: sorting what came in, summarizing it, drafting a first version.
I use Claude every day on these builds, mostly through Claude Code, which writes and changes the code, runs the tests and checks the live page after each deploy. The same tools do the reading and writing steps inside the automations we ship.
We hold one rule on every build. Nothing AI writes goes out without a person reviewing it. The Get Indiana episode flow shows it. The episode arrives as a draft and a human publishes it. If you want a feel for what the model side can do, the Claude API documentation is the primary source for what we build on.
How to pick the first thing to automate
Here are the five checks I run. A task that passes all five is a good first project.
- It repeats. It happens every week, every day or every time something specific occurs. A one-off task is cheaper to do by hand.
- It has a clear input and a clear output. A form comes in and a CRM record goes out. A post is published and a Google Business Profile update appears. If you cannot describe both ends in one sentence, the task is not ready.
- Forgetting it costs something. A lead that waits three days is a lost sale. A profile that goes quiet loses a little visibility every week. Pick the task where a missed step has a price.
- A person can check the result. You should be able to look at the output and tell in ten seconds whether it is right. That is how you trust it, and it is where AI review steps go.
- The data already exists in a tool. Automation moves information between systems. If the information lives in someone's head or on a sticky note, the first job is writing it down.
Here is how the common candidates score on those checks, in my opinion.
| Task | Repeats | Forgetting costs money | Easy to check | My take |
|---|---|---|---|---|
| Lead intake and routing | Every lead | Yes, directly | Yes, the lead is in the CRM or it is not | Start here |
| Publishing to site and Google Business Profile | Every post | Slowly, through visibility | Yes, look at the live page | Start here if leads already work |
| Weekly reporting | Weekly | Rarely directly | Yes, compare to the source | Good second project |
| Follow-up reminders and confirmations | Every deal or booking | Yes | Yes | Good once intake is clean |
| Cleaning up old spreadsheets and PDFs | Once | Rarely | Mixed | Do it when another project needs the data |
What I would leave alone at first
Do not start with the biggest, messiest process in the company. It has too many exceptions, and a first project that drags on teaches nobody anything. Do not start with a task only one person understands. And do not start with a customer-facing message that goes out with no review. Our automation page says the same thing: pick the one task your team does every week and hates, map it, automate it, and measure the hours it gives back. If it works, the next one is easier because the pieces are already connected.
One more caution, and it is the reason a partner matters after launch. Automations break quietly when an app changes. Ours run with failure alerts, so a broken step tells someone the same day. The full list of checks is in our automation breakdown. Before you buy another tool for a problem like this, read the tool sprawl post, because the answer is often connecting what you already pay for.
How we do it at Tuscan
Every client site we build runs on the same stack, so each automation has one place to plug in. Sites are Next.js on Vercel, with Supabase when a site needs a database or logins. Content lives in our own CMS at cms.tuscanagency.com, which we built. When we publish there, the live page updates within seconds and the new URL goes to Bing and the other IndexNow engines automatically.
The automations run on Make.com and scheduled GitHub Actions jobs. They land podcast episodes and YouTube videos on the site as pages, archive Beehiiv newsletters, sync Instagram feeds and post new blog posts and newsletters to a client's Google Business Profile. Email goes through Resend. Nightly health checks watch every site and job, and a failure posts to our Slack before a client notices it.
Every change runs automated tests before it goes live, and I check the real live page after. A passing build does not prove the page works, so I look at the page.
Content gets two AI passes and a person. A writer agent drafts from real material: the client's own words, transcripts and data. A separate reviewer agent that never saw the writer's reasoning then tries to fail the draft. It traces every number to a source and opens every link. A person approves before anything publishes. The same approach turned sermons into searchable text for The District Church, where 196 sermon transcripts were on the site as of November 16, 2026.
AI does the repetitive work. Judgment, client relationships and final calls stay with people, and we do not publish AI output nobody checked.
What a first project looks like with us
You tell us the task. We map it with you, build it on tools you own, add the failure alerts and a review step wherever AI is involved, and hand it over with written scope and a fixed price. You own every account and workflow. After launch you can keep us on to refine it or run it yourself. Our about page has the rest of how we work. If you already know which task you want gone, start a project and tell us what it is.
FAQ
What should a small business automate first?
Start with lead intake. Every inquiry should land in one place, reach the right person and alert you if something fails. It repeats, forgetting it costs sales, and you can check it in seconds.
Do I need AI to automate a business process?
Often you do not. Moving data between tools is plain automation. AI helps with reading and writing steps such as sorting, summarizing and drafting. We keep a person in the review step for anything AI writes.
Can I keep the tools I already use?
Usually yes. We connect your CRM, email, Google Business Profile and spreadsheets and only suggest a change when a tool costs more than it saves.
How do I know an automation is still working?
It needs failure alerts and a log of what it did. A quiet failure is the real risk, so every automation we build tells someone the same day it breaks.
What does an automation project cost?
Our automation page says projects are fixed price and based on scope, with a written proposal before you commit. A single workflow is a small project and a connected system across site, CRM and reporting is larger.
How do you use AI to build automations?
I build with Claude Code, which writes the code, runs the tests and checks the live page after each deploy. For content, a writer agent drafts and a separate reviewer agent tries to fail the draft before a person approves it.
Do you work with Indianapolis businesses?
Yes. Tuscan is based in Indianapolis, and our clients include Roots Realty, Get Indiana and the Illinois Street Food Emporium.