What should a small business automate first?

Run each candidate through five questions:

1. Does it happen at least once a week?
2. Does it follow the same steps most of the time?
3. Does the information already live in a system, like a form, inbox, CRM or spreadsheet?
4. Is a wrong draft cheap, because a person checks it before it goes anywhere?
5. Can you tell whether it worked?

The more yeses, the better the candidate. If a process fails the second question, fix the process before you automate it. Our first rule is to remove the friction before adding automation, because automating a messy process just gets you the same mess, faster.

Leads show up through web forms, email, phone calls and social messages. Somebody has to notice each one, log it, reply and follow up. It's repetitive and time-sensitive, and on a busy day it's easy to drop.

A typical first build captures every new lead in your CRM or a spreadsheet, uses AI to draft a reply and route it to the right person, and flags any lead nobody has followed up on. A person approves the reply before it goes out. It's the kind of workflow our Zapier, Make and n8n sprints are built for.

Do the boring part first. KC Enterprise Auto Sales takes bids through a web form, and when an email notification failed, staff could lose sight of a submitted bid. Our first fix separated storage from notification. Every bid now lands in Airtable even if the email doesn't, and staff check a protected Quick Bid inbox that reads straight from Airtable. Smarter automation comes after that foundation, in phases, with people reviewing it.

Scheduling usually breaks after the booking, in the confirmation, the intake form, the reminder, the note to whoever does the work and the CRM update. Every handoff is a chance for something to fall through.

For Lumik LLC, we built Suru, an executive assistant that lives in Slack. It checks availability, creates approved calendar events, updates CRM records and sends reminders from the channel where the team already works. Start with the handoff that breaks most often, not the whole calendar.

Yes, as long as it drafts and a person decides. Inbox triage means sorting, labeling, summarizing long threads and drafting replies. Document work means pulling the right details into a quote, form or contract draft.

Suru handles the inbox side for Lumik too. In the words of Adam Fajardo, Lumik's CEO and founder, it "drafts what needs drafting, and waits for one of us before anything goes out."

For Theiagen Genomics, we designed quote and statement-of-work workflows where pricing follows fixed rules and a person reviews every draft. Those workflows were still in beta testing when we published the case study. We'd rather a document workflow prove itself on reviewed drafts before anyone relies on it.

Usually, yes. What you want out of it is the decisions, owners and due dates, captured the same way every time and sent to wherever your team tracks work.

Our meeting-minutes workflow for Lumik LLC's federal program work turns recordings into human-reviewed minutes with decisions, owners and due dates. Christopher Sunga, Lumik's VP and co-founder, described what it was like before: "Minutes used to take hours every week, and how good they were depended on who was taking notes."

Transcription and action extraction can still be wrong, so a designated person reviews each draft before it counts as the record.

If your team answers the same questions from memory every week, a knowledge assistant is a good early project.

Staff at Brambleton Community Association kept researching the same parking, exterior-change and rules questions across covenants, bylaws and policies, and answers could vary depending on who handled the inquiry. HOAgent pulls the relevant passages with citations and drafts a response for authorized staff to verify, revise and approve. It doesn't give legal advice.

The catch is that it's only as good as your documents. Results depend on current, complete and correctly versioned sources, so someone has to own them.

The same idea works on messier material. For Keyllective, we built ThreadFinder, a research prototype that lets the team ask questions across interviews, podcasts and local video, with the source attached to each answer.

Anything you rebuild on the same day every week is a candidate.

For Aerial View Podcast, a recurring Monday workflow handles normal episode updates: the transcript, the episode page and promotional drafts that get reviewed before they go out. An operator steps in only when an exception comes up.

For Theiagen, we built an executive delivery board where outcomes link back to their source records, so status doesn't live across Notion, Google Drive and meetings.

For most small businesses, the first report worth automating is simpler: a weekly summary of new leads, open quotes and overdue follow-ups, pulled from the systems you already use.

Anything where a confident wrong answer is expensive, like legal conclusions, clinical or scientific claims, pricing exceptions and anything sent to customers without a review step. Our rule is to keep people in control where judgment matters.

ColoWatch, a virtual colonoscopy startup, uses an AI-assisted workflow to draft patient and provider content, and clinical and business reviewers approve it before it's published. Brambleton's video workflow works the same way. AI helps produce short videos for routine notices, and people check accuracy, tone and brand fit before anything goes out.

Also, don't make a rip-and-replace your first project. Build around the tools your team already trusts, the way KC Enterprise kept its Airtable base.

Pick the process that fails most often, automate the repetitive part, keep a person on the decision, and check it against the success measure you agreed on before you add the next one. If you want to budget it first, see what AI consulting costs in Northern Virginia. If you run a Loudoun County business, Launch Loudoun may subsidize part of that first project.

Book a free 15-minute intro at galang.ai/book and bring the one process you'd fix first. We'll tell you whether AI fits, what a responsible first step looks like and what not to automate.