Before the examples
What an AI workflow automation looks like.
An AI workflow automation is a chain of steps that starts on its own, uses AI to read, write or judge something a fixed rule can’t, and hands the result to a person or another system. The eleven examples below show what that looks like in a service business: three real projects we built, then eight everyday workflows that fit most plumbing, HVAC, electrical, roofing and remodeling companies.
Every example has the same five parts. A trigger starts it: a missed call, a new form, a finished job. An input feeds it: the caller’s words, the form, a voice note. An AI step does the part a simple rule can’t, like understanding a customer’s description or drafting a message. An action moves the result somewhere: a text, a calendar, your CRM. And a person checks or decides the part that needs judgment.
If any of the five is fuzzy, the workflow isn’t ready to build. For the definition in more depth, read what AI automation is .
Examples 1 to 3
Three real projects we built.
These are Lithium Marketing client projects, with the results each business saw. Each started from one specific bottleneck, not a wish to “use AI.”
Plumbing company
After-hours lead capture
- The problem
- 60% of after-hours emergency calls were going to voicemail, and those callers were going to competitors instead.
- What we built
- A custom AI phone system that triages how urgent each call is, dispatches the on-call tech by SMS for emergencies and books non-emergency jobs directly into the schedule.
- AI phone system
- SMS dispatch
- Schedule booking
The result
3x
more leads
Real estate agency
Lead‑qualification bot
- The problem
- Agents were losing hours chasing unqualified leads instead of working with the buyers who were ready.
- What we built
- An AI chatbot that pre-screens buyer inquiries on budget, timeline and location, then sends the hot leads straight to the right agent’s calendar.
- AI chatbot
- Lead qualification
- Calendar booking
The result
52%
conversion
Gravel contractor
GIS-powered lead generation
- The problem
- Search volume for the contractor’s services was too low for traditional SEO to bring in enough work.
- What we built
- A GIS-powered geospatial database that identified rural properties with gravel driveways, feeding an automated direct-mail campaign with AI-generated personalization.
- GIS database
- Data pipeline
- Automated direct mail
The result
16,000+
rural properties identified
Notice where people stay involved. The plumbing system texts a real on-call tech, who decides how to handle the emergency. The real estate bot books time with a real agent, who does the selling. The AI does the sorting, writing and reaching out; people make the calls that carry risk.
Examples 4 to 6
Workflows for the phones and the inbox.
The next eight are everyday workflows, described in general terms to show how they work. They are not reported client results.
4. Missed-call text-back
Starts when a call rings out or goes to voicemail. The AI sends a text within a minute (for example, “Sorry we missed your call, this is the office at [your company name]. What can we help with?”) and reads the reply to sort it: emergency, new job, existing customer or spam. A person gets the sorted reply and calls back, with emergencies first. Check how fast real requests get a call back and how many turn into booked jobs. The missed calls guide covers this one in depth.
5. Web form to a ready-to-quote job record
Starts when a customer submits a project form with a description and photos. The AI copies the customer’s details into a draft job record in your CRM, summarizes the request in two lines and flags anything missing, like the address or the size of the job. The estimator reviews the photos, asks the follow-up questions and decides whether to book a site visit. The workflow never sets a price. Check that each draft matches what the customer wrote and reached the right estimator.
6. Unanswered-lead reminder
Starts when a new inquiry sits in your intake system with no response logged. The AI spots the unanswered record after the window you set, and sends the office a reminder with the original request and a suggested reply. A person decides whether to call, text or close it out. Check how many leads sit unanswered each week, before and after.
Examples 7 to 9
Workflows for estimates and paperwork.
7. Voice note to draft estimate
Starts when a tech records a voice note after a service call. The AI transcribes it, pulls out the parts, labor and recommendations, prices them from your price book and builds a draft estimate in your field-service software. The office or the tech checks the quantities and prices, fixes anything misheard and sends it. Check how many drafts needed corrections and how much sooner estimates go out. The voice notes guide walks through the full setup.
8. Supplier invoices into the books
Starts when a supplier invoice lands in a shared inbox. The AI reads the PDF, pulls the vendor, date, line items and total, matches the parts to open jobs where it can and prepares an entry in your accounting software. Your bookkeeper approves each entry before it posts. Check how many invoices matched cleanly and how many needed a hand.
9. Estimate follow-up
Starts when an estimate has gone unanswered for a few days. The AI writes a short text and email from the estimate itself (the work, the warranty, the next open slot) and sends them on a schedule you set, stopping the moment the customer replies. A person takes every reply. Check your estimate close rate before and after. The lead follow-up guide has sample messages.
Examples 10 and 11
Workflows for the field and after the job.
10. Equipment manual search
Starts when a tech types or says a question on site, like “what’s the clearance spec for this furnace model.” The AI searches your library of approved manuals and bulletins and answers with the passage and the page it came from. The tech checks the source and follows the manufacturer’s instructions and your safety procedures. Check whether the cited passage answered the question and that outdated documents stay out of the library. The equipment manuals guide explains how it’s built.
11. Review request and maintenance reminder
Starts when a job is marked complete. The AI sends a thank-you text a few hours later with a direct link to your Google review form, and schedules a maintenance reminder on the interval you set for that job type. If the customer’s reply mentions a problem, it alerts a person right away. A person handles any issue and approves replies to new reviews. Ask every customer, never offer anything in exchange and never screen out unhappy ones; Google’s rules forbid all three. Check reviews per finished job and how many reminders turn into booked visits.
At a glance
All eleven examples on one page.
| Example | Starts when | The AI | A person |
|---|---|---|---|
| After-hours lead capture | An after-hours call | Triages urgency, texts on-call tech, books | On-call tech handles emergencies |
| Lead qualification bot | A buyer inquiry | Screens budget, timeline, location | Agent works the qualified lead |
| GIS lead generation | A property data run | Finds targets, personalizes mail | The contractor owns the campaign |
| Missed-call text-back | A missed call | Texts, sorts the reply | Calls back, emergencies first |
| Form to job record | A project form | Drafts the record, flags gaps | Estimator decides on a visit |
| Unanswered-lead reminder | No response logged | Reminds with a suggested reply | Office calls, texts or closes |
| Voice note to estimate | A tech’s voice note | Drafts and prices the estimate | Office checks and sends |
| Supplier invoices | An invoice email | Reads and prepares the entry | Bookkeeper approves |
| Estimate follow-up | An unanswered estimate | Writes and sends reminders | Takes every reply |
| Manual search | A tech’s question | Finds the passage and page | Tech checks the source |
| Review and reminder | A finished job | Sends review link, schedules reminder | Handles issues, approves replies |
Make it yours
How to turn an example into your workflow.
- 1
Pick one
Choose the example closest to your biggest daily headache, and count that headache today.
- 2
Write the five parts
Trigger, input, AI step, action and the person who decides, using your own systems and staff.
- 3
Pilot it
Run it on real work for two to four weeks, with someone checking every result.
- 4
Measure and decide
Compare against your starting count. Keep it, adjust it or drop it.
Lithium builds these as custom AI automation . Each project is scoped after a free discovery call, and most builds take one to two weeks. If your only problem is the phones, our managed AI receptionist answers every call 24/7 for $179 a month, with no contract and no setup fee. Bring one real example of the task to the call, along with the awkward cases. They show fastest whether the task is worth automating.
