Start with one well-chosen job, such as answering missed calls or sending invoice reminders, and see whether it saves staff time or wins back revenue before you add more. Bigger multi-step workflows that tie several systems together cost more and take longer to pay back.
That's the short answer. The rest of this post breaks the cost down by job, shows you how to run the payback math with Los Angeles labor costs, and covers the pitfalls that turn a modest project into an expensive regret.
The question behind the question
When an owner asks "what does AI automation cost?", they usually mean something more specific:
- "My front desk can't keep up with the phones. What does it cost to stop losing those calls?"
- "We re-type the same intake info into three systems. Can that go away, and for how much?"
- "I'm owed money and nobody has time to chase it. What would it take to automate reminders?"
- "Leads come in and go cold before anyone calls back. Is fixing that a small problem or a huge one?"
A lot of what you'll find online is either national averages built for software buyers or pitches for big custom builds. Neither helps a clinic in Pasadena or a contractor in the Valley decide whether to spend money on this next quarter. So we'll look at it job by job.
You're not alone in asking. Thryv's AI and Small Business Adoption Survey was conducted in April 2026 among 561 U.S.-based small and medium-sized businesses, all of them non-Thryv customers. It found AI adoption among U.S. small businesses has risen to 66%, up from 55% a year ago, and more than half (53%) of respondents spend at least $100 per month on AI tools.
What AI automation costs, job by job
Every automation project has three cost layers. Knowing them makes any quote easier to read.
- Setup (one-time). Mapping how the job works today, connecting your tools (phone system, calendar, CRM, accounting), writing the AI's instructions, testing with real examples, and training your staff. SOFTEN
- Software and usage (monthly). The subscriptions and per-minute or per-message fees for the AI and the tools it connects to. These often scale with volume.
- Upkeep (monthly or as needed). Adjusting prompts when your services or prices change, fixing a broken connection when a vendor updates its app, and checking that the outputs are still right.
Here's how those layers compare for the four jobs LA owners ask us about most, plus the bigger multi-step option. This is a relative guide, not a price list: your quote depends on call volume, how many systems are involved, and how messy the current process is. We'll give you real numbers for your situation during an audit.
| Job | What it does | Setup (one-time) | Monthly | Payback outlook |
|---|---|---|---|---|
| Missed-call answering and text-back | AI answers or texts back every call your team can't pick up, books or captures the request | Lower | Low to moderate | Often the fastest |
| Client or patient intake | Forms, calls or chats flow straight into your CRM, calendar or practice software, no re-typing | Moderate | Low to moderate | Fast to moderate |
| Invoicing and payment reminders | Invoices go out on time, polite reminders follow on a schedule, payments get logged | Lower to moderate | Low | Fast, if receivables are high |
| Lead follow-up | Every new inquiry gets a fast first reply and a short follow-up sequence until it books or opts out | Moderate | Moderate | Slower to prove, bigger upside |
| Multi-step agent workflow | Several of the above, tied together, with an AI agent handing work between systems | Highest | Highest | Longest |
Missed calls: the cheapest place to start
Industry benchmarks like that describe many sectors at once, not your business, so the useful move is to check your own call logs and count how many calls go unanswered in a normal week. For a lot of owners, that number is the biggest leak they have.
The software side is relatively cheap. Pricing for AI answering services varies widely, so get current quotes from a few vendors before you commit. Compare per-minute rates for AI answering and human answering services side by side, and ask each vendor for current pricing in writing. What you're paying for in setup is making the AI actually useful: knowing your services, your hours, your service area, which calls are urgent, and how to book into your real calendar instead of just taking a message.
Intake: where clinics and law offices save the most time
Intake costs more to set up because it touches more systems. A new patient or client fills out a form, someone copies it into the practice or case software, someone else checks insurance or conflicts, and someone sends the welcome email.
Invoicing and reminders: money you've already earned
This one is easy to justify. In a QuickBooks survey, 56% of surveyed small businesses reported being owed money from unpaid invoices, averaging $17.5K per business, and 47% reported a portion of their invoices were overdue by more than 30 days. Automated reminders don't do anything clever. They just never forget, and they never feel awkward asking twice. For a business carrying a large balance in receivables, even a modest improvement in how fast money comes in can cover the project.
Lead follow-up: the slowest to pay back, but often the biggest prize
Follow-up automation costs more because it involves judgment: what to say to whom, when to stop, and when to hand off to a person. It also takes longer to prove its value, since you're measuring booked jobs over weeks rather than hours saved this week. Done right, it's often the job with the most upside.
Start with one job, not a platform. Start with the single most painful, most repetitive task, automate it well, measure it for a month, and only then add the next one.
How to run the payback math (with Los Angeles numbers)
You don't need a spreadsheet model. You need three numbers:
- Hours this job eats every week, across everyone who touches it.
- What an hour of that work costs you.
- Revenue you lose because the job isn't done (missed calls, late invoices, cold leads), if you can estimate it.
In Los Angeles, the floor on that second number keeps rising. Effective July 1, 2026, the Los Angeles minimum wage rises to $18.42, an increase of 55 cents an hour. Minimum wage floors often rise over time, so build some headroom for future increases into your payback math. Front-desk and admin staff often earn more than minimum wage, and payroll taxes and benefits add to the cost, so treat the minimum wage as a conservative starting point for the math, not a ceiling.
A worked example
Hypothetical: this is not a client story, just the math with round numbers.
A two-location physical therapy practice in Los Angeles has front-desk staff spending about 10 hours a week returning missed calls, playing phone tag and re-entering intake details.
- Labor cost of the task: 10 hours ร $18.42 = $184.20 a week, or about $798 a month (ร4.33 weeks). That's at minimum wage. At a more realistic front-desk wage the number is higher.
- Automation cost: an illustrative $2,000 setup plus $150 a month for software and upkeep.
- Net monthly savings: $798 โ $150 = $648 a month.
- Payback on setup: $2,000 รท $648 โ 3.1 months.
That counts only staff time. It doesn't count a single new patient booked from a call that would otherwise have gone to voicemail. If the practice books even one or two extra new patients a month from calls it used to miss, payback gets much shorter. Use your own value per new client to check.
Plug in your own hours and wages rather than relying on anyone's averages, including ours. Owners' self-reported savings estimates are a useful sanity check, but your call logs, timesheets and receivables report are the real evidence.
Los Angeles factors that change the price
A few things specific to doing business here push costs up or down:
- Rising labor costs. Because the local minimum wage tracks inflation, an hour saved tends to be worth a bit more each year. That shortens payback over time.
- Bilingual callers. Many LA businesses take calls and intake in both English and Spanish, and sometimes other languages too. A setup that handles a second language well needs more testing, so budget toward the higher end of your quote.
- Call recording rules. Any AI phone system that records or transcribes calls should tell callers so at the start of the call. Check the consent rules that apply to you with your attorney before going live. This is a setup item, not a monthly cost, but it can't be skipped.
- Regulated data.
- Multiple locations. Restaurant groups and multi-office practices across the basin often have different hours, phone numbers and staff per site. Each location adds configuration, though not a whole new project.
Pitfalls that blow up the budget
The most expensive AI project is the one nobody uses. Before you buy a platform, decide what job it should do, and clean up a messy process before you automate it.
- Automating chaos. If three staff members handle intake three different ways, the AI will faithfully copy the confusion. Spend an hour agreeing on the right way first.
- Paying for seats and features you won't use. Many tools price by tier, and the one feature you need is often on the expensive tier. Ask what the bill looks like at your real volume, including overage fees.
- Ignoring per-minute and per-task fees. Usage-based pricing is fine until a busy season multiplies your bill. Get the overage rate in writing.
- No human handoff. An AI that can't pass an upset caller or a complex legal question to a person will cost you more in goodwill than it saves in hours.
- No owner for upkeep. Prices change, services change, vendors update their apps. Someone, on your team or ours, has to own keeping the automation accurate.
- Skipping training. Check how much training your team will need before you commit to a tool, and budget time for it. Budget an hour or two of staff walkthrough into every project.
A quick checklist before you get a quote
Bring answers to these and any quote you get will be faster and more accurate:
- [ ] Which single job hurts most right now (calls, intake, invoicing, follow-up)?
- [ ] Roughly how many calls, forms, invoices or leads does it involve per week?
- [ ] How many staff hours a week does it eat?
- [ ] Which tools does it touch (phone system, calendar, CRM, practice or accounting software)?
- [ ] Do you handle regulated data (health, legal, financial)?
- [ ] What languages do your customers use?
- [ ] Who on your team will own the automation once it's live?
- [ ] What would "working" look like after the first month (calls answered, hours saved, days-to-paid)?
What we'd do
If you're a Los Angeles owner weighing this for the first time, here's the approach we'd recommend. It's also how we run our own projects.
- Find the leak. Look at a week of call logs, intake forms and open invoices to see where time and money actually go. Usually one job stands out.
- Price that one job. Get a fixed setup price and a realistic monthly number at your real volume, not a vendor's best-case tier.
- Build it small and test it with real examples. A single job is a much shorter build than a multi-step workflow, so ask any provider for a timeline in writing.
- Measure for a month.
- Then expand. Once the first job pays back, connect the next one. That's when it makes sense to look at AI automation across your operations or agentic workflows that hand work between systems.
We work with owners across the region.
Next step
Pick the one job that bothers you most this week and write down how many hours it takes. Then book a Free Automation Audit. We'll look at that job with you, give you a real setup and monthly number, and tell you honestly if it's not worth automating yet.
Sources
- Intuit QuickBooks โ 2025 US Small Business Late Payments Report (May 28, 2025)
- CPA Practice Advisor โ SMB AI Use Continues to Rise, But They Need Help (July 14, 2026)
- Proskauer Rose LLP โ Coming to an Employer Near You: Los Angeles' Ever-Rising Minimum Wage (April 15, 2026)