AI for Admin. Never for Clinical. Inside One California Practice's Operating Rule.

Looking at the Data: In California, 70.4% of employers use AI weekly — but only 56.5% of their employees do. Dr. Givona Sandiford runs a 70-person speech therapy practice across the Bay Area, LA, and Southern California using AI to handle nearly everything administrative. The one place she won't let it operate: clinical decisions. Her rule isn't about distrust of technology. It's about where automation creates leverage versus where it erodes the work itself.

Watch the full interview with Dr. Sandiford

The data: California's AI adoption gap is bigger than it looks

TriNet's 2025 State of Work report surveyed 90 California employers and 46 California employees at small-to-medium businesses. The headline numbers:

A 14-point usage gap is not small. But the more interesting finding is buried underneath it: employers report AI as acceptable across nearly every HR function — recruiting, performance reviews, scheduling — at rates well ahead of what their employees actually feel comfortable with.

The question isn't whether to use AI. It's where to use it — and where to refuse.

The data point most California SMB operators miss: this isn't a trust gap or a tools gap. It's a clarity gap. Employers haven't told their teams what's automated, what isn't, and where the lines are. Sandiford's practice is one of the few that has.

How Dr. Sandiford uses AI — and where she stops. 

Melospeech, Inc., a Certified Autism Center, offers speech therapy services to clients across California. Dr. Givona Sandiford has grown the business to 70 employees working on remotely managed teams, and she is able to focus on strategy and growth while her teams manage the day to day. AI is one of the major reasons she can. 

For things like administrative tasks, we will usually replace any repetitive task with AI. We started in 2021, and one of the first tools we made was our MeloSuite. It would determine how far away clients are and group them all together so that [a therapist] wouldn't drive very far. Now, we have many more AI tools. Any clinical decisions are made by a human, and then anything that's a repetitive administrative task can be handled by AI because it can be programmed properly. - Dr. Givona Sandiford

The administrative load in a speech therapy practice is heavy. Intake forms. Insurance verification. Session scheduling. Progress note formatting. Without automation, a 70-person practice would need a back office almost as large as its clinical staff. With Arya, a AI chatbot for client scheduling, the MeloSuite, and SLP Spot, a clinician app to support their client visits, all built in-house at Melospeech, the practice runs lean.

The rule sounds simple. It isn't, in practice. AI vendors increasingly market clinical-decision-support tools to healthcare practices. The pressure to adopt is real. Sandiford's position is that the cost of getting clinical work wrong — for the client, for the clinician's credibility, for the practice's reputation — is too high to delegate to a system that doesn't carry the consequences. So, Dr. Sandiford’s team built it in house and left all clinical decisions and interpretation to a human.

“We've also made the most of our technology, SLP Spot, which does all of our notes. Our team will do the therapy, but when we go into the home, we can talk through the SLP Spot and it will do our notes for us in any language we choose. That's automated. The clinician can add notes and use AI to write the report, but it requires the clinician to read over the report and make sure there are no errors within it before signing off. So where we draw the line is about clinical decisions that require a human."  - Dr. Givona Sandiford

Why this matters for SMB operators in California

The State of Work data and Sandiford's playbook point to the same conclusion from two directions:

• The adoption gap isn't a tooling problem. Your team can access the same tools you can. They aren't using them at the same rate because they haven't been told which workflows are theirs to automate.

• A blanket "yes" to AI is as risky as a blanket "no." Both leave employees guessing or building tools that might not be the best fit for automation.

• AI adoption requires thoughtfulness about your business model and what processes are easily automated. Designing tools to streamline repeatable processes is the place to get started.

How to put this into practice

Operators looking at the Dr. Sandiford model can borrow the framework:

1. Write your AI line out loud. One sentence. Who does what. What's automated, what's human-only, and what's hybrid with a human reviewer.

2. Start with admin. Scheduling, intake, formatting, summarization, internal documentation. High leverage, frequent tasks, low judgment risk.

3. Name the protected zones. Anything that requires professional judgment, anything client-facing in a moment of care, anything regulated. These always get human ownership.

4. Tell every employee where the line is. Not in a policy doc no one reads — in a meeting, in onboarding, in the first conversation about a new tool, and then consistently as things evolve.

5. Audit quarterly. AI capabilities move fast. The line that made sense in Q1 may need to move by Q4. Revisit deliberately.

6. Watch the workload signal. If AI is reducing real administrative burden, retention metrics and business should both improve. If it isn't, the rollout may not be doing something productive for your business, even if usage is up.

What's next for California SMBs

California's State of Work data shows a workforce that's more mobile, more benefits-focused, and more skeptical of pay-first retention strategies than most operators assume. The AI adoption gap is one symptom of a broader pattern: California employers and California employees are looking at the same workplace and seeing different jobs.

Closing that gap doesn't require new technology. It requires clearer language about how the technology is being used.

Read the full California regional report from TriNet's 2025 State of Work

See how TriNet supports California SMBs with HR, benefits, and compliance

FAQ

How are California small businesses using AI in 2025?

According to TriNet's 2025 State of Work report, 70.4% of California employers at small-to-medium businesses use AI weekly or more. Adoption is highest in administrative workflows — scheduling, intake, billing, internal documentation — and significantly more cautious in client-facing or regulated work.

Why is there an AI adoption gap between California employers and employees?

The gap (13.9 points in California) is less about trust in the technology and more about clarity. Employees often haven't been told which workflows they're authorized to automate. Operators who name the boundary explicitly tend to see usage gaps close.

What are some thoughtful ways to introduce AI to a small team in California?

Start with administrative workflows that don't require professional judgment. Name the boundaries in writing and in conversation. Train the team on specific approved tools rather than leaving adoption to chance. Audit quarterly as capabilities evolve.

How do California SMBs stay compliant when using AI?

California's regulatory environment — including AB 2930 on automated decision-making and existing privacy laws like CCPA — requires careful documentation of how AI is used in employment decisions and risk assessment of high risk uses of automated decision making tools. Working with advisors  that track state-level compliance changes is one way SMBs can help reduce that exposure.