AI Coaching Assistant: Pilot With 5 Clients in 6 Weeks

An AI coaching assistant is software that helps coaching work happen. It does not replace the coach. It drafts summaries, sends reminders and prompts reflection between sessions. Our advice: pilot it on small, repeatable tasks first. Keep a person checking anything that reaches a client. Run a six-week test with five clients before you roll it out.

What is an AI coaching assistant?

Think of a kitchen helper. It chops the vegetables. It does not decide the menu.

You will meet four main types.

Type	What it does
Meeting assistant	Listens to the call and writes the summary
Between-session companion	Sends nudges and reflection questions when you are not there. Often sold as a virtual coaching chatbot
Admin automation	Handles email, booking and scheduling
Content engine	Tailors worksheets or curriculum to a client's goals

Most tools mix more than one type. Vendors sell them under many names: smart coaching solutions, a digital coaching platform, an artificial intelligence mentor. The label tells you little. For the case for the category, read TD Magazine's three reasons AI coaching has potential.

Do not buy on the label. Buy on the job. Ask one question: does this do the task I need done? A content engine will not fix a scheduling headache. A meeting assistant will not run your billing.

Match the feature list to your real bottleneck. If you are not sure which one that is, start with what to automate first in a coaching business.

What the AI handles, and what stays with the coach

AI is a good assistant. It is a poor coach. Get the split wrong and it costs you client trust.

Hand these to the tool:

• First drafts of session notes and follow-up emails.
• Scheduling reminders and simple check-in messages.
• Summaries of what was said on a call.
• Basic reflection prompts between sessions.

Keep these with you:

• Reading where a client really is.
• Anything emotionally complex or high-stakes.
• Calls about risk, safety or ethics.
• The coaching relationship itself.

The research points the same way. A 2026 randomised controlled trial compared accredited human coaches with automated AI coaches across 114 coachees. It found substantial effects across coaching outcomes, but only for human coaching.

A separate 2025 study looked at one 60-minute session. Clients rated their working alliance with an "AI coach" about as highly as with a human one. There is a catch. The AI was simulated. Trained human coaches were playing the part behind an avatar. So the study shows clients are open to the idea. It does not show that today's tools can coach.

The authors of the 2026 trial frame this as co-regulation. Our plain version: the AI holds the repeatable structure and you hold the person. We treat AI as an enthusiastic intern, not a magic button. Useful for the busywork. Not trusted with the hard conversations.

What you gain

The main gain is time. The assistant takes hours you were spending on admin.

• More clients, same hours. Automated check-ins let you run a bigger group without more coaching time.
• Clients stay engaged. Nudges between sessions keep the work moving.
• Hybrid programs. Human sessions every two weeks, AI-supported check-ins in between.
• Faster admin. Scheduling and reminder drafts happen without you.

Start with one repeatable task, like weekly check-in messages. Add the next one only when the first one works.

Pricing varies a lot. Tools for solo coaches are usually a flat monthly fee. Larger deals add custom integrations and support. Seats, data storage and CRM connections move the price. Marketplace listings show how vendors set out their tiers. CoachRx on Capterra, a platform for fitness coaches, is one example to compare against.

If your program already runs on a defined method, the bigger win is not buying a tool at all. It is encoding your own approach into a system, so you scale the program without scaling your hours.

Risks, privacy and a short governance checklist

Check the risky parts before you roll anything out. This is how you keep client trust.

1. Consent. Do clients know a tool is listening or reading their messages?
2. Recordings. Where do call recordings live, and who can open them?
3. Data mapping. Do you know what data goes where?
4. Bias. Could the tool treat some clients unfairly?
5. Accuracy. What happens when the tool gets something wrong?

The NIST AI Risk Management Framework gives you a four-part structure for this. It is voluntary guidance, and it asks for records and ongoing monitoring instead of a one-off check.

• Govern. Decide who owns the tool and who signs off its outputs.
• Map. Write down what data goes in and what comes out.
• Measure. Track errors, like a wrong summary or a missed detail.
• Manage. Fix the process when something goes wrong, and log it.

Add three controls on top. An approval gate before anything reaches a client. A log of what the tool did and said. A named person who owns escalation.

Think of the kitchen pass. Nothing leaves the kitchen until someone checks the plate. DevWiz has a clear guide on when to put a person in front of AI. For layering the controls, see our notes on AI agent security.

How to choose and pilot an AI coaching assistant

Picking a tool is about fit more than features. Run this checklist first.

1. Job fit. Does it solve your real bottleneck?
2. Integrations. Does it talk to your CRM and booking system?
3. Data ownership. Can you export your data, and who owns it?
4. Custom prompts. Can you shape prompts and workflows to your method? This is what personalised coaching AI should mean.
5. Vendor trust. What do the terms say about training on your client data? Read the terms, not the pitch.

It also helps to be clear on the kind of coaching you do. Leaderly's breakdown of four types of workplace coaching is a quick way to name yours before you match a tool to it.

Then run a small pilot.

• Pick 5 clients who are comfortable with the idea.
• Set two or three measures, like response time or client satisfaction.
• Put a human review step before any message goes out.
• Run it for six weeks.
• Decide go or no-go on the results.

If the pilot works, move to production with three rules. Permission tiers, so it is clear who can approve what. An audit trail of every action the tool took. Basic training, so everyone knows the limits.

Prompts and workflows you can test today

You do not need a developer for this. Start with one workflow: record the call, summarise it, check the summary, approve it, send it.

Try these prompts:

• Session summary. "Summarise this call in three bullet points: goal discussed, progress made, next action."
• Client check-in. "Write a warm, short check-in message asking how the client's week went. Refer to their last stated goal."
• Weekly recap. "Draft a one-paragraph recap of this week's session for the client's records."
• Nudge. "Write a one-line reminder about the client's agreed action step."

Test each prompt on a small group first. Track any wrong or made-up details before you trust it with more clients.

If you are new to prompting, a structured course helps. One paid option is this AI prompt engineer certification.

Making it work with the tools you already use

Most coaches already run a CRM, a booking system and a payment tool. The assistant should slot into that stack.

Check three things before you buy. Does it connect to your CRM? Does it read your calendar? Can it push summaries into your client records, so nobody copies and pastes?

Weak integration undoes the time saving. The tool works on its own, but someone still moves data between systems by hand.

If you run group programs, look for one dashboard across clients. If a team approves messages before they go out, Contentelli has free tools for content review and approval that are worth a look. If you plan to offer the assistant under your own brand, DevWiz explains white-label AI software for coaches.

Getting your team ready

A tool is only as good as the person using it.

Start small. Pick one task, like drafting check-in messages, and train on that alone. Add a second task once the team is confident.

Cover three things: how to write a clear prompt, how to spot a wrong or made-up answer, and how approval works before anything reaches a client.

Build a one-page sheet of your best prompts. New staff copy from it. Review it every few months, because prompts need small changes as the tool updates.

Give junior coaches and support staff narrower permissions. Someone drafting a summary does not need the access of someone approving a client message.

Keeping it fair and honest

A tool trained on general data may phrase things in a way that suits some clients better than others. Language, culture and style all play in. Check outputs across a range of client profiles.

Tell clients when a message or summary involved an AI tool. Hiding it risks trust, and trust is the product in coaching.

Hold one rule. An AI tool never makes the call on a client's risk, safety or readiness. That stays human, every time.

Read the outputs often. If something feels dismissive or oddly generic, raise it with the vendor and change how you use the tool.

How to tell if it is working

Watch three things.

Measure	How to track it
Time saved	Admin hours before and after, over a month
Client outcomes	Retention, attendance and feedback scores over a full program cycle
Error rate	How often the tool gets something wrong, like a mixed-up name or a made-up detail

Review all three together every six weeks. If time is saved but errors are climbing, slow down. Tighten the approval step before you expand.

Designing an assistant clients want to use

If a client has to think hard about how to use the tool, it is failing.

• Short, plain messages. No jargon.
• Clear sender. Clients know when a message comes from the assistant and when it comes from you.
• Easy off switch. A simple way to stop nudges.
• Sensible timing. Messages arrive at reasonable hours.

If a client complains about the tool more than once, treat it as a design problem.

When to orchestrate instead of bolting on a tool

A bolt-on tool saves admin time. It does not carry your method.

That takes a different build. You encode how you coach into an AI Operating System: a set of AI employees that each own one part of delivery, with you approving the parts that carry judgment. We build these with Claude Code, and it does not take a technical founder. See Claude Code for non-technical founders and the AI delivery stack for coaching businesses.

This fits founder-led coaching and education businesses at $1M or more, with a method worth protecting.

Run the six-week pilot first. It shows you where your method lives. That is the part worth building a system around.

James Killick

Your next step

If the pilot shows the ceiling is your hours and not your admin, a tool will not fix it. The AI Orchestrators Program turns your method into a working prototype and a scaling roadmap in 90 days.

Not sure which problem you have? Take the assessment. It takes a few minutes and shows where your program has the most room to scale.

Sources

• NIST, AI Risk Management Framework
• Amber S. Barger, Artificial intelligence vs. human coaches (Frontiers in Psychology, 2025)
• de Haan, Terblanche and Nowack, human and AI chatbot coaching trial (Human Resource Development International, 2026)