How to build a custom AI specialist for your industry

TL;DR

ExtraSeat's catalog covers common domains: financial analysis, legal advisory, technical architecture. Custom specialists cover everything else — your specific industry, your clients' specific questions, your company's specific knowledge. This article explains how to build one, what makes a good system prompt, and where custom specialists outperform generic ones.

Table of contents

1. Why generic specialists have limits
2. What a custom specialist actually is
3. How to build one: step by step
4. Writing an effective system prompt
5. Choosing a voice and model
6. What custom specialists do better
7. FAQ

Why generic specialists have limits

A generic financial analyst knows general financial principles. It can answer questions about burn rate benchmarks, gross margin norms, and revenue recognition — the kind of knowledge that applies across industries.

But if you run a logistics business, the questions in your client calls aren't generic. They're about freight margins, fleet depreciation, carrier cost structures, and inventory turnover benchmarks specific to your sector. A generic financial analyst knows what a good gross margin looks like in the abstract. It doesn't know what one looks like for a 3PL with a cold chain operation in Southeast Asia.

That gap — between general domain knowledge and specific industry expertise — is where custom specialists live.

The pre-built catalog is the right starting point for most teams. Custom specialists become the right tool when your business operates in a niche where general knowledge isn't enough, or when you want the specialist to know your company, your clients, and your way of working.

What a custom specialist actually is

A custom specialist is an AI meeting participant you configure yourself. You control three things:

The system prompt — the instructions that define the specialist's domain, scope, and behavior. This is where you tell it who it is, what it knows, how it should communicate, and what it should and shouldn't answer.

The voice — the synthesized voice it uses when speaking in the call. You choose from available options to match the tone and persona you want.

The AI model — the underlying language model that powers its responses. Different models have different strengths; you pick the one best suited to your domain.

Everything else — the audio pipeline, the floor management, the Google Meet integration — works the same way as a pre-built specialist.

How to build one: step by step

Step 1: define the specialist's role. Before writing the system prompt, answer three questions: What domain does this specialist cover? (Be specific — "healthcare compliance for medical device startups" beats "healthcare.") What questions will it most commonly be asked? What should it not answer? Defining the out-of-bounds prevents off-topic responses that erode trust.

Step 2: write the system prompt. Plain language instructions — who the specialist is, what it knows, how it should behave. Length can range from a few sentences to several paragraphs. Longer isn't always better; more specific always is.

Step 3: choose a voice. Browse available options and pick one that fits the persona. Consider tone, register, and clarity for the meeting context. Test it before using it with clients.

Step 4: select an AI model. The default model works well for most business use cases. For highly specialized domains, testing two or three options during setup is worth the time.

Step 5: test before your first live session. Run a test session before using the specialist with a real client. Ask the questions you expect to come up. Are the answers accurate, appropriately scoped, and calibrated to your audience? Adjust the system prompt based on what you find.

Writing an effective system prompt

The system prompt determines the quality of the specialist's responses. Here's what separates a useful one from a generic one.

Start with the role and domain

Tell it who it is and what it knows. Be specific about the industry or context.

Generic:

You are a financial analyst. Answer financial questions.

Specific:

You are a financial analyst specializing in early-stage B2B SaaS companies. You have deep knowledge of SaaS-specific metrics: ARR, MRR, churn rate, net revenue retention, CAC payback period, and LTV:CAC ratios. When asked about benchmarks, reference figures that apply to companies at Series A stage ($1M–$10M ARR) unless otherwise specified.

The specific version produces answers that are useful in the room. The generic version produces answers that could have come from a textbook.

Define the communication style

Describe how the specialist should speak — tone, length, format. It's answering out loud in a meeting, not writing a report.

Example:

Respond conversationally and concisely. Answers should be 2–4 sentences in a live meeting context — enough to answer the question clearly, not so much that the conversation stalls. Avoid jargon unless the person asking is clearly technical. When numbers are relevant, lead with the number.

Set scope boundaries

Tell it what it shouldn't answer — questions outside its domain, questions requiring licensed advice that shouldn't come from an AI without caveats, or topics sensitive in your client context.

Example:

Do not give advice on specific tax strategies or legal structures. If asked a question that requires licensed legal or financial advice, note that the answer requires a qualified professional and offer general context only.

Add company-specific context (optional but powerful)

If you want the specialist to know about your company or your clients, include it.

Example:

Our company works with mid-market manufacturing businesses in Germany and Austria. Our clients typically have 50–500 employees, are family-owned, and are evaluating digital transformation investments. When benchmarking, reference figures relevant to this segment.

This transforms a generic domain specialist into something closer to a real team member with context.

What the difference looks like in a live meeting

Same question, two specialists — one generic, one configured for early-stage SaaS.

Question: "What's a good gross margin benchmark for us at this stage?"

Generic specialist:

"Gross margins vary significantly by industry. For software businesses, gross margins in the 60–80% range are generally considered healthy, though this depends on your cost structure and business model."

Custom specialist (Series A SaaS):

"For B2B SaaS at Series A — roughly $1M to $10M ARR — gross margins in the 70–80% range are typical for pure software. If you have significant professional services revenue, 60–70% is more common. What's your current services-to-software revenue split? That changes the benchmark."

The generic answer is accurate. The custom answer is useful — it gives a number, contextualizes it for their stage, and asks the follow-up question that moves the conversation forward. The difference comes entirely from the system prompt.

Choosing a voice and model

Voice: The specialist will be heard by everyone in the call. A voice that sounds natural and clear in a meeting context matters more than one that sounds impressive in isolation. Choose one that fits the domain — a legal advisor might warrant a more measured delivery; a brand strategist might work better with a warmer tone. Test it in a mock session before using it with clients.

Model: The default model works across most business domains. For highly specialized technical areas — medical, legal, engineering — testing alternatives is worth the time. Different models have different depth in niche areas, and the difference is most visible in domain-specific questions.

What custom specialists do better

Compared to pre-built catalog specialists, custom specialists have four clear advantages.

Niche industry expertise. The catalog covers horizontal domains. Custom specialists go deep into vertical ones — healthcare compliance, logistics finance, construction law, SaaS sales strategy. The more niche the domain, the bigger the gap between generic and custom.

Company-specific knowledge. If you want the specialist to know your products, your pricing, your clients, or your methodologies, that context goes in the system prompt. Pre-built specialists don't know any of it.

Audience calibration. A specialist built for conversations with CFOs should communicate differently than one built for operations managers. Pre-built specialists use a general register. Custom specialists use the one you specify.

Boundary control. In regulated industries or sensitive client contexts, you need control over what the specialist will and won't say. Custom specialists give you that. Pre-built specialists have general guardrails — not the specific ones your context requires.

Frequently asked questions

How long does it take to build a custom specialist?

The technical setup takes minutes. Writing a good system prompt takes longer — expect 20–40 minutes for a first draft if you think carefully about the role, scope, and communication style. Testing adds another session. Total time from start to production-ready: a few hours.

Do I need technical knowledge to build a custom specialist?

No. The builder is entirely text-based. You write the system prompt in plain language — no code, no configuration beyond voice and model selection.

How many custom specialists can I create?

Multiple. You can build specialists for different domains, different clients, different meeting types. Each is stored separately and can be brought into any session.

Can I update the system prompt after the specialist has been used?

Yes. Changes take effect in the next session.

What's the difference between a custom specialist and just using an AI assistant in another window?

The integration. A custom specialist built in ExtraSeat joins the meeting as a voice participant with full conversation context. It's not a separate tool you switch to — it's in the call. The system prompt applies throughout the session without any copy-pasting or context re-injection.

Conclusion

Pre-built specialists cover the ground most teams need most of the time. Custom specialists cover the rest — the niche domains, the company-specific knowledge, the audience-specific register that a general catalog can't replicate.

If the questions in your meetings are specific to your industry or your clients, the generic version will feel generic. A custom specialist built with the right system prompt will answer those questions the way a real expert in your field would.

→ Make better decisions in your next meeting — buy credits to start

Related reading:

• The small team's guide to having every expert in every meeting
• How multi-bot AI meetings work: when you need more than one expert
• How AI voice bots join Google Meet calls

This article is part of The small team's guide to having every expert in every meeting -- a comprehensive guide to AI meeting specialists for small teams.