Copilot Studio Alternative: When to Use a RAG Chatbot

Two Products, One Name: Internal Copilot vs. Public-Facing Agent

"Copilot Studio" describes two very different deployments that happen to share a builder. The first is an internal copilot: an agent that answers employees inside Teams, SharePoint, or Microsoft 365 Copilot Chat, authenticated by Entra ID, grounded in Dataverse and Graph, governed by the same admin center that already manages your tenant. The second is a public-facing agent embedded on a marketing website, answering anonymous visitors who have never logged into anything.

Copilot Studio can technically do both. It was designed for the first. When a team is told to "just use Copilot Studio" for the public site, they are usually pointing a tool built around authenticated, licensed, internal usage at a channel that is the opposite: unauthenticated, unpredictable, high-volume, and public.

That mismatch is not a knock on the product. It's the same reason nobody expects a CRM to double as a CMS. Copilot Studio's identity, governance, and licensing model make perfect sense for 200 employees with Entra accounts. They make much less sense for 200,000 anonymous visitors a month who will never authenticate to anything.

The Consumption Model Explained: Copilot Credits, Packs, and Metering

Copilot Studio bills on a consumption model, and the unit of measurement has changed more than once, which is exactly why a website owner should verify current terms before scoping a budget rather than relying on a number seen in a blog post from last year.

As of September 2026, Microsoft prices standalone Copilot Studio as a tenant-wide license sold in prepaid capacity packs: $200 per pack, per month, for 25,000 Copilot Credits, according to Microsoft's Copilot Studio billing and licensing documentation. Alongside the prepaid packs, Microsoft also offers a pay-as-you-go meter, billed through a linked Azure subscription at $0.01 per Copilot Credit with no upfront commitment, and a one-year prepurchase plan for Copilot Credit Commit Units that Microsoft advertises can save up to 20% versus consuming the same credits on-demand, per Azure's Copilot Studio pricing details page.

Renaming matters here. Microsoft's own documentation notes that starting September 1, 2025, the billing currency for agents changed from "messages" to "Copilot Credits," with no change to the quantity per prepaid pack or to the pay-as-you-go rate at the time of the switch. If a vendor, a reseller, or an old internal spreadsheet still talks about "message packs," that terminology is legacy; budget in credits, and check the live documentation before signing anything.

Consumption per interaction is not flat. Microsoft states that "the number of Copilot Credits counted for each response or action depends on the complexity of the task the agent completes," meaning a simple retrieved answer costs less than a multi-step action that calls a connector or a custom skill. Microsoft publishes a public Copilot Studio agent usage estimator specifically because this variability makes back-of-envelope math unreliable without it.

One detail changes the calculus for internal vs. external use directly: if your organization already has Microsoft 365 Copilot licenses, agent usage inside Copilot Chat, Teams, or SharePoint for classic answers, generative answers, or Microsoft Graph tenant grounding is zero-rated, meaning it does not draw down your Copilot Studio credit pack at all. That zero-rating only applies to usage inside licensed Microsoft 365 surfaces. A public website embed sits outside that boundary and draws on the metered pack or pay-as-you-go balance for every interaction, regardless of how many employees in the building already carry a Microsoft 365 Copilot seat.

Capacity enforcement adds a second wrinkle: purchased credits don't roll over month to month, and exceeding your capacity triggers technical enforcement that can result in service denial rather than a quiet overage bill. For a predictable internal deployment, that's manageable with monitoring. For a public site where traffic spikes are the norm (a product launch, a press mention, a seasonal surge), it's a harder number to plan around.

Pricing and metering details verified against Microsoft's live documentation as of September 2026 (linked above). Microsoft has revised this model more than once; confirm current terms directly with Microsoft or your Microsoft partner before budgeting.

What Copilot Studio Does Better Than Almost Anything Else

None of the friction described above should be read as "Copilot Studio is weak." For the job it was built to do, it's arguably the strongest option on the market, and a mid-market IT team already paying for Microsoft 365 should take that seriously before looking elsewhere for internal use cases.

Native Microsoft 365 grounding

Copilot Studio agents can ground answers directly in SharePoint document libraries, Dataverse tables, and Microsoft Graph, without a separate ingestion pipeline. An HR policy stored in SharePoint, an approvals table in Dataverse, and an employee's own calendar and mailbox metadata via Graph are all reachable natively. For a company whose institutional knowledge already lives in Microsoft 365, that's a real head start, and it's the strongest case for keeping employee-facing knowledge tools inside the platform rather than duplicating content elsewhere. See how that plays out for internal support in our guide to an AI chatbot for employee questions.

Entra ID identity out of the box

Every internal agent inherits your existing Entra ID identity and access controls. There's no separate user database to stand up, no parallel login flow to secure, and permissions on the underlying SharePoint or Dataverse content carry through to what the agent can retrieve and surface. For a regulated or security-conscious IT org, that's a meaningful reduction in attack surface compared with a third-party tool holding its own credential store.

Power Platform connectors and low-code actions

Copilot Studio taps directly into the Power Platform's connector ecosystem, letting an agent trigger a Power Automate flow, write to Dataverse, or call an internal API without custom backend code. For internal workflow automation (an IT ticket, a leave request, an approvals chain), that connector depth is hard to replicate outside the Microsoft stack.

Centralized admin governance

Admins manage agents, data loss prevention policies, and environment access from the same Power Platform admin center already used to govern the rest of the tenant. Security and compliance teams that have already invested in Microsoft's governance model get a single pane of glass instead of a separate tool to audit.

Taken together, these four strengths explain why Copilot Studio is a strong default for internal agents specifically. The question this article is built around is what happens when that same tool gets pointed at a channel none of these four strengths was built for: an anonymous, public website.

Where Copilot Studio Struggles on a Public Website

The friction points below are less about capability and more about fit. Every one of them traces back to the same root cause: Copilot Studio's defaults assume a licensed, internal user, not an anonymous website visitor.

Cost predictability at anonymous scale

A metered, per-credit model is manageable when usage comes from a bounded set of employees. It's much harder to forecast when usage comes from an unbounded public audience, where a single viral moment, a paid ad campaign, or seasonal traffic can multiply conversation volume overnight. Because credits don't carry over and overage triggers enforcement, a traffic spike on a public site risks either a larger-than-planned bill or the agent going dark at the worst possible moment.

Branding and widget control

Copilot Studio's channel publishing was designed primarily for internal surfaces (Teams, Microsoft 365 Copilot, a custom website channel) rather than a fully brand-matched customer-facing widget. Getting a public embed to look, sound, and behave like a native part of your marketing site, rather than an obviously bolted-on Microsoft component, takes real customization work that a purpose-built customer-facing widget handles by default.

Setup effort for a non-Microsoft use case

Standing up an internal Copilot Studio agent is fast when the knowledge already lives in SharePoint or Dataverse. Standing up a public-facing agent that needs to answer from a product catalog, a pricing page, or marketing content that was never structured for Microsoft's connectors is a different, heavier lift, closer to a custom integration project than a same-day configuration.

Knowledge grounding outside Microsoft sources

Copilot Studio's strongest retrieval paths run through Microsoft's own ecosystem. A public website's real knowledge base (PDF spec sheets, a Shopify or WooCommerce catalog, scattered marketing pages, a support FAQ hosted outside Microsoft 365) sits outside that ecosystem by default, and connecting it cleanly takes more configuration than a tool built to ingest arbitrary business documents natively. For the mechanics of what "grounded" actually means, see our business guide to RAG.

None of this makes Copilot Studio a bad product for the public web; it makes it a product optimized for a different job. The same logic that makes a company choose a CRM for sales and a separate CMS for the website applies here: use the tool built for the surface it's actually serving.

The Hybrid Architecture Most Companies Land On

In practice, most mid-market IT and marketing teams don't pick one tool and retire the other. They split by surface: Copilot Studio stays inside the Microsoft 365 perimeter for employee-facing agents, and a lightweight, RAG-native chatbot runs on the public website for customers and prospects.

The split maps cleanly onto the differences described above. Internally, Entra identity, SharePoint grounding, and Power Platform connectors are exactly what an employee-facing agent needs, and the consumption model is easy to forecast against a known headcount. Externally, a flat-fee, no-code RAG chatbot avoids per-credit exposure to anonymous traffic, gives marketing full control over branding, and ingests a product catalog or a support FAQ directly without a Microsoft-shaped integration project.

Heeya is built for that second half of the split: upload PDFs, Word docs, spreadsheets, or point it at your website, and it answers customers from your own content on a flat monthly fee rather than a per-interaction meter, with a widget styled to match your site and a lead-capture or booking tool built in. It has no ambition to replace Copilot Studio inside Teams or SharePoint; the two sit on opposite sides of the same firewall, each doing the job it was designed for.

This isn't a novel pattern specific to Microsoft. Any organization that standardizes on one platform for internal tooling and a separate, purpose-built tool for the public-facing equivalent runs the same playbook, and our build vs. buy decision guide for custom AI chatbots walks through the same logic for teams evaluating whether to build the public layer in-house instead.

Decision Grid: Which One for Which Job

The table below lines up the two tools against the questions that actually decide the outcome for a given use case.

If your knowledge lives in Microsoft and your users already sign in, Copilot Studio wins on almost every axis. If your audience is anonymous and your knowledge lives in ordinary business documents, a RAG-native chatbot wins on cost predictability and setup speed. For a wider look at how Copilot Studio and a purpose-built agent differ conceptually, see AI agent vs chatbot: key differences.

Migration and Coexistence Notes

Teams rarely need to choose once and live with it forever. A few practical notes for running both, or moving between them.

Running both without duplicating work

Keep a single source of truth for each piece of content. Internal policy documents stay authoritative in SharePoint and feed Copilot Studio directly. Public-facing content (pricing, catalog, FAQ, support docs) stays authoritative on the website and feeds the RAG chatbot. Avoid maintaining the same document in two places for two different bots; that's where content drifts out of sync and one bot starts giving a stale answer.

If you already deployed Copilot Studio externally and want to pull it back

Retire the public-facing channel in Copilot Studio, keep the internal agent live, and stand up a separate public chatbot pointed at your marketing content. Because the two tools don't share a knowledge base or a conversation history, this is a parallel cutover rather than a data migration: point the new tool at your existing documents and website, test it against your real FAQ traffic, then swap the embed script on the site.

Budgeting for the internal side going forward

Track Copilot Credit consumption against Microsoft's usage estimator before committing to a capacity pack size, and revisit it after a quarter of real internal usage, since Microsoft has changed the metering model more than once and is likely to keep refining it. Our AI chatbot pricing breakdown covers how consumption-based and flat-fee models compare more broadly if you're building the business case either way. Copilot Credits are Microsoft's version of the same consumption billing that Intercom Fin, Zendesk, and Salesforce Agentforce charge per resolved conversation; see our per-resolution AI chatbot pricing breakdown for the worked math on those.

Where the two can connect

Some IT teams route a subset of public-site questions (an internal policy question a customer shouldn't see, a request that needs employee-only routing) to a human or an internal system rather than trying to force one bot to serve both audiences. That's a reasonable pattern: keep the public bot public, and use standard handoff tools like a contact form or a connected CRM (Heeya integrates with HubSpot) rather than exposing internal Copilot Studio agents to anonymous traffic.

FAQ: Copilot Studio vs. a Public RAG Chatbot

Final Verdict

Copilot Studio and a public-facing RAG chatbot are not competing for the same job, even when a team is told to pick one. Copilot Studio's Entra identity, native Microsoft 365 grounding, and Power Platform connectors make it a genuinely strong default for internal, employee-facing agents, and a mid-market company already paying for Microsoft 365 should use that advantage rather than duplicate it elsewhere.

The public website is a different surface with different requirements: anonymous traffic instead of authenticated employees, brand-matched presentation instead of an internal app shell, and a bill that needs to stay flat regardless of how a marketing campaign performs. That's the gap a purpose-built RAG chatbot fills, and it's why the realistic end state for most Microsoft 365 shops is both tools, each running the surface it was built for, rather than one tool stretched across both.

Further Reading

• AI Chatbot for Employee Questions
• HR Chatbot vs Intranet FAQ
• AI Agent vs Chatbot: Key Differences
• ChatGPT vs Custom RAG Chatbot
• Custom AI Chatbot: Build vs Buy
• What Is RAG? A Business Guide