Many mid-market Salesforce admins and solution architects face a familiar constraint. Their company already runs a custom B2B portal or legacy customer site, yet leadership wants the same 24/7 autonomous support that Agentforce promises inside Salesforce. Buying and standing up an entire Experience Cloud instance feels like overkill when the rest of the web presence already exists.
The good news is that Salesforce designed Agentforce to work outside its own portal product. You can add a fully functional autonomous agent to almost any HTML-based site in a matter of days rather than months.
The Real Problem Teams Are Trying to Solve
You manage a customer-facing site that sits outside Salesforce. Support volume keeps rising, yet hiring more agents is not an option. The business wants customers to open cases, check order status, or update billing details without leaving the existing portal. At the same time, the IT roadmap does not include a full Experience Cloud migration this year.
Traditional chatbots on these sites usually fall short. They follow rigid scripts, fail on edge cases, and still require human hand-offs for anything beyond the simplest lookup. Autonomous Agent technology changes the equation because it can reason over your CRM data, run Flows, and decide when to escalate. The question becomes how to bring that capability to a site you already own.
Why a Lighter Embedding Approach Usually Wins
The standard route for customer self-service inside Salesforce is Experience Cloud with its dedicated chat component. That path delivers deep integration and polished theming, but it also brings licensing, site configuration, and ongoing maintenance.
A lighter path exists. Salesforce supports the Messaging for In-App and Web channel for any external website. You build the agent once in Agentforce Builder, connect it to a messaging channel, and drop a code snippet into your current portal’s HTML. The agent still reaches your Salesforce data, triggers actions, and maintains conversation history.
This approach saves time because you avoid provisioning a new portal product. It also keeps ownership of the surrounding user experience with your existing development team. Many teams report faster time-to-value when they start with the embed rather than a full site rebuild.
How the Embedding Process Actually Works
Begin in a sandbox where Einstein is already enabled. Assign the AgentforceServiceAgentBuilder and AgentforceServiceAgentBase permission sets to the users who will configure the agent.
Open Agentforce Builder and create a new service agent. Define the topics your customers will ask about, such as order tracking or billing updates. For each topic, write clear instructions and attach the Flows or prompt templates that perform the work. Test the agent thoroughly inside the built-in console before moving to production.
Next, create a Messaging for In-App and Web channel in Setup. Link the activated agent to that channel and configure an Embedded Service Deployment. The deployment wizard generates a small JavaScript snippet. Paste that snippet into the footer or help section of your existing portal. The widget appears exactly where you place it and inherits your site’s styling through CSS overrides if needed.
Authenticated users can be passed through single sign-on so the agent sees the correct contact record. Guest users are also supported, though you will want tighter Guardrails on what data the agent can access. Routing logic lives in Omni-Channel Flows if you need to hand conversations to human agents during business hours.
Practical Capabilities You Gain Right Away
Once embedded, the agent can create cases, look up orders, update contact details, and answer questions grounded in your Salesforce records. When you connect Data Cloud, the agent gains a unified customer profile that spans multiple systems. This grounding reduces hallucinations and lets the agent give personalized answers instead of generic replies.
The same agent can escalate to a live agent when a topic exceeds its configured scope. Conversation history persists across sessions, so returning customers do not repeat themselves. Pre-chat forms collect context before the agent even starts reasoning.
You also keep access to Agent Observability dashboards. These show topic performance, escalation rates, and customer satisfaction scores so you can refine instructions over time.
Honest Limitations and Considerations
The embedded widget is still a chat interface. If your portal needs deeply custom visual components or complex multi-step forms inside the same pane, you may eventually want a more integrated solution. Licensing is another factor. Agentforce Service Agent licenses are sold separately from base Salesforce editions, and conversation-based pricing starts around two dollars per conversation at list price. Exact costs depend on volume and contract negotiations.
Data quality matters more than the deployment method. If your CRM records are incomplete or inconsistent, the agent will surface those gaps to customers. Spend time cleaning key objects before launch.
Finally, the approach works best on web-based portals. Native mobile apps require additional native SDK work that falls outside the simple embed path.
Decision Framework: Light Embed vs Heavier Custom Build
Choose the MIAW embed when your portal is already live in HTML, your main goal is autonomous chat for routine service tasks, and you want to validate value in under four weeks. It is also the right call when budget or timeline rules out an Experience Cloud project this quarter.
Consider a heavier custom build or full Experience Cloud deployment when you need pixel-perfect branding that matches complex design systems, when the agent must control large parts of the page beyond chat, or when regulatory requirements demand on-premise or air-gapped infrastructure. In those cases the extra investment in a purpose-built portal becomes justified.
Another trigger for heavier work is multi-channel orchestration at scale. If the same agent must seamlessly move between web, SMS, and voice with shared memory, the additional setup in Service Cloud and voice channels may require more architecture than a simple embed provides.
Getting Started This Week
Confirm your org edition supports Agentforce and that Einstein is enabled. Create a dedicated sandbox user with the required permission sets. Build and test one focused topic in Agentforce Builder, then activate it. Configure the messaging channel and drop the embed code into a staging version of your portal. Run end-to-end tests with both authenticated and guest users before promoting to production.
Monitor the first conversations through the observability tools and adjust topic instructions based on real usage. Many teams iterate weekly for the first month.
Want help implementing Agentforce on your existing portal? Book a meeting to discuss your needs. You can also explore support options on our Support Portal.
Read Next
Agentforce in Experience Cloud Portals: The Self-Service Revolution
Salesforce Agentforce agents integrated into Experience Cloud portals are shifting self-service from static FAQs to proactive, autonomous support that resolves most issues without human intervention.
Beyond Basic Bots: How to Connect Agentforce with Your Existing Business Systems
Learn practical strategies for integrating Agentforce with your existing business systems using MCP, APIs, and custom connections for seamless AI automation.
