Agentforce for Real Estate: 4 AI Agents That Pay for Themselves

Every real estate operator has now sat through an AI demo. The agent answers a question about a listing, the room nods, and nothing ships, because nobody in the room could say what the agent would be measured on or what it would cost per month to run.

This post answers both. It covers the four Agentforce agents we deploy for real estate teams, the one metric each has to hit before we call it a success, and what Salesforce actually charges to run an agent as of September 2026. If you cannot name the metric and the cost, you are not ready to fund the agent, and that is fine. Most AI projects that miss their return miss it because the metric was never written down.

What an Agentforce agent is, in operator terms

Agentforce is Salesforce's platform for AI agents that work inside your org. An agent has topics (the jobs it is allowed to do), actions (the things it can read or change in Salesforce, usually built with Flow), and guardrails (what it must never do without a human). It runs in Salesforce, on your data, under your permission model.

The difference from a chatbot is that an agent takes actions on records. It does not tell the salesperson to schedule a site visit; it creates the event, sends the confirmation, and logs it.

Agent one: 24/7 portal lead qualification

Who it is for: brokerages and developers receiving leads from portals, the website, and paid campaigns.

The problem: a portal lead that hits an inbox at 9 pm gets a reply at 10 am, if the salesperson is not at a site visit. By then the buyer has spoken to two other agents.

What the agent does: replies within seconds, asks the three or four questions that qualify the lead (budget, timeline, location, financing), creates or updates the Lead with the answers, and books a call or a site visit with the right person. Anything it cannot answer escalates to a human with the transcript attached.

The metric: speed to first qualified response, and the share of leads that reach a booked appointment. We built this pattern for a US commercial brokerage; the case study on this site covers how portal leads went from an inbox to a qualified record without a human retyping anything.

Agent two: listing and inventory inquiry

Who it is for: brokerages with active listings and developers with unit inventory.

The problem: the same questions arrive hundreds of times. Is unit 1204 still available. What is the price on the 2 BHK on floor 8. Is the warehouse on the north side still listed. Each one costs a salesperson a lookup and a reply.

What the agent does: reads Listing or Unit records and answers with live availability, price list version, and the next step. For developers, it respects holds: a unit on hold reads as unavailable until the hold expires. It never quotes a price that is not on the current price list, because it can only read the record.

The metric: inquiries resolved without a human, and the lag between a price list change and the first correct quote after it.

Agent three: lease and document intelligence

Who it is for: commercial brokerages, lease administrators, and property managers.

The problem: the terms of a lease live in a PDF. Critical dates, escalation schedules, renewal options, and CAM clauses are typed into a spreadsheet by an analyst, and the spreadsheet is wrong about one in every batch.

What the agent does: reads the uploaded lease, extracts the fields you define into the Lease record and its Critical Date children, and flags anything it is not confident about for a human to confirm. The human reviews exceptions instead of retyping every field.

The metric: hours of abstraction per lease, and the exception rate the reviewer has to touch.

Agent four: tenant service

Who it is for: property managers, residential and commercial.

The problem: tenant requests arrive by phone, email, and WhatsApp with no shared queue. Status questions generate repeat calls. Nobody knows the cost per work order because the vendor invoice and the request never meet.

What the agent does: takes the request on the channel the tenant used, creates the Case and the Work Order, dispatches to the right vendor by trade and SLA, answers status questions from the record, and closes the loop when the vendor marks it done. Anything involving a safety issue, a cost above threshold, or an angry tenant goes to a human immediately.

The metric: requests handled end to end without a human, and cost per work order once vendor cost is on the record.

What Salesforce charges to run an agent

This is the part most posts skip. The figures below are from Salesforce's published pricing as of September 2026; confirm them on Salesforce's own pricing page before you budget, because they change.

• Usage. Agentforce is metered. Flex Credits are sold at $500 per 100,000 credits, and a standard agent action consumes 20 credits, so roughly $0.10 per action. Conversation-based pricing is $2 per conversation.
• Seats. On September 3, 2026 Salesforce introduced Core, Advanced, and Max editions at $195, $395, and $550 per user per month, and each bundles Flex Credits into the seat. Existing Enterprise and Unlimited customers keep their pricing and add Agentforce through an add-on or Flex Credits.
• Prerequisite. You need a supported edition before an agent can go live. This is the single most common reason a pilot stalls in week one, so we check it in the free audit before anything is scoped.

Implementation is separate from all of the above, and ours is fixed price: an Agentforce Pilot in your sandbox is $1,500 for 7 days, and a production Launch is $4,500 over 2 weeks. Both numbers are on our pricing page.

Why most AI projects miss their return

Three reasons, in the order we see them.

1. No metric. The agent was funded on a demo. Without a number it cannot fail, which also means it cannot succeed, and the budget quietly disappears at renewal.
2. Bad data. An inventory agent on top of a spreadsheet-fed Unit object quotes wrong prices with total confidence. The data readiness check is not a formality; it decides whether the agent should be built at all.
3. No guardrails. An agent that can close an Opportunity or delete a Contact without confirmation will eventually do so. Guardrails are configured, tested, and signed off before go-live, not added after the first incident.

How we prove an agent works before it goes live

Salesforce ships a testing tool for agents. We write each test conversation with the topic, action, and reply we expect, run the batch in your sandbox, and hand you the pass and fail report. An agent that has not passed its scripted tests is a demo. An agent that has is a deployment.

Frequently Asked Questions

Which Agentforce agent should a real estate team start with?

Lead qualification, almost always. It has the clearest metric, the least data dependency, and the shortest path to a number the sales head cares about.

How much does Agentforce cost per month?

Usage is metered by Salesforce at published rates, currently $500 per 100,000 Flex Credits or $2 per conversation, on top of a supported seat. The exact monthly figure depends on volume, which is why we size it during the audit rather than guess.

Does an Agentforce agent need clean data?

Yes. An agent reads records and acts on them. If the records are wrong, the agent is confidently wrong. Data readiness is checked before an agent is scoped.

Can the agent work on WhatsApp?

Agents run on the channels your org is licensed for. WhatsApp requires an approved business messaging integration and is scoped as an add-on, not included in a standard pilot.

How long does it take to deploy a first agent?

A sandbox Pilot takes 7 days from the day we have access. A production Launch takes 2 weeks, followed by 30 days of fixes.