Agentforce reads the records you actually have. If half your deals closed without a comp and four pipelines were flattened into one, it will answer confidently and wrongly. Twopir Consulting scopes AI on Rethink by Buildout after the operating data is trustworthy — never before. Grounded in your CRE data, governed by your sharing model.
Trusted by 500+ organizations — including property and real estate businesses running their operations on Salesforce with Twopir Consulting.
AI Delivery Practice
An agent does not know what is missing. It answers from what it can see, in complete sentences, with no signal that the underlying record was half empty. Every failure below is a data condition rather than a model limitation.
Rethink creates a comp at close only when the deal carries a Property, and a Space or Unit for lease deals. Years of closed deals may have generated nothing. Ask an agent about market rates and it answers from the fraction that exists — confidently.
If Buyer, Seller, Tenant and Landlord Rep all share a stage list, "how long does a deal take" has no meaningful answer. The agent will still produce one, because the field it reads contains a number.
Owners imported as free text instead of real Company and Contact records. Ask which assets an owner holds and the answer depends on spelling. This is a migration decision made years earlier, surfacing as an AI problem.
Without email and calendar capture, the activity history reflects what brokers remembered to log. An agent summarising "what has happened on this deal" is summarising the logging habit, not the relationship.
An agent runs as a user and sees what that user's permissions and sharing rules allow. Point it at an org where desks were never properly walled off and it becomes a very efficient way to surface a confidential assignment to the wrong broker.
An agent gets switched on, produces plausible output, and nobody can say whether it helped. Without a measurable task and a baseline, an AI pilot cannot fail honestly — which means it also cannot succeed.
Agentforce is Salesforce's platform capability for building AI agents that act on your own CRM data. An agent reasons over the records in your org, takes actions through the automation the platform already has — Flow and Apex — and escalates to a person when it should not decide alone. It is built and governed by Salesforce; Twopir designs, configures and governs what yours does.
The part that matters for commercial real estate is where the agent gets its permissions. An agent runs as a user with its own permission set, which means your existing sharing model and field-level security apply to it exactly as they apply to a person. It is not a separate system with its own view of the data, and it cannot see past a sharing rule you already trust.
On Data Cloud — the question everyone asks. For agents that read and write ordinary CRM records, a Data Cloud modelling project is not a prerequisite: record access, Flow and Apex actions, and prompt templates that reference record fields cover it. Data Cloud becomes necessary when an agent must answer from unstructured content — a library of leases, offering memoranda or inspection reports. Most CRE agents worth building first are in the first category, which is why we usually recommend starting there.
Last reviewed September 2026. Agentforce capabilities and commercial terms change often — confirm current detail against Salesforce's own documentation before making a commitment.
This is the clearest practical argument for Rethink by Buildout over a closed commercial real estate tool, and it only became obvious once platform AI arrived.
| Question | CRE data inside Salesforce | CRE data in a closed tool |
|---|---|---|
| Can platform AI read it? | Yes. Deals, Properties, Spaces, Comps and Inquiries are ordinary records in your org. | Only if the vendor built and shipped an AI feature for it. |
| Who controls what it sees? | You do — your sharing model and field-level security, unchanged. | The vendor's own permission model, whatever that is. |
| What can it act on? | Anything your Flow and Apex can already do, including other systems you integrate. | The actions the vendor exposed. |
| Whose roadmap are you on? | Salesforce's platform roadmap, which you also get for every other business object. | One vendor's, for one product. |
| What happens to your data? | It stays in your org, under your existing data governance and retention. | Depends on the vendor's terms. |
This is an argument about architecture, not about Buildout. Rethink is a managed package inside your Salesforce org, so your CRE data is Salesforce data — and everything the platform gains, your brokerage gains too, without waiting for it to be built twice.
Every one reads records you already have and stops at a human before anything reaches a client or changes money. Start with the read-only ones — they earn trust before they need it.
| Pattern | Reads | Does | Human checkpoint |
|---|---|---|---|
| Deal-desk Q&A | Comps, Properties, Spaces, closed Deals | Answers market questions in natural language — recent lease comps in a submarket, achieved rates by asset class. | Read-only. Nothing to approve, which is why it is the right first agent. |
| Inquiry qualification | Inquiry, requirement fields, on-market Properties | Enriches an inbound inquiry, matches it to fitting properties, drafts a first response and proposes an owner. | Broker reviews the match and sends. The agent never emails a prospect unattended. |
| Deal summarisation | Deal, Activities, Offers, Site Selections | Produces a current-state summary of a deal for a pipeline review or a handover. | Read-only, internal. Useful precisely because it is low-risk. |
| Landlord report drafting | Deal, captured Activities, Inquiries | Drafts the narrative around the activity report — what happened this period and what it means. | Broker edits and approves before it reaches the landlord. Always. |
| Lease-expiry prospecting | Space/Unit expiry dates, Contacts, Companies | Identifies upcoming expiries, prioritises them and drafts tailored outreach for each. | Broker approves the list and the messages before any send. |
| Data-quality agent | Deals, Properties, Spaces, Comps | Finds the gaps that break everything else — deals closing without comp-critical fields, unresolved owners, duplicate properties. | Proposes fixes for an admin to apply. Quietly the highest-value agent on this list. |
Note the order. The first agent we recommend is almost never the most impressive one — it is a read-only agent on data you already trust, because an agent that is wrong in week one does not get a week two, whatever it could have grown into.
We run these five checks before scoping any agent. Failing one is not a reason to abandon AI — it is a reason to fix that thing first, and every fix is worth doing regardless.
Breaks any agent that answers market questions.
The most common failure, and the least visible. If comps were never generated, an agent will answer from a fraction of your history and sound certain doing it.
Breaks anything involving cycle time or forecasting.
Four representation types need four honest pipelines. Flattened stages make every timing question unanswerable — and answerable-sounding.
Breaks ownership, portfolio and relationship questions.
Owners and tenants have to be real Company and Contact records, not text. Otherwise the answer depends on spelling.
Breaks confidentiality, which you only discover afterwards.
An agent runs as a user. If your sharing model does not already hold, an agent becomes an efficient way to surface something it should not.
Breaks the business case, and therefore the second project.
A pilot without a defined task and a baseline cannot fail honestly — which means it cannot succeed either, and the budget quietly disappears.
Agentforce gives you the controls. Deciding how tightly to set them is a business decision, and in commercial real estate — where one confidential assignment leaking matters — it is the decision that needs the most care.
Every agent starts read-only, whatever it is eventually meant to do. It earns write access after it has demonstrably been right on real records for long enough that people trust it. No agent gets a write scope on day one.
The agent gets its own permission set scoped to the objects and fields its task genuinely needs — not a copy of an administrator profile because that was quicker. Its access is reviewed like any other user's.
Anything that reaches a client, a landlord or a counterparty, and anything that changes money, goes through a person. This is a design rule rather than a setting, and it is the one we will not negotiate away to make a demo smoother.
What the agent was asked, what it read, what it did. Without a trail you cannot investigate a bad answer, and you cannot demonstrate to a client that their information was handled properly — which in CRE you will eventually be asked to do.
We will not build an agent on data we have told you is not ready. If the readiness checks fail, the honest sequence is to fix the data first — and the fixes are worth making whether or not you ever deploy an agent. Selling the agent anyway would be the profitable choice and the wrong one.
We will not give an agent autonomy over client communication. In commercial real estate the relationship is the asset. An agent that emails a landlord unattended is risking something that took years to build, to save a few minutes.
We will not pretend the commercial model is simple. Agentforce has several pricing structures — per conversation, consumption-based credits, and per-user licensing — and which is cheapest depends on how your agents are actually used. We will model it against your expected volumes, and you should confirm current rates directly with Salesforce rather than from any consultancy's website, including this one.
We do not publish AI outcome metrics, because we do not yet have evidenced ones for a Rethink engagement and an invented figure on this page would be the least defensible claim we could make.
Buildout's own customer story on Rethink CRM, cited here as product evidence — VREI is Buildout's customer, not a Twopir engagement.
Note what produced this: consolidation, not AI. The reported saving came from not having to hunt for information. That consolidated, trustworthy record is precisely the precondition an agent needs — which is the argument of this entire page.
Read Buildout's customer storyRebuilding lead management and sales operations on Salesforce — the kind of data foundation an agent has to stand on.
The transferable part is sequencing. Clean capture, resolved entities and enforced fields came first. Everything intelligent that sits on top of a CRM is downstream of that work, and no model compensates for skipping it.
Read the case studyRepresentative engagement shapes, not measured client results — included so you can recognise which conversation you are having.
No specific problem yet. The useful output is a readiness assessment and a shortlist of tasks worth automating — which is a better board answer than a pilot nobody defined success for.
Almost always a data problem rather than a model problem — usually comp coverage or flattened pipelines. We diagnose against the readiness checks before touching the agent configuration.
The best starting point in the cluster. A specific, measurable, read-only task with an obvious baseline — which is exactly the shape of a first agent that earns a second one.
Every consultancy currently has an AI practice. The useful question is which of them will tell you your data is not ready, and lose the bigger project by saying so.
The five checks come before any agent is scoped. If they fail, the recommendation is to fix the data — work that pays for itself whether or not you ever deploy an agent, and work we would rather do properly than skip to reach the interesting part.
Knowing that comps are created at close only when specific fields are populated is what lets us predict exactly which questions an agent will answer badly. Generic AI consultancies discover that in production, on a client call.
An agent runs as a user, so the sharing model becomes an AI safety control. In a brokerage where confidential assignments and walled desks are real requirements, that is the part we spend the most time on.
Not an AI shop learning Salesforce. Agents act through Flow and Apex, run as users, and respect sharing — which makes agent design a platform-architecture problem before it is a prompt problem.
A read-only agent on data you already trust, with a measurable task and a baseline. An agent that is wrong in week one does not get a week two — so the first one is chosen to be right, not to be impressive.
Agentforce is Salesforce's platform capability for building AI agents that act on your own CRM data. It is built and governed by Salesforce; what a consultancy does is design, configure and govern your specific agents. It works with Rethink data for a structural reason: Rethink by Buildout is a managed package inside your Salesforce org, so Deals, Properties, Spaces, Comps and Inquiries are ordinary records in that org. An agent reads them the same way it reads any other object, acts through Flow and Apex, and respects the same sharing model.
Not necessarily, and this is the most commonly misstated fact about Agentforce. For agents that read and write ordinary CRM records — which covers most of the CRE patterns worth building first — record access, Flow and Apex actions, and prompt templates referencing record fields are sufficient. Data Cloud becomes genuinely necessary when an agent must answer from unstructured content, such as a library of leases, offering memoranda or inspection reports. Because this detail changes as the platform evolves, confirm the current position against Salesforce's own documentation before budgeting for it.
Only if your sharing model would already allow it. An agent runs as a user with its own permission set, so record-level sharing and field-level security apply to it exactly as they apply to a person — it cannot see past a restriction you already have. The real risk is the inverse: firms that assumed their desks were walled off and discover, through an agent, that they never were. That is why permission modelling is one of our five readiness checks, and why every agent gets a scoped permission set rather than a copy of an admin profile.
Salesforce offers several commercial models — per conversation, consumption-based credits where each action draws down a balance, and per-user licensing. Which is cheapest depends entirely on how your agents get used: a small number of long, action-heavy conversations prices very differently from many short ones. We deliberately do not publish rates on this page, because published AI pricing has changed repeatedly and a stale number on a consultancy website is worse than none. We will model the options against your expected volumes, and you should confirm current rates directly with Salesforce.
With a read-only agent on data you already trust, doing a specific task with a measurable baseline. A deal-desk question-answering agent over comps and closed deals is usually the right first one: it is useful immediately, nothing it does is irreversible, and it surfaces data gaps safely rather than in front of a client. Resist starting with the most impressive-sounding agent. An agent that is wrong in its first week does not get a second one, regardless of what it could have grown into.
Then we fix that first and say so plainly, even though it is the less exciting project. The five readiness checks — comp coverage, pipeline integrity, entity resolution, permission model and a measurable first task — are run before any agent is scoped. Every remedy is worth doing on its own merits: enforcing comp-critical fields, separating flattened pipelines and resolving owners to real records improve your reporting and your operations whether or not an agent is ever deployed. Building an agent on data we have told you is not ready would be the profitable choice and the wrong one.
Not on anything we design. Every pattern we build stops at a human before something reaches a client, a landlord or a counterparty, and before anything that changes money. The agent drafts, prioritises and prepares; a broker reviews and sends. This is a design rule rather than a platform limitation — the technology would permit more autonomy, and in commercial real estate we think that is the wrong trade. The relationship is the asset, and it is not worth risking to save a few minutes of review.
The main product page: what Rethink by Buildout is, who it is for, and how far it can be pushed.
Do This FirstDeterministic automation is cheaper, more predictable and usually the right step before agents.
SiblingWhere comp coverage and entity resolution are decided — the foundations this page depends on.
SiblingThe Flow and Apex actions an agent invokes, and the integration layer it reaches through.
Parent ServiceThe practice-level view across CRM, operations and reporting for a CRE business.
Next StepAsk us whether your data is ready. We will give you a straight answer, including "not yet".
A readiness assessment runs the five checks against your org and comes back with a straight answer: the agents worth building now, the data work that has to happen first, and what each would take. If the answer is "not yet", you will hear that.
Data first. Agents second. We will tell you which one you need.
