Every technology leader we talk to is living the same week.
The board wants an AI strategy by Friday. A vendor wants to sell you an agent that allegedly does everything short of refilling the coffee. And somewhere in your building, an analyst has quietly been running half their job through a chatbot and is praying you do not find out.
Meanwhile you are staring at the Salesforce investment you fought to fund. Does AI pay it off or make it obsolete?
Good news. It pays off. The answer is a triangle, and triangles are easy to remember at 8 am before a board meeting.
The Triangle You Can't Unsee
The Salesforce AI Stack: Three Layers, One Framework
Bottom Layer: Your Systems of Record
Core banking platforms, loan origination, and account opening. In real estate, your accounting and property platforms. These hold the truth. They are also allergic to plain-English questions and always have been.
Middle Layer: Salesforce
Where all that data finally agrees with itself and the part everyone forgets: where security lives. Salesforce already knows that Dave in the call center cannot see the CEO’s comp data. Pull Salesforce out of the middle, and you have just volunteered to rebuild access control by hand.
Top Layer: Intelligence
Where all that data finally agrees with itself and the part everyone forgets — where security lives. Salesforce already knows that Dave in the call center cannot see the CEO’s comp data. Pull Salesforce out of the middle, and you have just volunteered to rebuild access control by hand.
The industry name for this is headless Salesforce.
- The brains and the rules stay in Salesforce.
- The conversation moves to wherever your people already are.
The Part Nobody Wants to Hear
Point AI at messy data and it does not politely tidy up. It scales the mess.
Here is the upside. A dumb extractor reads a lease end date of “3099” and files it without blinking. A reasoning layer stops and says “that lease ends in the year 3099, you may want to check that.” Ask it for an average and it warns you three records are blank and will skew the math. Show it a customer who is apparently 120 years old and it raises an eyebrow instead of trusting the birthday.
Genuinely great. Also not a substitute for a foundation you never built. The triangle only stands on all three legs.
So What Is This MCP Thing
MCP is the connection between the top two layers. Officially, the Model Context Protocol is the emerging standard for how an AI tool talks to your Salesforce data.
Two ways to wire it:
The out-of-the-box connector. Copilot ships one. Log in, flip it on, about half an hour, no extra cost. The catch is a short list of blind spots. The well-known one: if a user gets access through a permission set group, the connector cannot see it. It never grants too much — it quietly hides things the user is allowed to see, which is arguably worse, because now your answers are confidently incomplete.
The custom MCP connector. What you build when those blind spots apply to you, when you live and die by permission set groups, or when your tool has no native connector yet. Light coding, heavy testing, because it is your data and it has to respect every single user’s permissions. Langley is Fortimize’s clearest example of why custom MCP is sometimes the only real path — their permission architecture made the standard connector unworkable.
BEFORE YOU GO ANY FURTHER
The Security Detail That Cannot Be Skipped
Set this up without care, clicking through the default steps, and you can hand your entire AI user base full admin access to all of Salesforce instead of access scoped to each person.
Shockingly easy to do.
Significantly less fun to discover three weeks later. Done right, the AI sees exactly what each user sees and not one record more.
On cost: the MCP connection on the Salesforce side is API usage. No separate charge today. Your real cost lives in whatever intelligence tool you chose.
What This Means for Banks and Credit Unions
The efficiency ratio is the number every institution is held to, and AI is the first tool in a long time with a real chance of moving it without adding headcount.
Agentforce surfaces relationship gaps before they become service failures. Copilot brings that context into Teams and Outlook where your bankers already work. Claude via MCP connects Salesforce to core banking platforms the standard connectors cannot reach. People’s Bank Indiana landed on Copilot. WestStar Bank is building toward a single customer view across seven systems with Agentforce on top. The platform decision is secondary to the foundation beneath it.
What This Means for Leasing Operations
Your leasing team should be able to ask about a renewal, a vacancy, or a tenant’s open legal matter and get an answer, not open four systems and build a report.
Agentforce surfaces that context in real time. New Quest is live on exactly this, with upcoming renewals and active legal matters tied to specific tenants surfacing without anyone pulling a report. Tanger is running Copilot across Salesforce, Yardi, and people data in one place. The VTS renewal window and the Yardi contract cycle are the moments when the foundation conversation gets real. Fortimize has watched the firms that treat those windows as a forcing function make the transition cleanly. The ones who wait absorb the cost.
How to Start Without Setting Anything on Fire
- Read before write: Let people ask questions of data they already see. The executive who pings an analyst for a report. The relationship manager who wants to know what is renewing next quarter. Low risk, high return. Write-back can wait.
- Check the base: Is the data flowing into Salesforce and clean enough to trust? If not, that is step zero. Skipping it is precisely how mess-at-machine-speed happens.
- Run the security analysis: Do the connector's blind spots hit your organization? If no, turn it on. If yes, build the custom one.
- Wire it so it respects each user's permissions: This is the one step where moving fast is the wrong instinct.
- Expand once it is proven: Then add the contextualized answers and the workflow kickoffs.
- Pilot small: Pick one front door. Five users, not five hundred. Do not run agents in two tools at once or you end up with two half-trained interns instead of one good one.
The Bottom Line
Your Salesforce investment is not what AI competes with.
It is what makes AI safe to switch on. The base keeps the data honest. The middle keeps it secure. The top finally makes it usable by the people who should not have to build a report to learn what is renewing on Tuesday.
Get the order right and AI becomes the best hire you made all year. Get it wrong and it becomes a very fast, very confident liability.
Fortimize works exclusively with banks, credit unions, and commercial real estate organizations. If you are working out where AI fits on top of your Salesforce, that is the exact problem we solve.
