Private AI: Keeping Deal Data Away From Public Models
Your brokers are already using AI. The question is whether merchant statements are being pasted into tools you do not control.
Ask around your floor and you will probably find it: a broker who copies a merchant's statement summary into a free chatbot to get a quick read on average daily balance, or pastes a call note in to draft a follow-up email. It works, it is fast, and nobody told them not to.
The problem is where that text goes. A merchant's name, revenue, account activity and sometimes identity details have just left your systems and landed with a third party whose data handling you have never reviewed and the merchant never agreed to. Multiply that by a dozen brokers and a few months, and a meaningful slice of your merchants' financial lives is sitting somewhere you cannot see.
AI is genuinely useful in a funding shop. This guide covers the specific risks of public tools, what a private setup changes, and how to set a policy your team will actually follow.
Pre-flight summary
- Pasting merchant data into public AI tools moves it outside your controls, logs and retention policy.
- Funding files combine identity and banking data, which makes uncontrolled AI use especially risky.
- Private AI keeps processing within boundaries you define and can be audited like the rest of your CRM.
- AI figures should link back to their source so a person can verify them.
- Give brokers an approved tool and a clear policy rather than a ban they will work around.
Leg 01What actually happens when someone pastes a statement
When a broker pastes data into a public AI tool, that data is transmitted to the provider and processed on their infrastructure. Depending on the product, account type and settings, it may be stored, reviewed or used to improve future models. Terms vary between providers and change over time, and a broker using a personal account has almost certainly not read them.
Even when a provider promises not to train on your data, you still have a copy of merchant information sitting outside your access controls, outside your audit trail and outside your retention policy. You cannot delete what you do not know exists, and you cannot tell a merchant where their information went.
Leg 02The specific risks for a funding shop
Funding data is a particularly bad fit for uncontrolled tools, for reasons that go beyond general privacy.
- Identity exposure: applications carry Social Security numbers, dates of birth and home addresses.
- Banking exposure: statements contain account numbers, balances and counterparties.
- Contractual exposure: funder agreements may restrict how you share merchant information.
- Regulatory exposure: privacy and data-security rules may apply; confirm which ones with qualified counsel.
- Accuracy exposure: a general chatbot can confidently misread a statement, and nobody checks its arithmetic against the source page.
Leg 03What private AI changes
Private AI means the model runs in an environment you control or that is dedicated to you, rather than a shared public service. Requests and data stay within boundaries you define, governed by the same access rules as the rest of your CRM.
That changes the questions you can answer. Who used the AI on this deal? What data did it see? Where is that data stored, and when is it deleted? With a public tool used ad hoc, those answers are unknowable. Within your own system, they are part of the record.
FundPilot offers an optional private AI deployment on private hardware for clients who want deal data to stay off shared public models entirely. Its assistant answers from live CRM data and can draft applications, emails and texts, but it never sends or saves anything without a person confirming.
You cannot delete merchant data you do not know left the building.
Leg 04Accuracy matters as much as privacy
Privacy is only half the issue. A broker who pastes a statement into a chatbot and gets back an average daily balance has no way to check where that number came from. If it is wrong, the deal gets priced or matched on bad information, which is unfair to the merchant and embarrassing with the funder.
Underwriting assistance should show its work. In FundPilot, statement analysis links each figure, such as deposits, balances, NSFs and existing positions, back to the page it came from, so anyone can click through and verify it. Consider a hypothetical: the tool flags a recurring debit as a possible existing position. A broker clicks the figure, sees the actual line on page three, and confirms it before discussing it with the merchant.
Leg 05Writing an AI policy your floor will follow
Banning AI outright tends to fail; people quietly keep using it. A better approach gives brokers a sanctioned option and draws clear lines around everything else.
- Ask your team, without blame, how they are using AI today.
- Provide an approved AI tool that works inside your controlled systems.
- State plainly that merchant names, statements, applications and identity data never go into unapproved tools.
- Require a person to review and confirm anything AI drafts before it is sent to a merchant or funder.
- Include the AI policy in onboarding and revisit it whenever tools change.
- Ask counsel whether your policy should address any specific legal or contractual obligations.
Questions from the floor
Is it ever acceptable to use a public AI tool at work?
Public tools can be fine for tasks that involve no merchant or company-confidential data, such as brainstorming a generic email structure. The line is the data itself: no merchant names, statements, applications, identity details or deal terms. Put that boundary in writing so brokers do not have to guess, and give them an approved tool for anything that touches real deals.
Does private AI mean the AI makes decisions on deals?
No. In FundPilot the assistant answers questions from live CRM data and drafts applications, emails and texts, but a person confirms before anything is sent or saved. Underwriting figures link back to source pages for verification. Decisions about whether and how to present a deal stay with your brokers and the funders who review it.
What should we ask an AI provider before using it with deal data?
Ask where data is processed and stored, whether it is retained after a request, whether it is used to train models, which staff at the provider can access it, how it is deleted on request and what happens to it if you stop using the service. Get answers in writing, and have counsel review any contractual terms that touch merchant data.
General information only — not legal, tax or financial advice. Rules for telemarketing, texting, recording, disclosures and data retention vary and change; confirm current requirements with qualified counsel.
Connecting flights