Almost every AI underwriting tool we cover is built for the borrower’s side of the table: the GP, the sponsor, the broker pricing a deal before they take it to a lender. LenderBox is aimed at the other side. It’s built for the banks and private credit shops writing the loans, the people who have to say yes, no, or renegotiate, and it’s the first tool we’ve broken down that lives entirely in the lender’s seat.
Who built it
LenderBox is founder Vijay Mehra’s second act. He spent two decades split between CRE and technology, most recently as founder of Rethink, a CRE deal management platform he bootstrapped over 14 years to a private equity exit in 2021. He has sat as a GP, an LP, and a borrower, and it was the borrower seat, watching how inefficient bank and private-lending underwriting actually was, that pushed him into building this. He spent about 12 months on it with five pilot customers and a six-person engineering team, and the company came out of stealth on January 27, 2026. So this is a five-month-old product, publicly, and worth reading with that in mind.
What it actually does
The center of the product is a deals view built like a CRM: filterable by asset type, loan type, and market, with a map layer that pulls in third-party LTV and interest-rate trend data. A chat interface sits on every screen. Lenders upload their own credit and loan policies into a knowledge base and query them directly, including a compliance check that runs live deals against those policies. That last part is the piece a borrower-side tool can’t replicate, because it’s grounded in the lender’s own rulebook.
Borrowers log into a portal and upload documents against a checklist that’s customizable by institution and loan type. Once a document lands, extraction starts. LenderBox says its data dictionary covers more than 70 CRE document types and extracts over 6,000 data points per deal, with page-level source citations on every field and a stated 99.9 percent accuracy rate.
That citation trail is the thing we keep telling people to demand from any extraction tool. If a number can’t show you the exact page it came from, you have to re-verify it by hand, and the tool hasn’t actually saved you the work. LenderBox says every field traces back. The 99.9 percent accuracy figure is the kind of number we’d want to watch prove itself on a genuinely messy file rather than take at face value, but the architecture is pointed the right way.
Risk flags and policy-compliance issues surface within 30 to 90 seconds of upload, each tied to a source citation. A separate scoring model runs per asset type, loan type, and document type, returning an approve, reject, or unknown call along with a stated risk level. Mehra calls this “the power of 10 PhD underwriters in your pocket.”
The newest piece: structuring, not just reading
The part released the same week as our demo is the most interesting. It’s a deal structuring module built to replace Excel for the underwriting itself, not just to feed a model someone else built. Loan terms auto-populate from the extracted documents, every calculated field carries its formula, and lenders can save multiple versions of a model as terms change. It runs sensitivity and stress tests (raise the rate 200 basis points, drop NOI 25 basis points, and see where the deal breaks), a Monte Carlo simulation, and a report generator LenderBox says can produce a fully branded output in under 60 seconds. An admin portal lets clients retrain the underlying models in natural language.
This is the difference between a tool that reads your documents and a tool that does the underwriting. Most extraction products stop at populating fields. Structuring, versioning, and stress-testing inside the same system is where the real time savings would live, if it holds up.
Where the claims get loose
LenderBox’s own marketing says the platform compresses “25+ hours of manual underwriting into 35 minutes,” triples deal volume for the same headcount, and improves what it calls “process health metrics” by “32% to 94%.” That last one is a range wide enough, on an undefined metric, that it reads more like a marketing spread than a measured result. Treat the time-compression and volume numbers as vendor claims worth pressure-testing on your own files, and ignore the “process health” figure until someone can tell you what it’s actually counting.
Security and compliance
For a tool asking banks to pipe loan files through it, this is the section that decides the deal, and it’s where LenderBox is strongest on paper. It’s SOC 2 Type II certified, audited by Securance Pro Assurance PLLC over an observation period from November 2025 through February 2026. The platform states 256-bit AES encryption, GLBA compliance, TLS 1.2-plus in transit, AWS KMS-managed keys under FIPS 140-2, and a no-cross-training policy across client data. The advisory board includes a VP of Lending at Texas Heritage National Bank, an EVP at Guaranty Bank & Trust, a former managing director at Santander Bank, and a partnerships lead at Plaid. Mehra positions SOC 2 Type II as a requirement for bank clients and a nice-to-have for private credit shops, which tracks with how those two buyers actually procure.
Pricing and onboarding
Onboarding starts by migrating a lender’s historic, closed loan files to build the initial intelligence layer. Mehra says that takes a few weeks rather than the three to six months typical of a bank SaaS rollout, with staff trained in about a day. Pricing runs on a one-time activation fee scaled to portfolio size and document volume, plus either a flat monthly fee tied to deal volume or a pay-as-you-go, outcome-based model, with no annual or multi-year contract requirement. LenderBox doesn’t publish dollar figures anywhere, so price is the open question you’ll have to surface yourself. It goes through sales@lenderbox.ai or a booked demo.
Who it’s for
LenderBox is built for banks and private credit teams underwriting commercial loans, not for the brokers or GPs pricing them. If you sit on the lending side and you’re still underwriting in Excel off documents someone extracts by hand, this is squarely aimed at you, and the policy-compliance check plus the new structuring module are the two features that separate it from a generic extraction tool. If you’re a broker or a sponsor, it’s not your seat, though it’s worth understanding what your lenders may soon be running your deals through. The roadmap includes an Excel plug-in and an expansion into C&I lending, both weeks out as of the demo.
You can see the platform for yourself at lenderbox.ai.
We break down a tool like this every week and actually run them on real CRE deals. If you want the running list of what’s worth paying for, that’s what the AI for CRE Collective is built around: operators testing and sharing what holds up. Join here: https://www.skool.com/ai-for-cre-collective/about
