Preview of the new creaid.ai. Not live yet. Yellow = to confirm with you.

AI in CRE finance · 7 min read

How AI Lender Matching Works

Short answer

AI lender matching takes a description of your deal, turns it into structured criteria (property type, location, loan amount, loan type, loan-to-value and so on), filters a database of lenders by those criteria, and ranks the lenders that remain by how closely they fit. The result is a shortlist to contact. It is not an approval or a quote, and it is only as good as the lender data behind it.

A commercial bank building
A commercial bank building. Example photo: Joe Mabel, CC BY-SA 3.0, via Wikimedia Commons.

The three parts of a matching system

Every lender matching tool, whatever it looks like on the screen, has three parts.

  1. A lender database: records of lenders and what each one lends on.
  2. A deal profile: the facts of your request, in a form that can be compared with those records.
  3. Matching logic: rules and ranking that decide which lenders fit and in what order.

The AI part mostly sits in the second and third steps. A language model reads your plain-language description and pulls out the facts, and it can help rank and explain results. The first part, the database, is ordinary data work, and it matters most. Without it you have a general chatbot guessing at names, a problem described in the guide on using a general chatbot to find a lender.

What the lender database holds

A lender's credit box can be written down as a set of fields. The more of these a database records, and the more accurately, the better the matches.

FieldExample valuesWhy it matters
Lender typeBank, credit union, agency, life company, debt fund, privateSets typical terms and speed
Loan sizeMinimum and maximum loan amountThe most common reason for a pass
Property typesMultifamily, industrial, retail, office, hotel, storageMany lenders exclude some types
GeographyStates, metro areas, or nationwideBanks often lend near their branches
Loan typesPermanent, bridge, construction, SBAA permanent lender will not do a lease-up
LimitsMaximum LTV or LTC, minimum DSCR or debt yieldDecides whether your request fits
StructureRecourse or non-recourse, fixed or floatingHas to match what you can accept

These fields follow the real differences between the types of commercial real estate lenders. A local bank and a national bridge lender may both lend on apartments and have almost nothing else in common.

Turning your description into a deal profile

Older lender search tools made you fill in a form with dropdowns. With a language model you can write the deal the way you would say it to a colleague.

Say you type: I am buying a 40-unit apartment building in a mid-sized Texas city for $6,000,000. It is 94% occupied, NOI is $420,000, and I want a $4,200,000 loan with a fixed rate and no personal guarantee. The system extracts a profile from that: purchase, multifamily, 40 units, Texas, stabilized, a $4,200,000 loan, 70% loan-to-value ($4,200,000 / $6,000,000), 10% debt yield ($420,000 / $4,200,000), fixed rate, non-recourse preferred.

Two things can go wrong here. The model can misread something, such as treating the purchase price as the loan amount. Or your description can leave out a fact the match depends on, such as whether the property is stabilized. Good tools show you the profile they extracted or ask follow-up questions. Check it before you trust the list. It also helps to know your own numbers first, using how to size a commercial loan.

Filtering, then ranking

Matching usually happens in two passes.

Hard filters remove lenders that cannot do the deal. A lender whose minimum loan is $10,000,000 is out. So is one that does not lend in the state, or does not finance the property type. These are yes or no tests, and they do most of the work. From a database of well over a thousand lenders, the filters may leave a few dozen.

Ranking orders the lenders that remain. A system can weigh things such as these:

  • How central the deal is to the lender's range. A $4,200,000 request sits better with a lender whose range is $1,000,000 to $10,000,000 than with one that starts at $4,000,000.
  • Whether the lender type suits the deal: an agency or bank lender for a stabilized apartment building, a bridge lender for a vacant one.
  • How well the structure fits what you asked for, such as non-recourse or a fixed rate.
  • How much room there is between your request and the lender's limits on loan-to-value or DSCR.

Some tools also explain each match in a sentence. Read those explanations. They show whether the system understood the deal.

What a match is not

A match means the deal appears to fit the lender's stated criteria. It does not mean the lender will quote it.

  • Appetite changes: a lender can be full on a property type or a market this quarter, and no database sees that in real time.
  • Criteria are simplified: real credit decisions depend on the sponsor, the tenants, the condition of the property and the story. A record of fields cannot hold all of that.
  • Data ages: lenders change programs and people move. Any contact or criterion may be out of date.
  • No pricing: a matching tool does not issue rates. Pricing comes from the lender's quote.
  • No underwriting: the tool has not verified your numbers. The lender will.

So a match list is a starting point for calls. It replaces neither the lender's underwriting nor the judgment of an experienced broker about who is actually closing deals. Confirm all terms with the lender and your own counsel.

How to get better matches

The list you get back reflects what you put in. Six habits make it more useful.

  1. Give complete facts: property type and size, location, purchase or refinance, price or value, loan amount, occupancy, NOI, your experience, timing.
  2. Say what you must have and what you would merely like. Non-recourse as a requirement gives a different list than non-recourse as a preference.
  3. Describe problems honestly: vacancy, a short lease, a credit issue. Hiding them only produces lenders who will pass later.
  4. Run more than one version if the structure is open, for example a bridge loan and a permanent loan.
  5. Work the list in order, but do not stop at the first name. See how many lenders to approach.
  6. Record what each lender says. Their answers improve your next search, and for brokers they become part of the lender network.

Where CREaiD fits

CREaiD's Lender Matchmaker works this way: you describe a deal and it returns lenders that fit from a database of 1,500+ commercial real estate lenders, covering banks, credit unions, agency lenders, private lenders, hard money and private equity.

It has a free version. See the Lender Matchmaker.

Brian Gramlich
About the author

Brian Gramlich is the CEO and founder of Terrydale Capital, a commercial real estate finance firm based in Dallas, Texas, and the founder of CREaiD.

Key takeaways

  • Lender matching combines a lender database, a deal profile and matching logic.
  • The database matters most, because AI cannot match against data it does not have.
  • Hard filters remove lenders that cannot do the deal, and ranking orders the rest.
  • A match is a lead to call, not an approval or a quote.
  • Complete, honest deal facts produce better matches.

Questions people ask

Is AI lender matching the same as asking a chatbot for lenders?

No. A matching tool searches a structured lender database using the facts of your deal. A general chatbot without such a database writes an answer from its training text and can produce lenders that are outdated or do not exist.

Does a match mean the lender will approve my loan?

No. It means your deal appears to fit that lender's stated criteria. The lender still has to want the deal, underwrite it, order third-party reports and approve it.

How many matches should I expect?

It depends on the deal. A stabilized apartment building in a large market at a common loan size fits many lenders. A specialized property, a small market or an unusual structure fits few. A short list is useful information in itself, because it tells you the deal is hard to place.

Can matching tools show me rates?

A matching tool identifies who may lend, not what they will charge. Rates depend on the index, the lender's spread and your specific deal on the day it is quoted. Ask each lender for a quote.

What if my deal has a problem, like low occupancy?

Include it in the description. The match should then point toward lenders that handle transitional deals, such as bridge lenders, instead of permanent lenders that will decline. Leaving the problem out only delays the pass.

Find lenders for your deal in minutes

CREaiD shows the lenders that really made your kind of loan in your state, from public loan records.

Start free