An AI agency is a firm you hire to design and build custom AI software and automations for your business, the way you would hire an architect to design a building rather than buying a house that already exists. That is the useful definition, and it comes with an immediate warning for a 4-to-20-person commercial real estate firm: most of the time you do not need one yet. The word “agency” also gets stapled to marketing shops that use AI to write copy, which is a different service entirely. This guide separates the two meanings, places a real AI agency next to the cheaper things you could hire instead, and names the specific conditions under which a small CRE firm should actually commission custom work.
What an AI agency actually is
An AI agency is a services firm that builds custom artificial-intelligence software for other companies. You describe a workflow, they scope it, design it, build it, and hand it over, the same relationship you would have with a firm that builds you a custom database or a custom website. What makes it an AI agency is that the thing they build is powered by a large language model or another AI system: a tool that reads your lease PDFs and pulls out the terms, an automation that drafts a first-pass market write-up, a workflow that consolidates a stack of rent rolls into one sheet.
The confusion starts because “AI agency” has a second, unrelated meaning in the wild. A lot of the results you find under that phrase are marketing agencies that use AI internally to produce content, ads, or SEO. They are not building software for you; they are selling you a marketing service and using AI to deliver it faster. That is a legitimate business, but it is not the thing a CRE principal means when they ask “who can build me a tool that reads my leases.” Keep the two apart: one sells you marketing output, the other builds you software you own.
The kind that matters for this article is the build shop. It typically brings a small team, engineers who know how to wire AI models into real workflows, a project lead who translates your process into a specification, and it works project by project rather than selling you a subscription. You are buying custom work, and the honest question is whether custom work is what your problem needs.
Product, agency, or consultant: the three things you can hire
A small firm with an AI problem has three genuinely different things it can pay for, and confusing them is the most common way money gets wasted. The table lays them out.
| What you hire | What it is | What you own at the end | Best when |
|---|---|---|---|
| AI product | A finished tool you subscribe to (ChatGPT, Microsoft Copilot, a proptech feature) | Nothing; you rent access | The task is common and a product already does it well |
| AI agency | A firm that builds custom software or automation for you | The software, and the job of maintaining it | No product fits a specific, valuable workflow and you can describe the result |
| AI consultant / fractional help | An advisor or fractional operator who trains your team and guides decisions | Skills and a plan, not software | You need to get fluent and decide what to build before building anything |
The distinction that trips people up is between the agency and the consultant. An agency builds and leaves you something to run. A consultant or fractional AI lead makes you capable and leaves you a decision. A firm that trains your brokers to use existing AI well is doing consulting work; a firm that ships you a lease-abstraction tool is doing agency work. Many small firms need the first before they need the second, and hiring an agency to build something you have not learned to specify is how projects go sideways.
None of these is automatically the right answer. The right answer is the cheapest one that actually solves the problem in front of you, which almost always means starting further up the list than “commission a custom build.”
Why a small firm exhausts the cheaper options first
The reason to reach for an AI agency last, not first, is that everything below it on the ladder costs less and burdens you less, and a small firm feels both of those more sharply than a large one. Work down the rungs in order and you spend an agency’s fee only when the problem genuinely requires it.
Rung one: buy a product. If the task is standard across the industry, storing comps, tracking lease dates, running a property ledger, a proptech product already does it, maintained by the vendor. You commission nothing. The full map of what is already for sale, and where the gaps between products sit, runs through our tour of the commercial real estate technology landscape.
Rung two: configure what you already pay for. Most serious CRE tools have custom fields, templates, and workflow rules you can shape to your process without writing code or hiring anyone. A firm that says “the CRM doesn’t do what we need” has often not exhausted what the CRM already does.
Rung three: train your people. A surprising amount of what looks like a software problem is a fluency problem. A team that knows how to prompt ChatGPT or Copilot for a lease summary, an LOI draft, or a market write-up handles by hand what you were about to pay to automate. A focused fluency workshop is the cheapest intervention on this list and often the highest-return.
Rung four: commission custom work. Only when a specific, valuable workflow survives all three rungs, no product fits it, configuration cannot close the gap, and a trained person still cannot do it in reasonable time, does an AI agency earn its place. The full decision, with the cost math behind it, is laid out in the buy-versus-build playbook for small CRE firms. The difference between renting a finished product and commissioning your own is the same fork covered in our guide to custom software versus off-the-shelf.
The discipline of doing less, but doing it exactly, is what lets a lean firm out-operate bigger ones, the argument in our manifesto on how 4-to-20-person shops out-operate institutional giants. Hiring an agency for a problem a workshop would have solved breaks that discipline and drains the budget you needed for the build that actually matters.
The four signals a CRE firm may actually need one
An AI agency becomes the right call when a workflow is valuable, repetitive, and unserved by anything you can buy or configure. Four patterns show up again and again at small CRE firms, and each is a fair reason to commission custom work.
A document workflow no product handles cleanly. Your firm receives leases, LOIs, and offering memoranda in every format imaginable, and someone re-keys the same fields out of them every week. Off-the-shelf abstraction tools exist, but if your asset class or your terms are unusual enough that none of them fit, a custom tool that reads your documents into your fields is agency work.
A data-consolidation job that eats a person’s week. Rent rolls from a dozen property managers, each in a different template, hand-merged into one investor sheet every month. No product consolidates your specific mess, but a scoped automation can, and the hours it returns are real.
A repetitive drafting task with a clear output. First-pass market write-ups, standardized deal summaries, tenant-communication drafts, work where the input is data you already hold and the output has a consistent shape. This is squarely in reach of a custom automation built on a current AI model.
A workflow that is genuinely your edge. The specific way you screen deals or underwrite a niche asset class, the thing a competitor cannot copy off a subscription. When the process is how you win, owning the software that runs it, rather than renting a generic version, can justify the cost and the upkeep.
If your painful workflow does not match one of these, an agency is probably premature. If it matches one, the next question is not “which agency” but “how do I hire one without getting burned.”
How to hire an AI agency without getting burned
Hiring a build partner at small-firm scale is a different exercise than the enterprise procurement most agency content assumes. You have no IT department to run what gets built and confidential deal data flowing through it, so a handful of criteria matter more than any portfolio.
Insist on a fixed, narrow first scope. The first project should be one workflow with one clear output, not a platform. A good partner will happily scope small; one that pushes you toward a sprawling build before you have shipped anything simple is optimizing for their revenue, not your risk.
Name who owns it on Monday. Any software someone builds you has to be run by someone at your firm after the agency leaves. Ask, before signing, who maintains it when a document format changes or an integration breaks, and make sure the answer is a real person or a standing relationship, not a hopeful silence.
Check data handling explicitly. Your files hold LOIs, financials, and deal terms that cannot leak. Ask where your data goes, which AI models process it, whether anything is retained, and get it in writing. A serious partner answers this crisply; vagueness here is disqualifying.
Prefer a reversible first project. A first engagement priced and scoped small enough that walking away costs you a modest sum, not your year’s technology budget, lets you learn how the partner works before you commit to anything larger. Reversibility is a feature, not a compromise.
A partner who welcomes a small, fixed, reversible first project with clear data terms is showing you how they will behave on a big one. That behavior tells you more than any case study.
What it costs, in real ranges
Cost is where agency shopping frightens small firms, because the numbers quoted publicly come from enterprise projects. At CRE-firm scale the ranges are far more human, and they line up with the ladder above.
Training your team to use existing AI well, a focused fluency workshop, runs roughly $2,000 to $15,000 in the current market. It is the cheapest way to close a gap, and often the first thing a good partner will point you to instead of a build. A scoped custom automation covering one workflow, the kind of thing an AI agency delivers as a first project, runs roughly $25,000 to $150,000 depending on complexity, with a sensible first engagement deliberately aimed at the low end.
Read those as direction, not a quote. The point is the shape: training costs a fraction of a build, a narrow build costs a fraction of a platform, and matching the spend to how much the workflow actually differentiates you keeps a small firm out of the six-figure mistakes that generic guides quietly assume you can afford.
Frequently asked questions
What is an AI agency?
An AI agency is a services firm you hire to design and build custom AI-powered software and automations for your business, then hand them over for you to own and run. It works project by project rather than selling a subscription, and the thing it builds is powered by an AI model, for example a tool that reads your lease documents into structured fields or an automation that drafts market write-ups. It is distinct from a marketing agency that merely uses AI internally to produce content, which is a different service that leaves you with no software.
What is the difference between an AI agency and an AI product?
An AI product is a finished tool you subscribe to and share with thousands of other firms, such as ChatGPT or Microsoft Copilot; you rent access and own nothing. An AI agency builds custom software for your firm alone, which you then own and are responsible for maintaining. A product is cheaper and someone else keeps it running, but it only does what it does for everyone; an agency’s work fits your exact workflow but costs more and leaves the upkeep to you. Most small firms should exhaust products first and reach for an agency only when no product fits.
Does a small commercial real estate firm need an AI agency?
Usually not yet. Most of what a 4-to-20-person CRE firm needs is already handled by an off-the-shelf product, by configuring a tool you already pay for, or by training your team to use existing AI well, all of which cost far less than commissioning custom work. An AI agency earns its place only when a specific, valuable workflow survives those cheaper options: no product fits it, configuration cannot close the gap, and a trained person still cannot do the job in reasonable time. The honest first step is figuring out which rung your problem actually sits on.
How much does it cost to hire an AI agency?
At small-firm scale, a scoped custom automation covering one workflow runs roughly $25,000 to $150,000 in the current market, depending on complexity, with a first project deliberately aimed at the low end. That is well below the six- and seven-figure numbers quoted from enterprise projects. Before that, a fluency workshop that gets your team using existing AI well runs roughly $2,000 to $15,000 and often removes the need for a build entirely. Treat both ranges as direction rather than a quote.
What is the difference between an AI agency and an AI consultant?
An AI agency builds you software and leaves you something to run; an AI consultant or fractional AI lead makes your team capable and leaves you a decision. Consulting work includes training your brokers to prompt AI tools well and helping you decide what, if anything, to build. Agency work includes designing and shipping a custom tool. Many small firms need the consulting first, because hiring an agency to build something you have not yet learned to specify is a common way projects drift over budget and miss the mark.
Can I just use ChatGPT instead of hiring an AI agency?
Often, yes, and you should try before commissioning anything. A team fluent in ChatGPT, Claude, or Microsoft Copilot can handle lease summaries, LOI drafts, and market write-ups by hand for zero build cost. The point at which a tool stops being enough is when a task is repetitive and high-volume enough that doing it by hand, even with AI help, still eats a person’s week, and when the output needs to be consistent and wired into your files automatically. That is where a custom automation, and possibly an agency, starts to pay for itself.
What should I look for when hiring an AI build partner?
Four things matter more than any portfolio at small-firm scale: a fixed, narrow first scope rather than a sprawling platform; a named person or standing relationship who maintains the software after the partner leaves; explicit, written answers on where your confidential deal data goes and which models process it; and a first project small and cheap enough that walking away is survivable. A partner who welcomes a small, reversible, clearly-scoped first engagement is showing you how they will behave on a larger one.
Is an AI agency the same as a marketing agency that uses AI?
No, and conflating them is the most common confusion around the term. A marketing agency that uses AI sells you a marketing service, content, ads, SEO, and uses AI internally to deliver it faster; you end up with marketing output and no software. An AI agency in the sense that matters for building tools designs and ships custom software you own, such as a document-reading automation or a data-consolidation workflow. When you search the phrase, check which one a firm actually is before you talk to them.
How do I know if my workflow is worth automating with a custom build?
A workflow is a candidate when it is valuable, repetitive, and unserved by anything you can buy or configure. Concretely, that looks like a document task no product handles cleanly, a data-consolidation job that eats a person’s week every month, a repetitive drafting task with a consistent output, or a workflow that is genuinely your competitive edge. If your painful task matches one of these and simpler options have failed it, a custom automation is worth scoping. If it does not, an off-the-shelf tool or better team fluency is the cheaper fix.
Where to start
The first move is not to shortlist agencies. It is to figure out whether you need one at all, which means placing your most painful workflow on the ladder in this guide: is it a standard task a product already handles, an 80%-right tool you could configure, a fluency gap a trained team would close, or the rare workflow that genuinely earns a custom build. A free AI-readiness assessment does exactly that triage with you: it identifies the workflow, tests whether a product or better fluency already solves it, and only points toward commissioning custom work when nothing simpler will do. Book a free AI-readiness assessment before you hire anyone.
Arthur Wandzel
