Before you sign another proptech contract or request a custom software quote, answer one question: is the workflow you want to fix standard across the industry, or specific to how your firm makes money? Standard workflows belong to off-the-shelf vendors. Firm-specific workflows are where custom automation earns its price. Most buy-vs-build advice is written for companies with engineering teams; this playbook is written for the 4–20 person commercial real estate firm with no IT department, where the decision has a third option nobody mentions: wait, and build fluency first.
This is the anchor for our buy-vs-build and vendor economics cluster, part of the broader small CRE firm AI manifesto. By the end you will have a workflow-by-workflow decision matrix, 3-year cost math at 5, 10, and 20 seats, and a set of questions that settle the argument in one partner meeting.
Three Options, Not Two
The buy-vs-build debate presents a false binary. A small CRE firm has three moves, and the sequencing matters more than the choice.
Buy means subscribing to off-the-shelf proptech: Buildout for listing marketing, AppFolio or Yardi Breeze for property management, Dealpath for pipeline, Prophia for lease abstraction. You pay per seat, per unit, or per document, and the vendor owns the roadmap.
Build means paying a development partner to create automation specific to your firm: a pipeline that turns your lease stack into a rent-roll database, a screening model tuned to your acquisition criteria, an inbox triage system that routes tenant requests your way. Market pricing for this kind of scoped automation runs roughly $25,000–$150,000 depending on complexity.
Wait means neither. If your team cannot yet get consistent results from ChatGPT, Claude, or Gemini on everyday tasks like LOI drafts and lease summaries, you are not ready to evaluate vendors or scope a build. A team fluency workshop (market range: $2,000–$15,000) costs a fraction of either path and changes what you end up buying. Deloitte’s 2026 CRE outlook, surveying more than 850 executives at institutional owners and investors, found 27% still report AI implementation challenges tied to technical issues, lack of expertise, or resistance to change. Those are firms with IT departments. A 10-person shop that skips the fluency step is taking the same risk with fewer safety nets.
We see the sequencing mistake constantly: a firm buys an AI-branded subscription, usage collapses after month two because nobody trusts the output, and the tool joins the graveyard of per-seat licenses on the corporate card. The tool was rarely the problem. The sequence was.
When Off-the-Shelf Proptech Genuinely Wins
Buying wins more often than custom-development firms like ours would prefer to admit. Here is where the case for off-the-shelf is close to unbeatable.
The Workflow Is Standard Across the Industry
Listing syndication, tenant screening, rent collection, showing scheduling, email campaigns: thousands of firms run these workflows identically, which is exactly why mature products exist for them. Buildout’s AI assistant generates property and location descriptions from uploaded documents and pushes listings across its syndication network. AppFolio’s Realm-X agents respond to prospects, schedule showings, and dispatch pre-approved maintenance vendors. You cannot build a better version of these for less than the subscription costs, and you should not try.
The test: if you can describe your workflow to a peer at a conference and they nod because theirs is identical, buy.
The Economics Work at Your Head Count
Per-seat and per-unit pricing punishes scale, but a 4–20 person firm does not have scale. Yardi Breeze pricing starts around $1 per unit per month for residential portfolios with a $100 monthly minimum. Prophia offers per-document lease abstraction starting near $20 per document. At small-firm volume, these numbers are hard to argue with. A 300-unit management portfolio on Breeze costs less per year than one week of custom development.
The Vendor Carries the Maintenance Burden
This is the argument most buy-vs-build articles underweight for firms without technical staff. When CoStar changes its data format or a new lead source appears, your vendor ships the update. With custom software, someone has to own that change, and in a 10-person firm that someone does not exist unless your build contract creates them. Off-the-shelf transfers the maintenance problem to a company with hundreds of engineers. That transfer has real value even when the tool itself is mediocre.
You Have Not Proven the Volume Yet
Custom automation pays back through volume: documents processed, deals screened, emails triaged. If you abstract 15 leases a year, Prophia’s per-document pricing beats any build. If you underwrite eight deals a quarter, an analyst with a well-built prompt library beats a custom screening model. Buy first, measure the volume honestly, and let the usage data make the build case for you.
When Custom Automation Wins
Now the other side, and it is a narrower set of conditions than custom shops advertise. But when these conditions hold, off-the-shelf tools lose decisively.
The Workflow Crosses Systems That Do Not Talk
The most common breaking point for small CRE firms is not a missing feature. It is the gap between tools. Your deal flow lives in CoStar exports, your comps in Excel, your correspondence in Outlook, your leases in a shared drive of scanned PDFs. No single vendor owns that mess, so no single vendor fixes it. Every off-the-shelf tool assumes it is the center of your universe; your actual workflow runs through the seams between them.
Custom automation is at its best exactly here: a pipeline that reads lease PDFs from your drive, extracts the 40 fields your asset manager cares about, and writes them into the Excel model you already use. Our sibling playbook on turning lease stacks into structured data covers this pattern in depth.
The Workflow Is How You Make Money
Deal screening criteria are a good example. Every acquisitions shop screens differently: that is the edge. An off-the-shelf screening tool encodes someone else’s criteria, and configuring it to match yours usually means fighting the product. If a workflow is genuinely proprietary, encoding it in software you own compounds your advantage instead of renting someone’s average. The same logic runs through our deal analysis playbook: screening more deals with the same head count is a competitive weapon, not a convenience.
The inverse also holds. Rent collection is not how you make money; it is table stakes. Never build table stakes.
Per-Seat Pricing Has Become a Tax
SaaS pricing in the AI-feature era commonly runs $50–$500 per user per month once AI add-ons are included. At 20 seats and $200 per user, that is $48,000 a year, every year, forever. A one-time build in the $25,000–$150,000 market range with modest running costs can cross below the subscription line within two to three years. The math does not always favor building (we run it in the next section), but in our view the flip is reliable past a mid-range price tier and ten seats.
Confidential Deal Data Cannot Leave Your Control
Rent rolls, LP commitments, off-market pricing, tenant financials: small CRE firms handle information whose leak would cost real relationships. Some off-the-shelf AI tools have solid data agreements; others train on your inputs or route them through subprocessors you have never heard of. Reading those terms is its own job.
A custom system built on API access to frontier models can be configured so documents stay in infrastructure you control, with no training on your data, and the data-handling terms concentrated in one contract you negotiated. For firms where confidentiality anxiety is the blocker to AI adoption generally, this is often the deciding factor, independent of cost.
The Workflow-by-Workflow Decision Matrix
Applying those tests to the workflows that dominate a small CRE firm’s week produces a starting-point matrix. Your volumes will move individual rows, but in vendor evaluations this table settles most arguments quickly.
| Workflow | Default call | Why |
|---|---|---|
| Listing marketing & syndication | Buy (Buildout, Crexi) | Standard workflow, mature tools, network effects you cannot replicate |
| Property accounting & rent collection | Buy (AppFolio, Buildium, Yardi Breeze) | Table stakes; per-unit pricing is cheap at small scale |
| Market data & comps | Buy (CoStar, LoopNet, Placer.ai) | The data moat is the product; you cannot build the data |
| Lease abstraction | Volume-dependent | Under ~50 leases/year: per-document services like Prophia. A large stack you re-query constantly: build the pipeline |
| Deal screening & underwriting prep | Build (if acquisitions is your business) | Your criteria are your edge; generic tools encode someone else’s |
| Tenant/inbox triage & drafting | Wait, then build | Start with fluency on ChatGPT or Claude; automate once patterns are proven |
| CAM reconciliation & investor reporting | Volume-dependent | Painful in Excel, but only worth building at real portfolio volume |
| CRM hygiene & follow-up | Buy (HubSpot, Apto), automate at the edges | The CRM is commodity; the discipline of using it is the hard part |
Two patterns worth pulling out of the table. First, “buy” dominates wherever the vendor owns proprietary data or network effects: nobody should build their own comps database. Second, the “build” calls cluster around document-heavy and judgment-heavy workflows, the places where your firm’s specific way of working is the asset. The communications playbook and the back-office automation playbook walk the middle rows in far more detail.
The 3-Year Cost Math
Sticker prices mislead in both directions. Subscriptions look cheap until you multiply by seats and years; builds look expensive until you amortize. Here is the honest comparison, using market ranges instead of any one vendor’s quote.
The buy path costs subscription × seats × 36 months, plus setup and the internal hours to configure and adopt it. AI-tier SaaS commonly lands between $50 and $500 per user per month.
The build path costs a one-time project fee in the $25,000–$150,000 market range, plus running costs (hosting and model API usage, typically low hundreds per month for small-firm volumes), plus a support arrangement with the builder.
| Scenario | 3-year total | Notes |
|---|---|---|
| SaaS at $100/user/mo, 5 seats | ~$18,000 | Hard to beat with any build; buy |
| SaaS at $200/user/mo, 10 seats | ~$72,000 | Mid-range build breaks even inside 3 years |
| SaaS at $300/user/mo, 20 seats | ~$216,000 | Even a high-end build wins on cost alone |
| Focused custom automation ($40K) + $300/mo running | ~$50,800 | Cheaper than the 10-seat SaaS line by year 3 |
| Larger custom build ($100K) + $500/mo running | ~$118,000 | Only justified against high-seat SaaS or revenue-side gains |
Three caveats keep this table honest. The build column assumes a scoped, single-workflow automation, not a platform; firms that try to build their own AppFolio spend platform money and get a worse AppFolio. The buy column understates hidden costs: annual price increases, AI add-on tiers, and the per-seat charge for the assistant who uses the tool twice a month.
And neither column captures the largest number in the room, which is the labor the automation gives back. A workflow that returns ten hours a week to a principal is worth more than the entire spread between these columns; that return, not the software line item, is usually what decides the question.
One more line for completeness: the fluency workshop at $2,000–$15,000 is not in the table because it is not an alternative to either path. It is the cheap insurance that makes whichever path you pick perform.
Integration Reality When You Have No IT Department
Here is the part of the decision that generic buy-vs-build guides skip, and it matters more than the cost math for a 4–20 person firm.
What “No IT Department” Changes in Practice
It does not mean you cannot own custom software. It means three specific requirements move into the build contract that a larger firm would handle internally:
- The builder maintains it. Your contract needs a support arrangement with named response times, not a handoff of source code you cannot read. Ask what happens when the automation breaks on a Saturday during a closing.
- A non-technical person can operate it. Insist on a plain-language runbook: how to check that yesterday’s run worked, what to do when a document fails, who to call. If the vendor cannot produce one, they have never served a firm like yours.
- It fits the tools you already run. The output should land in Excel, Outlook, and the shared drive, because that is where your firm lives. Any proposal that starts with “first you’ll migrate to…” has doubled its real cost and halved its odds of adoption.
The Same Reality Cuts the Buy Direction Too
Off-the-shelf is not integration-free. Somebody still has to connect the new tool to your existing stack, migrate the data, configure the workflows, and get eight skeptical colleagues to change habits. Vendors quote a two-week onboarding; the honest adoption timeline at a small firm is a quarter. When we scope automation projects, the discovery conversation spends as much time on the firm’s existing tool stack as on the new system, because the integration surface, not the AI, is where projects succeed or fail. Buying does not exempt you from that surface. It hides it inside the onboarding instead.
The One-Sentence Integration Test
Whichever path you evaluate, ask the vendor or builder: “Walk me through what my office manager does on the Monday morning after this goes live.” A good off-the-shelf vendor answers in concrete steps. A good build partner answers with the runbook. Anyone who answers with a feature list is selling to a firm with an IT department, and that is not you.
The Staged Path: Fluency First, Buy Second, Build at the Ceiling
For most small CRE firms we recommend a sequence, not a single decision.
Stage one: make the team fluent. Before any spend on tools or builds, get everyone competent with ChatGPT, Claude, or Gemini on daily work: LOI drafts, lease summaries, market write-ups, email handling. This costs a workshop, not a project budget, and it changes the decisions downstream, because a fluent team can tell which vendor AI features are substance and which are demo-ware. The 90-day fluency playbook covers what that looks like in practice.
Stage two: buy the standard layer. Fill the commodity rows of the matrix with off-the-shelf tools, sized to your actual volume. Resist the bundle upsell; buy the rows you need.
Stage three: build at the ceiling. After a quarter or two, one workflow will emerge as the bottleneck the tools cannot fix, usually a cross-system, document-heavy process specific to how your firm operates. That is the build candidate. By then you have usage data to size it, a fluent team to adopt it, and a clear-eyed view of what off-the-shelf cannot do. This staged shape also protects you from the most expensive failure mode we see in the field: firms that commission a build to solve a problem that fluency or a $200-a-month subscription would have solved.
The stages are not rigid. A firm drowning in lease abstraction backlog with a deal closing next month skips ahead, and should. But when there is no forcing event, sequence beats speed.
Seven Questions That Decide It
Bring these to the partner meeting. They compress everything above into an argument-settling checklist.
- Is this workflow standard across the industry, or specific to how we make money? Standard: buy. Specific: build candidate.
- Does a vendor own data or network effects we cannot replicate? If yes, buy. You will not out-build CoStar.
- What is the honest 3-year cost at our seat count? Run the table above with real quotes, including the AI-tier pricing, not the entry tier.
- Where does our confidential deal data go? Get the answer in writing from any vendor. If the answer is unacceptable and the workflow matters, that is a build signal.
- Who maintains it on a Saturday? For a build: is support contractual? For a buy: is there a human, or a ticket queue?
- Can our least technical colleague run it from a one-page runbook? If no, adoption will fail regardless of path.
- Have we proven the volume? No volume data means you are guessing. Buy cheap or wait; do not build on a guess.
A useful pattern from vendor-evaluation work: if the room cannot agree on question 1, the firm is not ready for questions 2 through 7, and the honest next step is the fluency stage, not a procurement decision.
Frequently Asked Questions
Should a small CRE firm build custom software or buy proptech?
Buy for standard workflows (listing marketing, accounting, market data), build only for workflows that are specific to how your firm makes money and cross systems no vendor connects. For a 4–20 person firm, the default is buy; the build case must be earned with volume data and a workflow the off-the-shelf market demonstrably does not serve. Most firms should also sequence a team fluency step before either spend.
How much does custom AI automation cost for a real estate firm?
Market pricing for scoped custom automation runs roughly $25,000–$150,000 as a one-time project, plus running costs that typically land in the low hundreds per month for small-firm volumes, plus a support arrangement. A single-workflow automation (say, lease data extraction into your Excel models) sits at the lower half of that range; multi-system builds with underwriting logic sit higher. Any quote should come with a named maintenance arrangement, because a 10-person firm cannot maintain software itself.
When is off-the-shelf proptech enough?
When the workflow is standard, the vendor owns data you cannot replicate, and the per-seat math is cheap at your head count. A 300-unit portfolio on Yardi Breeze at roughly $1 per unit per month, or a firm abstracting a handful of leases a year on per-document pricing, has no build case. Off-the-shelf is also the right way to prove volume before committing build money.
Can we start with off-the-shelf and build custom later?
Yes, and for most firms that staged path is the recommended shape: fluency training first, off-the-shelf for the commodity layer, then a custom build once one workflow proves itself the bottleneck. The subscription period generates the usage data that sizes the build correctly. The one thing to protect along the way is data portability: prefer vendors that let you export your records cleanly, so the eventual build is a migration, not a hostage negotiation.
Who maintains custom automation if we have no IT department?
The builder does, under contract. A small firm should not accept a code handoff as the end state; the engagement needs an ongoing support arrangement with named response times, a plain-language runbook for daily operation, and a defined procedure for when the underlying AI models are updated. If a development partner cannot describe their post-launch support model in the first conversation, keep looking.
Is our deal data safe in off-the-shelf AI tools?
It depends entirely on the vendor’s terms, and you have to read them. The questions that matter: is your data used to train models, which subprocessors touch it, and what happens to it at contract end. Established proptech vendors generally publish answers; newer AI point tools vary widely. Where the terms are unacceptable and the workflow involves rent rolls, LP data, or off-market pricing, that is one of the strongest arguments for a custom system running in infrastructure you control.
How long does a custom CRE automation project take?
Scoped single-workflow automation commonly delivers in one to three months; larger multi-system builds run four to six or more. Beware quotes that promise a complex build in two weeks (that is a demo, not a system) and engagements without a defined acceptance test. The calendar risk for small firms is rarely the build itself; it is discovery, because the builder has to learn your workflow before automating it. Firms with a fluent team and documented workflows cut that phase dramatically.
What does a proptech subscription cost over three years, all-in?
Sticker × seats × 36, plus the parts the sticker hides: AI-feature tiers priced above the base plan, annual increases, onboarding fees, and seats for occasional users. AI-tier SaaS commonly runs $50–$500 per user per month, so a 10-seat firm at a mid-range $200 tier is committing roughly $72,000 over three years. Run the multiplication before the demo, not after.
What should we do before buying or building anything?
Get the team fluent with general AI tools on your actual work product, and inventory the workflows where hours go to die. Fluency (a workshop is $2,000–$15,000 at market rates) upgrades every later decision: you will recognize which vendor AI features are real, scope any build against proven usage instead of hope, and avoid paying project money for problems a prompt library solves. An outside AI-readiness assessment can compress that inventory into a ranked list in a week.
Where to Start
The buy-vs-build call is downstream of a more basic question: which of your workflows are worth automating at all, in what order, and by which path. That inventory is the actual starting point, and it is the thing we build in a free AI-readiness assessment: a working session that maps your firm’s workflows against the matrix above and returns a ranked plan with honest buy, build, or wait calls, including the ones where the answer is a $100-a-month subscription and not a project. Book a free AI-readiness assessment if you want that map for your firm. If a build is not the right answer, the assessment will say so, and you will still leave with the plan.
Arthur Wandzel
