What it costs to build your own AI for real estate operations: a build vs buy cost model

Decide which parts your team needs to own#

Before comparing prices, define the work both options must deliver: the documents they need to read, the systems they need to connect, the decisions they need to support, and the actions they need to complete. A lease-extraction tool and a production workflow that prepares a recovery claim are different scopes.

Then decide which capabilities your team needs to build and maintain. One option is to buy the shared foundation for document processing, portfolio data, evidence, and ongoing upkeep, while retaining your own approval chains, scoring rules, and reporting requirements.

Ask who builds and maintains those company-specific workflows. Some platforms expect your team to do it. With REAL, your team defines the business rules and REAL's engineers build and maintain the agreed custom workflows.

Whatever you choose, hold it to one standard. Reading a lease is not enough, and finding an exception is not enough. What counts is the action that follows (a recovery claim, a notice, a journal entry, a work order), the evidence behind it (the clause, sheet or ledger line), and an outcome you can measure against today's process. That is what we mean by AI execution for enterprise real estate, and it is the standard the cost model below is built to price.

Compare both options over the same horizon#

A fair comparison counts both options from the same starting point, the day you decide, and ends them on the same date. A build that takes 12 months to reach production carries 24 months of operating cost in a 36-month comparison. Enter the expected deployment time for each option, for the same scope, rather than assuming either one reaches production first.

This model assumes the vendor subscription is billed from signing, so it counts the subscription for the full comparison period. That is an assumption of the model, not a universal contract rule. If your agreement starts billing at go-live or on another date, adjust the subscription you enter to match.

Costs continue after launch. AI systems need ongoing testing, maintenance, and review as models, documents, and integrations change. Model providers retire versions and change behavior. Document formats drift as new landlords, new lease forms and new amendment styles arrive. Accuracy on long documents is uneven: Stanford researchers found that language models perform best when the relevant information sits at the start or end of a long input and degrade when it sits in the middle (Liu et al., "Lost in the Middle," TACL 2024). The finding is a reason to test retrieval across long leases and amendments, including clauses located deep within a document. It does not establish an error rate for a particular lease workflow.

The cost model, line by line#

Choose one comparison period, measured from the decision date, and use it for both options. The calculator starts with 36 months, which you can change.

How do you price the people lines?#

Start with the roles and time the project needs. Estimate the engineering, integration, testing, and subject-matter work required before launch, then identify who will maintain the system afterward. Use the share of each person's time allocated to this project, rather than automatically counting their entire salary.

For a U.S. salary reference, the Bureau of Labor Statistics reports a median annual wage of $135,980 for software developers in May 2025 (BLS Occupational Outlook Handbook). That is a national wage benchmark, not an AI-specialist rate or a minimum salary. Add your organization's employer costs and overhead using a multiplier agreed with finance.

The calculator's example uses that wage with a 1.3 multiplier, producing an estimated loaded cost of $176,774 per FTE per year. Replace both values with costs appropriate to your team. The calculator applies one loaded rate to every FTE, which is a simplified blended-rate assumption: if engineers, analysts and lease administrators cost different amounts, use a blended rate agreed with finance.

How do you price the buy side fairly?#

Price the bought option on the same horizon and include the work your team still does. Internal effort is real on the buy side too: someone owns the vendor relationship, reviews exceptions and approves actions. This model bills the subscription from signing, so it counts the subscription for the full period; check when billing starts in your agreement. Enter the vendor's expected deployment time for the full scoped workflow.

REAL, for example, prices on the number of locations managed and the solutions deployed, with no per-user fees, and states that initial agents go live in two weeks, running in parallel with the existing process first (REAL pricing, REAL onboarding). Initial agent availability is not the same as completing the entire scoped deployment, so enter the go-live date for the whole workflow. Put your own vendor quotes and timelines into the same formula, and test them with the scenario controls in the build vs buy calculator.

What does outside evidence say about build vs buy?#

The studies below describe enterprise adoption patterns and deployment challenges. They do not establish which option will cost less for your workflow.

• Deployment rates. MIT NANDA's "The GenAI Divide: State of AI in Business 2025" reported that purchased tools and vendor partnerships reached deployment about 67% of the time, against about 33% for internally built tools. The figures are self-reported, and they measure reaching deployment, not financial return (Fortune, Aug 18, 2025).
• Where enterprises are going. Menlo Ventures' 2025 survey of 495 U.S. enterprise AI decision-makers found 76% of AI use cases were purchased rather than built internally, up from 53% purchased in its 2024 survey (Menlo Ventures).
• Where projects stall. Gartner reported in January 2026 that at least 50% of generative AI projects were abandoned after proof of concept by the end of 2025, due to poor data quality, inadequate risk controls, escalating costs or unclear business value (Gartner, January 2026).

None of these is about real estate specifically, and none replaces your own numbers. Gartner's four reasons do show up in lease and CAM work: inconsistent source documents, audit exposure, usage costs that scale with portfolio size, and value that depends on someone acting on the output.

When does building make sense?#

Building deserves closer consideration when the following conditions apply:

1. 01The scope is narrow and stable. One document type, one workflow, one region, with little change expected.
2. 02The team is permanent. Engineers and an analyst are staffed for the run years, not only the build.
3. 03The data is already clean. Lease data, ledgers and asset registers are consistent, so data connections and test set creation are small.
4. 04The workflow is unusual. No vendor covers it, and it is central to how you compete.
5. 05Deployment timing is acceptable. The expected build schedule fits your operational deadlines, and the financial impact of waiting has been considered.

Use these questions to identify constraints and responsibilities. Then compare the costs of the same scope using your staffing plan, vendor quote, and deployment assumptions.

What changes when you buy the foundation#

Buying the shared foundation can shift responsibility for document processing, model infrastructure, evaluations, and connection maintenance to the vendor, where those capabilities are included in the agreement. Your team still needs to define requirements, review exceptions, approve actions, and manage the relationship.

Price any company-specific workflow development and maintenance explicitly. Depending on the agreement, that work may sit with your team, the vendor, or another implementation partner. In the calculator, include the relevant setup work in implementation costs and ongoing internal responsibilities in annual internal effort.

With REAL, REAL's engineers build and maintain the agreed custom workflows, on agents where every output links to its governing clause or ledger line and critical decisions require human approval.

Build, buy, or buy and build on top: who carries each cost#

This table compares responsibilities, not prices. The calculator models two options, build and buy; price a buy-and-build-on-top arrangement as the buy option, with the workflow setup in implementation and the work your team keeps in internal effort.

Questions to ask before you sign off on a build budget#

• Who is on the run team in year three, and is that headcount approved?
• Are both options priced over the same horizon, counted from the decision date?
• How many leases (or invoices, or drawings) are in the test set, and who checked them?
• What happens to accuracy when the model provider retires the version we tested?
• What is the incremental financial impact of each month before go-live, and which parts are permanently lost rather than received later?
• Can an auditor trace any output to its source document today?

For function-by-function versions of this decision, see build vs buy for AI lease abstraction, build vs buy for CAM reconciliation and the series hub, build vs buy AI agents for corporate real estate. For how REAL's agents share one record, see the REAL platform.

The useful comparison is not build against buy in general. It is the same scope for both options, your own inputs, and the same comparison period.