Six months after a CRE investment firm greenlights an internal AI build for deal management, the prototype works. Pipeline is tracked, OMs are being parsed, and leadership is impressed.
Then the person in charge of building the prototype leaves for a new opportunity, and no one’s left to help explain where a number came from or reconfigure the system when the firm’s strategy shifts.
That failure doesn’t prove the firm was wrong to build. It proves the firm didn’t know, going in, everything that would come after the initial prototype. The real question in the build-vs-buy call for deal management was never whether a team could build something that works. It’s whether the firm knows which tradeoffs it’s signing up to own for the long run.
The Vibe-Code Moment in CRE
This is playing out across institutional CRE right now, and for good reason. AI has genuinely lowered the barrier to building internal tools; a small team no longer needs a full product organization to stand up something functional in a matter of weeks.
The impetus usually comes with a single, narrow problem, such as quickly extracting details from OMs. Then the project grows in scope, and a small team ships something that works. Once leadership resources the project appropriately, a proof of concept quietly and quickly becomes the firm’s operational infrastructure for real capital decisions.
None of that is a mistake. Building is a reasonable response to a fast-moving market, and plenty of firms have the engineering depth to pull it off well. What’s worth examining is what a firm is committed to, whether it chooses to build in-house or invest in a prebuilt platform.
Accountability and Risk: Who Answers When Something Breaks?
Think back to that stale comp. With most internal builds, there’s no SLA to invoke and no contractual remedy to pursue: just the people who wrote the code, however many of them are still at the firm. That’s structurally what an internal build is: full ownership, and full exposure, sitting inside the firm.
Some firms are genuinely equipped for that because they have a real internal accountability structure: someone on call, someone who signs off before a number reaches an IC deck, someone prepared to explain a failure to an LP without pointing at a vendor.
Buying shifts that exposure outward. A platform built specifically for institutional deal management typically comes with a contractual SLA and a defined escalation path. Dealpath, for instance, operates under SOC 2 Type II with documented audit trails, because its clients answer to LPs who ask about exactly this. That protection comes with a different dependency: the firm is now trusting a vendor’s accountability structure instead of its own.
Neither path is inherently safer. They distribute the risk to different places, and only one of those places is inside the building.
Total Cost of Ownership: Why “Built” Doesn’t Mean “Done”
“We’ll build it” is a decision about a working prototype. It’s rarely a decision about everything that comes after — and that’s not an accident of oversight. Build costs are visible from day one: an engineer’s salary, a sprint timeline, a launch date. Soft costs aren’t, which is exactly why the comparison to buying looks so favorable at the start. The real ledger includes:
- Data governance (e.g., role-based access controls, data standardization, and audit trails) that scales over time
- Compliance work if the firm pursues SOC 2 or an equivalent certification
- LLM token costs that climb as usage and data volume grow
- Performance tuning as the database expands
- Roadmap decisions that need designated owners
- Knowledge transfer when key personnel who built the system leave the firm
That last one compounds in a specific way: the cost isn’t just that a person leaves, it’s that the logic they embedded in the system needs to be documented and maintained. A build that isn’t documented independently of its builders isn’t really owned by the firm as much as it’s on loan from whoever’s still there to explain it.
None of this makes building wrong. It’s also the price of something real: no recurring license fee, no vendor roadmap dictating what happens next, and no negotiating leverage sitting with someone else.
Buying trades that structure for a different one. The cost bundles into a subscription, but it recurs indefinitely and scales with seats and usage. Over a long enough horizon, for a firm with strong existing engineering capacity, it can add up to more than a lean internal build would have cost, provided that build was actually budgeted for its full lifetime, not just its first release.
Building an AI-Ready Foundation: The Layer Both Paths Depend On
Every deal management solution worth its salt must have AI capabilities, but whether AI actually makes a difference depends on the quality of the firm’s data foundation.
Dealpath’s 2026 State of AI in CRE Investing survey found that 83% of institutional professionals rate their data as AI-ready, but paradoxically, 90% say data quality or fragmentation has already limited AI’s impact at their firm. This gap needs to be addressed. In other words, to carry AI into the work that matters, like surfacing comps and informing underwriting, data has to be centralized, structured, current, connected to the deal record, and traceable back to its source.
The risks of poor AI implementation, such as runaway token costs, are real. The more AI has to hunt for the right data, the more token costs can overwhelm a firm. In May, Uber revealed it burned through its entire 2026 AI budget in just four months. Though open-source models are tempting — they promise lower costs than proprietary models developed by companies like OpenAI and Anthropic — there’s no promise open-source models are secure or aren’t using a firm’s sensitive data for training purposes.
Building in-house means constructing an AI-ready data foundation yourself: the structure, the access controls, the lineage that lets someone trace a number back to where it came from. For a firm with unusual deal structures and a strong product team, shaping that foundation exactly around how it underwrites is a real advantage. It also means owning the work of keeping the foundation AI-ready as the database grows — the part most likely to be underestimated at the prototype stage.
Buying means inheriting a foundation that already carries that structure and configuring it to the firm’s workflow rather than building the governance from scratch. Configurability and governance aren’t opposed here; a platform like Dealpath is built to be shaped around how a firm actually works while keeping the underlying data structured, connected, and traceable.
Either way, firms must move off generic, ungrounded AI and use AI anchored in the firm’s own structured deal data, where outputs come back attributed and verification shrinks to confirmation.
Measuring the Trade-Offs to Determine the Best Path Forward
The question to sit with is this: Beyond the build itself, have I accounted for the full lifetime cost of owning this — the data foundation and governance that make AI’s outputs trustworthy — and can my firm carry that cost for the long run?
The build-vs-buy decision was never about which path is more capable. Both can work. It’s about which of these tradeoffs — accountability, total cost of ownership, and the data foundation underneath — a firm is actually equipped to own over time. That’s worth answering deliberately, before the first line of code gets written or the first contract gets signed.
Dealpath’s team has worked through this tradeoff with institutional firms across the spectrum. Request a demo to talk through where your firm lands.
Request Demo
