Should we build our own AI underwriting system?

Build your own AI underwriting only when the workflow itself is your edge: proprietary model structures, mixed asset classes, an IC process no vendor template covers, or data that cannot leave the firm. If your workflow is standard for your seat, buy vertical SaaS instead: lender credit workflows, institutional pipeline management, and sell-side deal materials are solved product categories. Either way, plan for who operates the system after it ships. A custom build the team cannot run becomes a vendor dependency with worse economics than the subscription it replaced.

When does vertical SaaS beat a custom AI build for CRE teams?

Vertical SaaS wins when a vendor has already productized your exact workflow. CRE lenders underwriting loans at volume fit Blooma, institutional investment managers standardizing pipeline and IC process fit Dealpath, and sell-side brokerage teams producing OMs and pitch decks fit Henry.ai. In those categories the vendor has debugged the workflow across hundreds of customers and spreads the maintenance cost across all of them, which a single firm's custom build cannot match. Buying is also faster: a subscription deploys in weeks, while a build is a project. The test is fit, not features: demo the tool on your own messy documents and see whether your process survives contact with the template.

What does a custom CRE AI system actually cost to maintain?

The build is the smaller commitment; ownership is the real cost. A custom AI underwriting system needs ongoing attention in three places. Maintenance, because document formats drift, integrations break, and extraction pipelines fail silently until someone notices. Model churn, because the AI model that was best when the system shipped gets superseded and must be re-evaluated and swapped without breaking the workflow. And key-person risk, because a system only one developer understands has a lifespan equal to that person's tenure. None of this makes building wrong, but it changes how you should build: either pair the system with a maintenance retainer you trust, or have the builder transfer the capability so your own team can operate and extend it.

How do CRE firms build AI capability without hiring engineers?

The practical path is done-with-you enablement rather than hiring. In NextAutomation's AI Team Program, a CRE firm's existing team, analysts, associates, and operations staff, builds working AI systems in weekly group working sessions: prompting, agent workflows, and underwriting and reporting automations, on the firm's real deals. The deliverable is capability transfer: the team learns to ship and own the systems rather than receiving code it cannot maintain. Firms that also want senior direction add a fractional Chief AI Officer, an embedded operator who owns the AI roadmap, makes the build-versus-buy calls, and transfers judgment to the team as they go.

Should You Build Your Own AI Underwriting? Build vs Buy for CRE Teams

The Short Answer

Most CRE teams should buy vertical SaaS when their underwriting workflow matches what a vendor already sells, and build custom when the workflow is proprietary, the data cannot leave the firm, or per-seat pricing breaks at their scale. The tiebreaker nobody prices in: who at your firm can actually run the system after it ships.

One disclosure up front: NextAutomation builds custom AI underwriting systems for CRE firms, so we have a direct stake in this question. That is exactly why this framework starts with the cases where buying software is the right call, because for most standard workflows it is. (For the tool-by-tool landscape, see our guide to the best AI underwriting tools for CRE.)

When Buying Vertical SaaS Wins

If your underwriting workflow is standard for your seat, a vendor has already built it, debugged it across hundreds of customers, and spread the maintenance cost across all of them. Concede these three categories without a fight:

Lender underwriting at volume. Blooma automates CRE loan origination analysis and portfolio monitoring for banks and debt funds: deals are parsed at intake, scored against the lender's risk appetite, stress-tested, and re-underwritten continuously on the portfolio side. Its plan tiers are gated by team size and origination volume (per Blooma's plans page), and the company raised a $15M Series A led by Canapi Ventures in June 2021 (per the BusinessWire announcement). If you are a lender, start with Blooma before you consider building anything.
Institutional deal pipeline. Dealpath is the system of record for institutional acquisition pipelines, reporting 300+ firms and more than $10T in transactions (per Dealpath's homepage), with quote-based per-user plans that typically carry a five-user minimum (per Dealpath's plans page). If you are standardizing pipeline, diligence, and IC process across a large team, Dealpath is the category leader; do not rebuild it.
Sell-side deal materials and research. Henry.ai runs underwriting on the firm's own Excel model through a two-way add-in and generates OMs, BOVs, and pitch decks on brand, and it is SOC 2 Type II certified (per its homepage). If you are an investment-sales team that wins on the speed and quality of listing materials, buy that capability rather than building it.

The pattern: these are solved product categories. A custom build that re-implements one is an expensive way to arrive, a year later, where a subscription would have taken you in weeks.

When a Custom Build Wins

Building your own AI underwriting makes sense in four situations, and they share one trait: the vendor template does not fit.

The workflow does not map to any SaaS template. Mixed asset classes, a proprietary model structure, an IC memo format your committee actually reads: if the way you underwrite is part of how you win deals, forcing it into a vendor's workflow objects gives that edge away.
The data cannot leave the firm. LP records, deal terms under NDA, or a comps history you consider proprietary. Most vertical platforms compound value inside their own database; that is the vendor's moat, not yours.
Per-seat or per-volume pricing breaks at scale. Dealpath's typical five-user minimum (per its plans page) and Blooma's volume-gated tiers (per its plans page) are rational vendor economics, and they also mean the software bill re-tiers as you grow. A system you own does not charge you more for closing more deals.
The firm wants owned capability, not a subscription. A custom system built around your workflow is an asset the firm keeps, extends, and controls, instead of a seat license that ends when the contract does.

What that looks like in practice: a custom AI underwriting copilot shaped to your model, your counterparties' document formats, and your review process, with a human check on every extraction.

The Part Both Sides Undersell: Maintenance

Honest words about what building means, from a company that builds: the build is the cheap part. What you are really signing up for is ownership of a living system.

Maintenance burden. Document formats change, broker templates drift, integrations break, and extraction pipelines need monitoring. Someone has to notice when the rent-roll parser starts failing silently.
Model churn. The AI model that is best-in-class when your system ships will be superseded. Someone has to re-evaluate the replacement, re-test it on your own documents, and swap it in without breaking the workflow.
Key-person risk. If one developer, or one vendor, holds all the knowledge of how the system works, the system's real lifespan is their tenure.

This is why the sharper version of the question is not build versus buy. It is build with capability transfer versus build as vendor dependency. A custom system your team cannot operate is just a subscription with worse economics. There are two honest ways out: the AI Team Program, where your team builds and learns to own the systems in weekly working sessions, or a fractional Chief AI Officer, an embedded senior operator who owns the roadmap and transfers capability as they go.

How to Decide

Three questions settle most cases:

Is your workflow standard for your seat? If a vendor demo matches how you already underwrite, buy it. Test it on your worst-formatted rent roll and a scanned T-12 first, not the vendor's demo file.
Is the workflow itself your edge? If your underwriting standards, model structure, and IC process are how you win deals, a build molds to them instead of flattening them into a template.
Who will run it in year two? If the answer is "the vendor," make sure that is a vendor you trust with the workflow. If the answer should be "our team," build with capability transfer from day one.

If you want a straight answer for your specific firm, that is what a paid AI audit is for: we map your underwriting workflow, tell you where a SaaS platform is the better buy, and only propose a build where custom genuinely wins. If the real gap is capability rather than software, the AI Team Program trains your team to run AI-native underwriting in-house. Book an intro call, or start with the free AI roadmap.

Build this with NextAutomation

See how a custom underwriting system runs, then start building your own. Walk the AI underwriting copilot demo, install the free AI underwriting copilot template, and read the CRE tech leader's guide to AI implementation for a pilot-to-production framework.