01 · Case study
A capital firm replaces its CRM and cuts AI spend 8x.
A private-credit investment firm had already burned through developer after developer trying to get off its legacy CRM. Three months of architecture consulting later, the CRM was gone, AI costs were down roughly 8x, and two institutional investors were ready to fund the resulting platform as its own product.
Last reviewed:
Industry
Private credit, distressed residential mortgage notes
Portfolio
Nearly 400 active assets, 8-year track record of over 1,000
Engagement
AI systems consulting, 3 months
The situation
The client runs a $100M+ portfolio of distressed mortgage notes. Every part of the business — loan servicing, investor reporting, CRM, marketing — ran through a third-party SaaS platform and a senior analyst who manually reconciled reports by hand.
The company had already tried to fix this. Multiple developers had been hired to build custom middleware and get off the legacy platform. None of them delivered. “We’ve just been burning money on fire with developer after developer, and none of them came through with proper middleware,” the CTO told us.
The CTO wanted to build the fix himself with AI tools, but needed someone who could tell him what to actually build, in what order, and where the expensive mistakes were hiding.
What we did
Over three months of weekly working sessions, we acted as the technical architect behind the build:
- Diagnosed and eliminated a recurring platform crash that was costing hours every week
- Designed the multi-agent development structure the CTO used to build the entire platform
- Identified the real technical gap blocking cross-portfolio AI analysis and pointed to the right category of solution, instead of the wrong one the team had been chasing
- Stopped a premature rebuild that would have stalled the project for months, in favor of a staged migration off the legacy system
- Corrected the retrieval architecture for the AI chat feature: the team’s instinct was to build a document-search system, when the actual need was structured database queries. Building the wrong one would have meant slower, less reliable answers and a costly rebuild later
- Designed a data-access approach that excludes sensitive borrower information at the database layer, which turned a $20,000–$40,000 hardware requirement into an optional upgrade rather than a blocker
- Hardened two production automations (daily data feeds) with retry logic, failure alerts, and structured logging
The results
- Legacy CRM eliminated. $10,000 per year in software costs removed, replaced by an internal platform built during the engagement.
- AI operating costs cut roughly 8x. Monthly AI spend went from around $1,600 to $200.
- A single source of truth. All P&Ls, investor reports, CRM data, and nearly 400 assets now live in one system instead of scattered across spreadsheets and a third-party platform.
- Two production automations running daily, replacing manual data handling, with zero failures since launch.
- A fundable product. Two institutional investors are ready to fund the platform as a standalone offering, pending one clean quarter running on the client’s own books.
- Headcount reduced. One analyst role fully displaced by the new system; a second moved from salary to commission-only.
What this would have cost with a dev team
A platform of this scope — a multi-module operating dashboard, a middleware layer integrating four external systems, an investor portal, an admin panel with role-based access — would typically run $250,000 to $500,000+ built by an outside development team over 6 to 12 months, at standard market rates for senior full-stack developers plus project management and design.
The client’s own CTO, after using the platform daily for months, put it more bluntly: “This would have cost you a couple million dollars to build by developers.”
What it actually cost: no developers hired, no dedicated build budget. The only recurring cash cost was AI tooling, which started around $1,600/month and was optimized down to $200/month during the engagement. The platform itself was built by the CTO, guided by our weekly architecture sessions.
They told me five weeks… it took about an hour total. Everything was done.
What’s next
The client has engaged us for a second phase of work: a natural-language chat feature that lets non-technical leadership query the entire portfolio live, in front of investors and banks, plus two additional automations that fully replace the outgoing analyst’s remaining manual work.
Ready to build a platform your team actually owns?
Book a strategy callWe'll compare notes and see whether an assessment or a build makes sense for you.