Building AI in-house vs DealSage: the real cost of DIY.
A working prototype takes a weekend. Keeping one running, secure and trusted for three years is a different job entirely, and it's the one that rarely shows up in the original pitch to build. Here's the honest cost, time and risk breakdown.
A demo that answers one question about one document is a weekend project. A system a whole investment team trusts with real deal work looks like this, and every piece keeps needing attention after launch, not just at launch.
01
Ingestion
Pull email, CRM, data rooms, drives and call transcripts into one pipeline that keeps working when a vendor changes its API without warning.
02
Entity model
Decide what a deal, a contact, an organisation actually is in your data, and keep that model consistent as your process changes under it.
03
Permissions
Mirror who can see what, deal by deal, so the model never surfaces something a junior analyst was never meant to read.
04
Lineage
Trace every number back to the page it came from, or accept that nobody on the investment team will trust the output.
05
Evals
Build a way to test whether last week’s model upgrade actually made the answers better, or just changed them.
06
Interface
Build, and keep rebuilding, the screens, the Excel plugin, the email integration: whatever your team actually works inside.
07
Agent orchestration
Get multiple steps of reasoning to hand off to each other reliably in production, not just once in a demo.
THE HONEST EXCEPTION
When in-house is actually the right call.
None of this is an argument that building is always wrong. It's the right call for a narrower set of firms than the DIY instinct usually assumes.
A real engineering org
You already run a software team that ships and maintains production systems day to day, not a data scientist who also files your expenses. The maintenance burden in the table below is a cost you're already carrying for other systems.
A genuine data moat
Your proprietary data is the actual product, and owning the full stack, not just the interface, is the point of the business. That's rare. Most firms' edge is judgment applied to data, not the plumbing that moves it around.
A hard regulatory requirement
Some mandates genuinely require every system to run inside your own four walls, no exceptions. Worth checking first: DealSage's on-premise deployment covers most of what firms actually mean when they say this, so the honest remaining case is usually the first two.
SIDE BY SIDE
In-house vs DealSage, on paper.
Build in-houseDealSage
Time to liveTypically 6–12 months to a working baseline, by reported accountsWeeks
Total cost shapeHeadcount: typically 2–4 engineers for a year to reach parity with a platform baseline, plus infrastructureSubscription plus a scoped embedded-team engagement
Who maintains itYour team, indefinitelyDealSage’s engineers, as part of the platform
Model churn riskYours to absorb every time a frontier lab ships a new modelAbsorbed by the platform; routing updates without you touching anything
Security burdenYours to review, patch and prove, every timePlatform-level: DealSage Cloud, private VPC, or on-premise
Key-person riskHigh: the one or two engineers who understand it can leaveNone; the platform is the institutional memory
What you own at the endThe code, assuming it still runsYour data and the ontology built on it
Time and headcount figures are typical ranges reported by firms who've made this build, not a quote for any specific case. Your firm's actual cost depends on scope, existing infrastructure and how much of this you can borrow from other teams.
The deeper argument, if you want the full case
This page keeps to the framework: what you'd have to build, what it actually costs, and when in-house still wins. The long-form version goes further into why the build feels so tempting right after the first good AI result, and what the biggest, best-resourced firms in the industry actually did when they had this exact debate with far more budget and far more at stake than most mid-market funds. Read the full argument on build vs buy AI for the rest of it.
Frequently asked questions
How long does it take to build an AI platform in-house?
Reported timelines for a mid-market fund to reach a working baseline, not full parity with a mature platform, typically run six to twelve months. That’s before the ongoing work: retuning retrieval, patching security, and upgrading every time a frontier lab ships a new model, which happens every few months indefinitely.
Can we build custom things on top of DealSage instead of building from scratch?
Yes, and that’s most of the point of the platform rather than a workaround. DealSage connects over MCP so your own tools and models can query the ontology directly, supports custom objects alongside Deal, Contact and Organisation for whatever your firm needs to track, and lets you bring your own LLM rather than being locked to one vendor’s model. You get the foundation built, then extend it, instead of building the foundation yourself first.
What happens to our data if we ever leave DealSage?
You keep it. The ontology is built on your firm’s own data, and a churn conversation should be a data export conversation, not a hostage negotiation. Ask any vendor, DealSage included, to put that in writing before you sign, and be wary of anyone who won’t.
How many engineers does it actually take to build and run this in-house?
Reported figures for reaching parity with a platform baseline, not exceeding it, typically land around two to four engineers for the first year, then ongoing headcount indefinitely after that for maintenance, model upgrades and security review. Treat that as a typical range from firms who’ve done it, not a quote for your specific case.
Is DealSage cheaper than building in-house?
Usually, once you count the full three-year cost rather than the first quarter. The build itself is rarely where in-house projects go wrong. It’s the maintenance that never shows up in the original business case: infrastructure, model churn, security review, and the salary of whoever has to keep it running after the person who built it moves on. DealSage’s cost is a subscription plus a scoped embedded-team engagement, and both are quoted up front.
