Building the replacement took three years, and we cancelled the contract in year four. Most of year one went to a sync decision we had to undo.
Salesforce set the ceiling
Matthews could not adopt anything Salesforce had not adopted first. Every new capability waited on someone else’s roadmap.
Our data had to be shaped the way Salesforce shapes data, which doesn’t fit how a commercial real estate business works. As AI capability moved, the wait on Salesforce’s roadmap and the mismatch in data shape both got worse.
It was also hard to use. New agents took months to learn it.
We started in 2022 with two product leads and four engineers.
Our sync design cost us a year
The first plan was to build Artemis on the data structure we wanted. Salesforce data would be transformed into that shape on the way in. Brokers worked in both systems during the migration, so the two had to stay in sync in both directions.
They didn’t. Brokers reported that the systems didn’t match. They would update a record in one and check the other to be sure it had changed. Our data shape kept changing while we built, and the Salesforce API has quirks of its own. Together they made two way sync nearly impossible. The transforms threw constant errors, and the systems drifted apart.
It cost us trust. A migration only works if users believe the new system holds the right data, and every mismatch gave brokers a reason to doubt it.
Our platform architect made the call. We mirrored Salesforce’s structure for everything we synced, ran that way until the last user was off, then refactored. The mirror cost experiential quality. Brokers and AI agents both worked with Salesforce’s data shape until the refactor ran.
The mistake took most of year one, alongside moving the front end from Angular to React. If you are replacing a system people still use, mirror its structure first and refactor after the last user leaves. If you’ve run a migration like this, I’d like to hear how you handled the sync.
Matthews runs on its own software now
Artemis replaced Salesforce across the company. By early 2026 we had 11 engineers, and two leaders from marketing and operations had joined the product team. We cancelled the contract in year four. The data refactor we had deferred ran once nothing synced to Salesforce anymore.
New agents took months to learn Salesforce. Artemis takes about a day.
$1.3M a year to $0Salesforce contract
The contract was $1.3M a year when the work started. By the time the last user moved off, renewal would have been $1.9M.
A broker who builds alone needs us less
Matthews takes a percentage of what its brokers close. They accept that because closing requires things they cannot easily get alone: marketing, legal, and the data and software behind prospecting. Remove those and there is nothing to charge for.
AI is making those things buildable by one person. Every broker who runs their own marketing or handles their own pipeline has less reason to give up a percentage.
Our subject matter experts know these services at a depth a broker cannot replicate, and we hold data no independent has. Both matter less if brokers build their own tools anyway.
So we could discourage building and defend the old arrangement, or make Matthews the place where building happens. We took the second.
A broker building on our platform starts with that data, and with marketing and legal expertise already inside the tools. A broker building alone starts with a blank page. So does a broker at a competitor that has not allowed any of this.
Everything built in house becomes property. It fits how the company works, moves at the company’s speed, and waits on nobody’s backlog. The difference is like paying rent vs. paying a mortgage.
The rebrand was a technical decision. The old identity ran on gradients, texture, imagery, and layered composition, which need a person in Photoshop and don’t generate reliably. With a cleaner system, I estimate builders produce on brand work without manual design about 90 percent of the time.
Safety had to be built in
Every department runs its own portfolio, staffed and maintained by its own subject matter experts. The platform underneath keeps that from creating exposure.
A builder never has to know the rule for the rule to hold. It arrives with the tools.
- Identity and permissions: Okta sign in, PKCE, and row level security at the shared database layer
- Embedded policy: compliance ships inside the tools as skills
- Shared components: design tokens and UX judgment built into the library
- Write access: off by default, granted per environment, revocable
It changed engineering’s job too. They used to serve one team or one initiative at a time. Now they build the platform everyone else builds on.
Year one displaced $750K in vendor contracts
| Outcome | Result |
|---|---|
| Vendor spend displaced by builder apps, validated by finance | $750K in the first year |
| Content creation cycle time | Weeks to minutes |
| Legal document creation | Nearly automated |
| Operations teams | Same size while throughput grew |
Where I sit at Matthews
Matthews is a commercial real estate firm. The harder half of the work has been moving a services business into operating like a technology company.
I led the design of Artemis and owned every experience decision from inception through the full migration and the contract cancellation. I direct product for the app ecosystem and co-lead platform strategy. The program covers the framework, the ownership model, the team that supports departments, and the platform work that makes the builder tools usable.
The job has two halves now. One is running that nomadic product organization. The other is acting as product manager for the builder platform itself.
