We went first. Being Customer Zero.

Michael Rouleau

Chief Executive Officer

Twelve to fifteen months. That's how long it took to run a full AI transformation on our own business, and it's the first number I give a financial services leader who asks, because it's the one nobody publishes.

Last time I wrote that the institutions pulling ahead run experience, data, and AI as one connected system, and that we'd built that system inside our own walls before asking a client to follow us. Here's what it took.

Why we went first

We sit in an unusual seat. Across 600+ Salesforce projects and 200+ data projects, nearly all of them in financial services, we could see where the work was heading, and it was heading toward AI faster than most firms were ready to admit. That left us with a choice about what kind of firm we wanted to be. We could sell that future from a slide, or live it first and sell what we learned.

I spent part of my life in the military, and one principle from that time has never left me. You rehearse the mission before you run it, and you don't send people into something you haven't done yourself. If we were going to guide clients through an AI transformation in a regulated industry, we'd be customer zero. So we ran it on the whole business, pre-sales through delivery through back office, and gave ourselves nowhere to hide.

‍

The order we did it in

The order turned out to matter more than any tool we picked. That's the part I'd hand you first.

We started by turning our own maturity assessment on ourselves, the same structured review we'd built for clients. You can't sequence work you haven't measured, and you can't measure a business by asking people how they think they spend their time. That review surfaced 16 internal processes ready to automate.

We built a reasoning layer on top of those 16 and stopped there, because the temptation at that stage is to automate everything you can instead of everything you should. Those early wins mattered less for the hours saved than for the proof. The approach had now worked on a real business, ours.

Then the plumbing. Our own MCP server went in so agents could reach our systems properly, rather than through workarounds none of us would want to explain to an auditor.

Then governance, before scale. Every agent we run moves through six stages. We prescribe it, activate it, register it, govern it, measure it, and only then go build the next one. Registration is the step most firms skip and the one that saves you later, because an agent nobody registered is an agent nobody owns.

Last, we used our own AI to build our AI roadmap, with a human checkpoint at every decision that carried real risk.

Read that order back. Assessment, narrow build, plumbing, governance, roadmap. Tooling decisions came near the end, and the programs I watch stall are almost always the ones that started there.

What it produced

New business deals used to take 40 days to reach proposal. They now take 14.

The bigger shift was in how we sell. A quarter of our deals last quarter were fixed price, up fourfold in a year, and margin held. Pricing is where conviction shows up. Bill by the hour and your client carries the risk, quote a fixed fee and you carry it yourself.

The efficiency gains are real, and we measure them, but they aren't what our teammates talk about. What they talk about is the work they got back. A consultant who used to spend two days assembling an assessment now spends those two days with the client, applying judgment no model has.

The stack, and why it's this one

Our business runs on Claude, Agentforce, Databricks, Data 360, Salesforce, and Twilio. One connected flow, one version of the truth, governed the whole way through.

The rest of the market moved the same direction while we were building. Anthropic and Salesforce expanded their partnership last fall, making Claude a foundational model inside Agentforce 360 and the first LLM contained fully within Salesforce's trust boundary, with financial services named as the first industry for joint work. Databricks spent the year hardening the data layer. Twilio built persistent memory into the engagement layer. We were building on that architecture while they were building it, which beats reacting to it afterward.

What it cost us

The optimism above is only credible next to the reality, so here it is. Those twelve to fifteen months did not run in a straight line.

Six months in, a frontier model change forced us to rework things we thought were settled. Of the long line of delivery tools we tested, most got killed. Measuring results before scaling anything slowed us down and saved us twice. Our enablement spend went past what felt reasonable, which turned out right. Change management is where we made our mistakes, and it taught us more than any of the wins.

Speed and precision show up first in a transformation like this. Full funnel conversion and captured margin come later, and we're still proving those.

Your version of this

If you lead a regulated or private-equity-backed business, the pressure we felt is the pressure you're feeling now. The upside is obvious, the path to it isn't, and a stalled pilot or a governance gap carries real cost.

Most institutions I talk to already have the technology somewhere in the building. The gap is activation, turning a license into governed workflows that produce measurable value.

Four things you can learn from our own experience:

Sequence the work before you buy anything, because order beats software.
Decide how you'll measure return before you automate a single process. Programs that skip that step rarely survive their second budget cycle.
Treat governance as engineering from day one.
Spend more on enablement than feels reasonable, because the technology was never the hard part. The people side is where we made our mistakes.

Start with the sequence, because you can't set one without a baseline. Our Digital Maturity Assessment scores your three layers against your peers and hands back a roadmap in 48 hours, ordered by what has to come first. We've run 130 in the last 90 days. It costs you nothing, and the output is yours to take to whichever partner you choose.

We're still building, too. There's more coming on our partner ecosystem over the next few months, and I'll write about it when I can.