You remember the last call where the other person asked you to start from the top. You had already typed it into a form.
This is the path a booking follows here, and the path that opens the morning after both sides sign. One workflow, from the booking to the next morning.
Before the call
A booking starts a research pass on its own. By the time I join, there is a written brief: who the firm appears to be, why they booked, and what is still unknown. The questions for that half hour are written down too.
When we can tell which firm it is, the call deck is in the folder. That deck is the one we use for these conversations. What is written for that firm is the brief and the questions. When we cannot tell, the brief names the doubt, and the deck stays out of the folder.
The agreement goes out after a yes, before we hang up. It is our standard audit agreement, with that firm's scope written in.
The morning after both sides sign
A second pass opens the working room.
A private Slack channel exists, for this engagement only. A Drive folder holds the signed agreement. A shared document asks who owns each system, whether access is ready, and the link. Passwords stay out of that document. A welcome email points at it, and it says the Slack invite is on its way.
Three things are still a person, the same day. I invite them into the channel. I put two holds on the calendar: 20 minutes within 48 hours, and the walkthrough of the map after the window we agreed. I write a one-page brief for the person who will do the work. They receive the client's files only after they have signed a confidentiality term. I introduce them on that short call. That is the first time they write to the client.
The looking that follows is read-only. We stay out of the way during the window we agreed. They keep the map after that window, whatever they decide next.
A defined project
NVIDIA asked large financial institutions how they get AI work done, in a survey fielded from August to September 2025. 839 people answered, half in management and half practitioners, across asset managers, banks, broker-dealers, insurers, fintech firms, payments companies, and regulators, among others. A third of the companies report more than $250 million in annual revenue. 40% have more than 500 employees. The survey went out through NVIDIA's lists and social channels.
44% said they bring a third party in for a specific project. 31% said they rely heavily on outside parties for the AI infrastructure itself. The top reason was speed: how fast something goes live, named by 50% of respondents, and by 60% in fintech. Scale of compute was 45%. Cost through shared or managed infrastructure was 42%.
The partners they named were mostly infrastructure. Cloud providers were the answer for 75%. Software vendors were 31%. Equipment makers were 25%. Managed-service firms were 19%. Global integrators were 18%.
A cloud contract and a single workflow are different purchases. The line that travels is the reason, and the shape: one defined piece of work, hired because someone wanted it live sooner.
A form should create the next room. A signature should too. The half hour is that shape, small enough to watch. Inside a firm, the same shape is one workflow with a clear start and a clear handoff.
Source: NVIDIA, State of AI in Financial Services: 2026 Trends. Survey fielded August to September 2025.