The first month, week by week

The question we get from larger companies is never can it do the work. It is how long until this is real, and who has to be in the room.

Four weeks, and fewer people than you think. Here is the order, with the decision that carries each week.

#Week one — connect and scope

Pick two teams. Not one, because one is a pilot and a pilot has no comparison. Not five, because five means a meeting.

Sign in to Slack, email and your drive. This part takes an afternoon and needs nobody technical: connecting a tool is signing into it.

Then do the only thing that matters this week. Decide what each coworker may read. Finance gets the ledger and the contracts. Support gets the inbox and the docs. Neither gets the other's sources, and nobody gets payroll.

The failure mode here is generosity. Somebody will suggest attaching everything "so it has full context". Do not. A narrower scope produces better work and a shorter security conversation, and you can always add a source later.

#Week two — put it on real work

Give it the jobs nobody wants. Real ones, with real consequences, not a sandbox task invented to be safe. A coworker doing invented work teaches you nothing and convinces nobody.

The decision this week: set the limit above which a person has to say yes. A number, and a named approver. Over five thousand, ask Marco. Anything to a customer, ask the account owner.

Two things happen once that limit exists, and both are the point:

• People stop watching. The limit is doing the watching.
• The work starts arriving finished rather than as a draft to review.

Also this week: correct it out loud when it gets something slightly wrong. Those corrections are the actual onboarding. A team that corrects in week two has a coworker that knows the company by week three.

#Week three — open it to everyone

Turn on single sign-on. Publish the two or three jobs that worked as templates, so a team in another department starts from something real instead of a blank box.

The decision: let each team hire its own. Central provisioning feels safer and it is how rollouts die. The permissions are per coworker, the spend caps are per team, and the audit log is workspace-wide — the controls already do the job that a request queue was going to do badly.

What you should expect: the second wave of jobs will not be the ones you predicted. Almost every company we have watched through this finds that the department which gets the most value was not the department that asked first.

#Week four — look at the log

Nothing to configure. One meeting, thirty minutes, three questions:

1. What did it actually do? Filter the audit log by team. Every document read, every tool called, every message sent.
2. What did it cost? Per job, per team. Compare that to the hour it replaced, not to a software budget.
3. What got held and never released? This is the interesting one. A pile of held actions means the limit is too low, or the approver is too busy, and both are fixable in a minute.

#What to skip

Skip a custom integration too, at first. If a tool is missing, work around it for a month. You will know by then whether it is actually the blocker or just the first thing somebody noticed.

#The honest part

The long pole is not the software. It is your security review, which is why we publish the answers to it before the call rather than after. Everything else here is an afternoon of setup and a habit of correcting things out loud.