Catalyst StudioCatalyst Studio

In the open

How We Operate

This is how Catalyst runs its own business: the method behind every audit, how the work is organized, how AI is kept in check, and how we build software. We’re publishing more of it over time, and holding back only the handful of things we’d lose if a competitor had them.

01

The method

The same five steps run against a different part of the business each time, whichever function is being diagnosed. What counts as “spending without a return” is defined once, by the Waste Index, so a Sales finding and a Finance finding add up on the same terms instead of each function inventing its own math.

  1. 01Collect the evidenceWork from what the business produced: the records and the output.
  2. 02Find the patternWhere things wait, repeat, get handed off, or quietly leak.
  3. 03Put a number on itAnything that can’t be measured stays an open question.
  4. 04Name the causeWhat’s causing it underneath the symptom, and what would prove that wrong.
  5. 05Decide what changesFix it, investigate further, or leave it alone. A person makes the call.
02

How the work is organized

Four layers, stacked on top of each other: what gets set once, what we’re steering toward, what holds the work, and the work itself.

  1. FoundationNorth StarMissionVisionCore ValuesBusiness Goals
  2. StrategyStrategies→Goals→Initiatives→Experiments
  3. Where work is groupedProgramsProjects
  4. Work itemsTasksRoutinesListsCommitmentsNotes

Programs and projects both hold the same kinds of work items, so work is described the same way wherever it shows up.

03

How AI is kept in check

Every layer above is something an AI assistant might touch. This is what keeps that in check: what it’s allowed to use, how often it checks in, and how far it can go before it needs you to say yes.

  1. ToolsEach AI assistant has a fixed set of tools, with no open access to your data. Everything it does is checked against what you’ve allowed it to touch.
  2. AccessNo access at all until you agree to let AI help. Then it can read broadly, and create and edit its own kinds of work: tasks, lists, routines, and goals. Your client and customer records stay read-only until you give more access, one kind of record and one permission at a time.
  3. Always-onSome assistants run all the time, others on a schedule, watching for things starting to slip and telling you as it happens.
  4. ApprovalFour steps, and it stops there by design: read, suggest, draft, propose. Almost nothing important happens without your approval of that specific item. The one exception, stated openly: a single low-risk kind of recommendation, with a spending cap, applies automatically without waiting for you.

This is what runs underneath the platform. See how the assistants fit together.