The four-week rule.

If a project takes longer than four weeks to ship, we don't take it.

We get pushback on this every other discovery call. Six month roadmaps look serious in a deck. They die in real life. The four-week rule is the single most important filter we run a project through, and it has saved more client relationships than any technical decision we have ever made.

Most operators we work with run businesses with 30, 60, 90 day rhythms. Quarterly reviews. Monthly board updates. Weekly all-hands. The work that earns trust is the work that produces a result inside one of those cycles.

A project that takes six months produces nothing for five and a half of them. That is not building. That is consulting theatre billed by the hour.

The four-week number is not arbitrary. Every project we have shipped that succeeded was scoped to under four weeks of build time. Every project we declined that ran past that line, when we checked back, had stalled, gotten cancelled, or quietly turned into a perpetual retainer that nobody could point to a result for. So we made it a rule.

Shipped means the system is in production. Real users or real operators are using it. The numbers it produces are being read by the people who asked for it.

Not deployed to staging. Not demoed on a call. Not ready for the next sprint. Used.

This raises the bar on scoping more than it raises the bar on building. If something cannot be in someone's hands and changing their workday in four weeks, we cut features until it can.

• Week 1. Discovery, real data, real workflow mapping. No code yet.
• Week 2. Build the smallest version that solves the actual problem.
• Week 3. Integrate with the systems they already use. Slack, HubSpot, the CRM, the data warehouse, whatever the operator opens every morning.
• Week 4. Hand-off, training, documentation. The numbers start landing.

If any one of those weeks slips, the whole timeline slips. The discipline is in week 1. If discovery surfaces a fuzzier problem than what was in the proposal, we say it then. We do not push it three weeks deeper and hope it resolves itself. It never does.

Some clients hear four weeks and want longer. They want a roadmap, multiple phases, a six month engagement they can present to the board.

Two patterns show up:

• They have been burned before by an agency that promised fast and shipped slow. Adding more time is their hedge.
• They are using the project to justify a hire or a budget line that needed a bigger number to clear procurement.

The first is fixable. We show the contracts, the past projects, the actual delivery dates. The trust shows up. The second is not our problem to solve. We pass.

The rule breaks when the problem genuinely needs more than four weeks of build. Real RAG systems on top of large unstructured corpora. Multi-agent orchestrations with state machines. Custom model fine-tuning loops. These are not nights-and-weekends builds.

In those cases we do not extend the project. We split it.

Phase one ships in four weeks with a working prototype on a smaller dataset. Phase two builds on the prototype with the real production data. Phase three handles scale. Each phase has its own ship date. Each phase produces something the operator can use. Nobody waits six months for a deliverable.

The rule does two things to us, the agency.

First, it forces us to stay honest about scope. We cannot pad a project with research weeks or exploratory phases. If it does not contribute to a four week ship, it does not go in. That is a constraint on us, not a license for us.

Second, it filters the client list. Operators who want to ship find us. Operators who want to procure consulting theatre go elsewhere. Both groups get what they wanted. Everyone is happier for it.

The clients who push back on this rule the hardest are the ones who needed it the most.

If your AI initiative has been in flight for more than four weeks without something a real person uses every day, the project is not real. It is a sunk cost wearing a different name.

Cut the scope until it fits inside a quarter you can actually feel. Ship that. See if the numbers move. Then keep going. Or stop and start over. Both are better than the third option.