CEO Column | C7 | Milestone Design: When to Double Down and When to Cut Losses
A good capital structure doesn’t make you rush forward; it makes “decisions” simple.
Last month I met with government agencies and a fund in Brno, Czech Republic. We laid the entire PSF compute center timeline, engineering, operations, talent and energy agreements on the table. Halfway through, a fund partner asked a crucial question: “If something goes wrong, which milestone tells us to stop, and which one tells us to double down?” I rarely see a single sentence freeze the air in a meeting so instantly. Because the core of that sentence is this: what the capital market truly values is not growth but “decision nodes.” You cannot hit the accelerator by feel, and you cannot hit the brakes by hope. You need a “company-level” structure that makes every decision so simple that it needs no argument.
01 | Decision-Making in the AI Era: Not Discussion but Gates
In the past:
* Managers discussed
* Teams argued
* Investors went by feel
* Founders went by intuition
But not now. An AI agent will not listen to your story; it will only ask: “Has the next Gate been achieved?”
Not achieved → stop.
Half achieved → downgrade.
Achieved ahead of schedule → double down.
This is the cleanest way of governing.
02 | What Is a Gate? Not a Date but a “Verifiable State”
You cannot say:
* Q3: complete the server room
* Q4: complete deployment
This is a “schedule,” not a “milestone.”
A real milestone has three conditions:
1. Input
What resources will be used?
2. Output
What exactly is delivered?
3. Verification
Who decides pass or fail, and on what basis?
If any one is missing, it is not a milestone, just a “progress chart.”
03 | PSF’s Eight-Milestone Design Method (You Can Use It Directly)
This is the formula we now use in all projects in the Czech Republic, Japan, Singapore and the UAE. It is also the version funds and governments buy into most.
M1 | Legitimacy Gate
Acceptance: government, regulation, energy, land and building permits all in place
If it cannot pass → the project stops immediately.
This is the required “0/1 Gate.”
M2 | Resource Gate
* Equipment lead times
* GPU allocation
* Supplier contracts
* Energy price fixed
**Acceptance: all resources can be quantified + traced**
Unconfirmed → no construction.
M3 | Construction Gate
* Server-room structure
* Redundancy
* Power supply
* Cooling
**Acceptance: risk matrix > 85% controllable**
Below standard → the project is downgraded, no doubling down.
M4 | Workflow Gate
* SOPs
* SLAs
* Audit
* Rollback
* Tiered permissions
**Acceptance: all processes can be interpreted by an Agent (machine-readable)**
Passing this gate means the company can expand.
M5 | Model Gate
* Version control
* Data zoning
* Drift monitoring
**Acceptance: no cascading-incident risk from the model**
Not met → cannot proceed to putting customers live.
M6 | Delivery Gate
* Customer onboarding
* SLA attainment rate
* Complaint ratio
**Acceptance: at least 90% stable delivery**
Pass this gate → funds usually double down.
M7 | Financial Gate
* Cash flow
* Utilization
* Break-even
**Acceptance: can withstand 2 major shocks**
This is the key to whether the company can “survive.”
M8 | Scale Gate
* Second site
* Cross-border replication
* Department automation
**Acceptance: the whole system can be replicated without relying on the founder**
This is when the multiple jumps.
04 | Doubling Down vs. Cutting Losses: Here Is the Formula (the Most Important Part)
You must have clear, mechanical rules.
When do you double down? When at least two conditions are met:
* The next Gate is achieved early
* SLA is above expectations
* Customer renewal rate > 90%
* Incident rate declines
* Cash flow increases
* Agents can run the process automatically ≥ 70%
This is called “the strong get stronger.”
When do you slow down / downgrade? Any one of these conditions is enough:
* More than two version incidents
* Metrics become more volatile
* Engineering delay of more than 30 days
* Energy cost changes by more than 20%
* Key talent turnover
* New compliance restrictions
* Incomplete zoning
* Unpredictable Agent behavior
This is called “downgrade mode.”
When do you cut losses? If any one holds → stop immediately:
* M1 (Legitimacy Gate) not achieved
* Supply chain disruption with no alternative
* Cross-border compliance that cannot be fixed
* Finances that cannot return to a positive track
* Security incidents (data leaks, systemic loss of control)
This is called “death mode.”
Cutting losses is for survival; doubling down is for positioning.
05 | Investors in the AI Era Don’t Ask You “Can You Succeed” — They Ask:
“Do the Gates you designed ensure you won’t die?” That sentence matters more than any KPI, because real corporate governance is not about doing things fast, growing the market big or telling a beautiful story; it is about turning uncertainty into a series of “decidable doors.” Which door you pass, you double down; which door you are stuck at, you slow down; if you hit the door, you cut losses. No emotion, no luck, no assumptions — only a system.
06 | Conclusion: The Real Milestone Is Not Progress but a “Node”
I give you this line: “Milestones are not for reporting to customers; they are for keeping you alive.” Because in the future all cooperation will become: you → Gate check → Agent → investor → decision.
In the end AI will reach a conclusion about a company:
> **“Clear decision gates → investable.”**
> **“Unclear decision gates → high risk.”**
This is the capital logic of the AI era: not advancing by time, but surviving by Gate.
