The Five Hidden Cost Drivers
When an AI-augmented team is interrupted, the cost is not the idle hours. It is the compound effect of five distinct cost layers, each invisible in standard billing:
1. Human Ramp-Up Cost
Each interruption forces the team to re-orient. Context is lost. People spend hours rebuilding mental models of where the project was. For a 50-person enterprise team with a 60-day recovery window, ramp-up hours alone represent a six-figure annual cost.
2. Token Waste Cost
Interrupted AI workflows lose cached context. The model must rebuild its understanding from scratch. At the Opus tier, the waste factor is 2.8× — meaning for every euro of productive token spend lost, the rebuild costs nearly three. With frequent interruptions, context-rebuild waste can exceed the steady-state token budget.
3. Momentum Loss Cost
Flow state is not a metaphor — it is throughput. During recovery from an interruption, teams operate at 35% reduced velocity (sprint changes) to 60% reduced velocity (key-person departures). The gap between potential and actual throughput is pure opportunity cost.
4. Recruitment Cost (Key-Person Departure)
When a key person leaves, six months of recruitment cost enters the equation — job postings, interviews, onboarding, the slow ramp to productivity. No per-token or per-seat metric captures this. The deployer’s annual envelope absorbs it silently.
5. Sprint Drag (Key-Person Departure)
The remaining team compensates for missing expertise at 25% reduced efficiency for four months. This is the distributed cost of knowledge loss — every sprint runs slower because tacit knowledge left with the person.
The Superlinearity Finding
The five-driver model reveals a non-obvious relationship: interruption cost is superlinear in model tier and sublinear in team size.
Moving from Sonnet to Opus triples the token-waste component because higher-tier models have larger context windows and more expensive cache rebuilds. But doubling the team size less than doubles the human ramp-up cost because coordination overhead is already priced into the team’s baseline.
This has a direct implication for enterprise cost planning: teams running on expensive frontier models are disproportionately exposed to interruption cost. A single CTO departure in a 50-person Opus-tier team can wipe out months of token-cost savings.
Five Scenarios, Five Realities
| Scenario | Type | Team | Model | Key Finding |
|---|---|---|---|---|
| Startup: Lead dev leaves | Departure | 8 | Opus | 2-month standstill, 45-day recovery |
| Agency: Client reprioritizes | Sprint | 6 | Sonnet | 5 interruptions/month, 3 sprints wasted |
| Enterprise: CTO departs | Departure | 50 | Opus | >30% of annual budget consumed |
| AI Team: Context eviction | Sprint | 4 | Opus Max | 8 interruptions/month, 3.2× waste factor |
| Consulting: Partner exit | Departure | 3 | Sonnet | Tacit knowledge loss, 30-day recovery |
The Owned-Compute Advantage
This is where our cost-surface discipline (the CSB-track from our AI Pricing Capitulation analysis) intersects with interruption economics:
Owned-compute architectures reduce the token-waste component of interruption cost because cached context on owned infrastructure has deterministic lifetime — not the variable eviction policies of metered-API providers.
On a metered API, your cached context competes with every other customer for cache slots. When the provider evicts your cache, you pay the full rebuild cost. On owned EU-sovereign compute (our DGX Spark / Jetson / Acer GN100 stack), the cache lifetime is under your control. The interruption cost of the token-waste component drops to near-zero for planned pauses and stays predictable for unplanned ones.
Try It Yourself
The Keskeytyslaskuri (Interruption Cost Calculator) is live at lifetime.fi/laskuri. Available in five languages (FI, EN, SV, DE, FR). Plug in your team’s parameters, pick a scenario, and see the real cost surface of your next interruption.
Read Also
Market Signal
The AI Subsidy Era Just Ended — and Our Approach Is on the Right Side of It
The demand-side companion. Token-based billing has crossed the enterprise budgeting threshold. Why DWS IQ 6’s owned-compute discipline is structurally on the other side.
Field Note
Baseten Raises $300M — Inference Engineering Is a Discipline Now
The inference cost layer beneath every token-pricing decision. Why the infrastructure matters more than the price tag.