Most internal AI copilots don’t fail loudly.
They fail quietly — by becoming an always-on background cost.
No single request is expensive.
No team abuses it.
But usage compounds.
The Scenario: Internal AI Copilot (500 Employees)
Imagine an AI copilot rolled out internally across a mid-sized company.
Employees use it to:
- Draft documents
- Summarize tickets
- Answer policy questions
- Assist with light coding or data queries
It’s embedded into daily workflows and available to everyone.
Assumptions
- ~500 employees
- 5–10 requests per employee per day
- Usage spread evenly across the month
- Context grows throughout the day
- Retries and re-prompts are normal
Nothing extreme.
Nothing malicious.
Step 1 — Start With Daily Habits
Internal tools don’t spike like customer traffic.
They accumulate.
A few requests during standup.
A few more while writing.
A few follow-ups after meetings.
Individually, each request looks harmless.
Collectively, they define your baseline.
Step 2 — Define a Planning Baseline
In ModelIndex, this is the Expected scenario.
Expected means:
- Consistent daily usage
- Normal prompt iteration
- Gradual context growth
This is the number finance and engineering should both be comfortable with.
Not optimistic.
Not worst-case.
Step 3 — Identify the Hidden Cost Drivers
Internal copilots surface different risks than external systems:
- Long-lived context windows
- Repeated “minor” prompts
- Higher output verbosity
- Light retries that go unnoticed
Cost doesn’t jump.
It creeps.
Step 4 — Explore Best and Worst Boundaries
Now look at Best and Worst.
These are not usage tiers or performance modes.
They exist to answer one question:
How does cost behave as habits drift over time?
- Best assumes short prompts and disciplined usage
- Worst assumes growing context, verbose outputs, and habitual re-prompts
Worst isn’t abuse.
It’s normal usage left unchecked.
Step 5 — Ask the Right Question
The useful question is not:
Can we afford this per employee?
It is:
What happens when this becomes invisible infrastructure?
That’s when cost surprises show up.
Why This Matters
Internal AI tools feel cheap because no one sees the bill directly.
But once deployed broadly, they become a permanent operating cost.
Modeling that cost upfront lets teams:
- Set expectations
- Add guardrails intentionally
- Decide whether to ship, limit, or defer
Before the surprise shows up in finance.