Maintenance and Total Cost of Ownership

Upkeep over the full horizon

After launch there are ongoing costs: prompt and model maintenance as providers update models, integration fixes when upstream systems change, evaluation re-runs, monitoring and on-call, vendor licences, and periodic retraining of staff. Total cost of ownership (TCO) adds one-off and recurring costs over the full horizon (for example three years). TCO, not launch cost, is what the benefits must beat.

Three-year total cost of ownership, run

I ran this with plain Python 3. All figures belong to one worked example (an invoice-processing team) with invented but internally consistent numbers; prices are placeholders. With $53,000 one-off and $3,016 per month for running, monitoring and maintenance, the three-year total cost of ownership is $161,576, three times the launch cost.

one_off = {"discovery & design": 8000, "build & integration": 30000, "evaluation set & testing": 6000,
           "change management & training": 5000, "security & legal review": 4000}
monthly = {"run cost (LLM, OCR, infra)": 916, "human review time": 0, "monitoring & on-call": 900,
           "prompt/model maintenance": 1200}
print(f"one-off total: ${sum(one_off.values()):,}")
print(f"monthly total: ${sum(monthly.values()):,}  (review time counted in the savings estimate)")
print(f"3-year total cost of ownership: ${sum(one_off.values()) + 36 * sum(monthly.values()):,}")

Output:

one-off total: $53,000
monthly total: $3,016  (review time counted in the savings estimate)
3-year total cost of ownership: $161,576

Budget for model changes

Providers retire and update models; include time for re-evaluation and prompt updates every year.

Quick check: What does total cost of ownership include?

• Only the launch cost
• One-off and ongoing costs over the full time horizon
• Only token fees
• Only salaries of the AI team