AI Cost Modeler
Overview
Estimate, track, and optimize AI-related costs across the organization. Model total cost of ownership and build business cases with realistic financial projections.
Core principle: AI costs are often underestimated. Comprehensive modeling prevents budget surprises and enables informed decisions.
When to Use
- Planning new AI projects
- Building business cases
- Budgeting for AI programs
- Optimizing existing AI costs
- Vendor negotiations
Output Format
ai_cost_model:
project: "[Project name]"
date: "[YYYY-MM-DD]"
timeframe: "[Months/Years]"
summary:
total_year_1: "[$]"
total_year_3: "[$]"
monthly_run_rate: "[$]"
cost_per_transaction: "[$]"
development_costs:
one_time:
- category: "Labor"
description: "Development team"
calculation: "[FTEs × months × rate]"
amount: "[$]"
- category: "Infrastructure"
description: "Initial setup"
amount: "[$]"
- category: "Data"
description: "Data preparation, labeling"
amount: "[$]"
- category: "Vendor"
description: "Implementation fees"
amount: "[$]"
total_development: "[$]"
operational_costs:
recurring:
- category: "Inference/API"
description: "[Provider] API calls"
unit_cost: "[$X per 1K tokens]"
volume: "[Projected volume]"
monthly: "[$]"
annual: "[$]"
- category: "Compute"
description: "Hosting, training"
monthly: "[$]"
annual: "[$]"
- category: "Storage"
description: "Data, models, logs"
monthly: "[$]"
annual: "[$]"
- category: "Labor"
description: "Ongoing support"
fte_equivalent: "[X FTEs]"
annual: "[$]"
- category: "Vendor"
description: "Subscriptions, licenses"
annual: "[$]"
total_annual_operational: "[$]"
scaling_projections:
- year: 1
volume: "[Usage]"
cost: "[$]"
- year: 2
volume: "[Usage]"
cost: "[$]"
- year: 3
volume: "[Usage]"
cost: "[$]"
roi_analysis:
benefits:
- benefit: "[Benefit description]"
type: "[Hard savings | Soft savings | Revenue]"
annual_value: "[$]"
confidence: "[High | Medium | Low]"
total_annual_benefit: "[$]"
payback_period: "[Months]"
year_3_roi: "[%]"
sensitivities:
- variable: "[What might change]"
base_assumption: "[Current assumption]"
impact_if_doubles: "[$ impact]"
recommendations:
cost_optimization:
- "[Optimization opportunity]"
Cost Categories
Inference Costs (LLMs)
llm_cost_drivers:
input_tokens:
rate: "[$X per 1M tokens]"
factors:
- "Prompt length"
- "Context window usage"
- "RAG retrieval size"
output_tokens:
rate: "[$Y per 1M tokens]"
factors:
- "Response length requirements"
- "Streaming vs batch"
volume:
daily_calls: "[N]"
avg_tokens_per_call: "[Input + Output]"
monthly_projection: "[Total tokens]"
example_calculation:
api: "GPT-4"
input_rate: "$30/1M"
output_rate: "$60/1M"
daily_calls: 10000
avg_input: 500
avg_output: 200
daily_cost: "$15 + $12 = $27"
monthly_cost: "$810"
annual_cost: "$9,720"
Compute Costs (ML)
compute_cost_drivers:
training:
gpu_type: "[A100, H100, etc.]"
hours_per_run: "[N]"
runs_per_month: "[N]"
hourly_rate: "[$]"
monthly: "[$]"
inference:
instance_type: "[Type]"
instances_needed: "[N]"
hours_per_month: "[N]"
monthly: "[$]"
development:
notebooks: "[$]"
experimentation: "[$]"
Data Costs
data_costs:
storage:
training_data: "[GB] × [$X/GB]"
model_artifacts: "[GB] × [$X/GB]"
logs_monitoring: "[GB] × [$X/GB]"
preparation:
labeling: "[Hours] × [$/hour]"
cleaning: "[Effort estimate]"
enrichment: "[Vendor costs if any]"
acquisition:
external_data: "[$]"
api_costs: "[$]"
Labor Costs
labor_costs:
development:
data_scientist: "[N FTE] × [$/year]"
ml_engineer: "[N FTE] × [$/year]"
product_manager: "[N FTE] × [$/year]"
duration: "[Months]"
operations:
mlops_engineer: "[N FTE] × [$/year]"
support: "[N FTE] × [$/year]"
overhead:
management: "[%]"
training: "[Hours × rate]"
ROI Calculation
Benefits Categories
| Type |
Examples |
How to Quantify |
| Hard savings |
Labor reduction, error reduction |
FTEs × salary, error cost × reduction % |
| Soft savings |
Time savings, productivity |
Hours saved × rate × adoption % |
| Revenue |
New capability, faster deals |
Revenue × attribution % |
| Risk avoidance |
Compliance, quality |
Potential cost × probability reduction |
ROI Formula
ROI = (Total Benefits - Total Costs) / Total Costs × 100
Payback Period = Total Investment / Monthly Net Benefit
Example Calculation
roi_example:
costs:
development: 200000
year_1_operations: 100000
year_1_total: 300000
benefits:
labor_savings: 400000 # 5 FTEs reassigned
error_reduction: 50000
year_1_total: 450000
year_1_net: 150000
year_1_roi: "50%"
payback_months: 8
Optimization Strategies
Inference Optimization
| Strategy |
Potential Savings |
| Prompt compression |
20-40% on input tokens |
| Response length limits |
Variable on output |
| Caching common queries |
10-30% on volume |
| Model tiering (smaller when possible) |
50-80% on some calls |
| Batch processing |
10-20% vs real-time |
Compute Optimization
| Strategy |
Potential Savings |
| Spot instances |
60-90% on training |
| Right-sizing |
20-50% on inference |
| Auto-scaling |
Match cost to demand |
| Reserved capacity |
30-50% on steady state |
Sensitivity Analysis
sensitivity_scenarios:
base_case:
volume: 100000
cost_per_call: 0.01
monthly: 1000
volume_doubles:
volume: 200000
monthly: 2000
change: "+100%"
price_increases_50:
cost_per_call: 0.015
monthly: 1500
change: "+50%"
both:
monthly: 3000
change: "+200%"
Checklist
