Legal AI Cost Modeling · V2

Forecasting Tool Calculating Token Based Costs Per Month

Customize the inputs below to forecast your firm's pre-tax monthly costs.

PRISM by Feit Consulting
User Counts & Fixed Costs
Assign & review work
Complex analysis, heavy use
Routine support, moderate use
Base platform access fee
Avg. billable days per month
Model Usage Mix — % of each group's usage per model · rows must total 100%
Snaps to cost-optimised Haiku/Sonnet weighting based on Harvey's routing strategy — instantly see potential savings
User Type Opus 4.8
Ambitious work		 Sonnet 4.6
Everyday tasks		 Haiku 4.5
Quick answers		 Total
Partners % % % 100%
Attorneys % % % 100%
Paralegals % % % 100%
Consumption Assumptions — per model, adjust to match your firm's expected workflow intensity
Opus 4.8 Sonnet 4.6 Haiku 4.5
Tasks / User / Day
Model Calls / Task ?
Tokens / Call (avg) ?
↳ Tokens / User / Day 288,000 45,000 12,800
↳ Cost / User / Day $2.59 $0.14 $0.01
⚡
The Agentic Multiplier — This is the number that matters most One attorney "task" can trigger 8–20 model calls as agents loop, re-read documents, and verify outputs. This single row drives more of your monthly bill than any other input — including the number of users. Stress-test your forecast by pushing "Model Calls / Task" to 12–15 for Opus-heavy workflows.
Opus 4.8
Most capable · ambitious work
$ / 1K tokens
Sonnet 4.6
Efficient · everyday tasks
$ / 1K tokens
Haiku 4.5
Fastest · quick answers
$ / 1K tokens
⚠ Token prices are rising. Frontier models (Opus 4.8) have seen price increases in 2025–26, and with OpenAI/Anthropic holding near-monopoly positions on enterprise usage, further increases are likely. Build in a 20–30% annual escalation factor when projecting Year 2 costs — use the Price Escalation input in the Learning Curve section below. Confirm current enterprise rates with your vendor before presenting to clients.
Monthly Cost Breakdown

Fixed Costs

Total Licenses1,000
License Fees$20,000

Consumption by User Type

Partners (100)—
Attorneys (600)—
Paralegals (300)—
Total Consumption—

By Model

Opus 4.8—
Sonnet 4.6—
Haiku 4.5—
Estimated Monthly Invoice (pre-tax)
$0
Fixed + Consumption · 22 working days
Per license/mo: $0.00
Learning Curve Forecast — efficiency improves as users master prompting, but usage expands as AI embeds deeper into workflows

Phase 1 · Onboarding

Months 1–2 · This IS your baseline
M1 = Invoice Box Above
Phase 1 is your current assumptions with no adjustment. M1 and M2 on the chart always match the estimated monthly invoice above. Efficiency gains start in Phase 2.

Phase 2 · Developing

Months 3–6 · Efficiency gains emerge

Phase 3 · Proficient

Months 7–12 · Steady state · tighter prompts
% increase in tasks/day as adoption deepens
Users defaulting to Haiku for routine queries
Controls rate of phase transitions
Default 25% — AL article recommends 20–30%
Scenario Comparison

No scenarios saved yet. Adjust inputs and click "Save Scenario" to compare.

Est. Monthly Invoice
$0
$0.00
per license / month