Key finding
In the worked example, saving four minutes valued at $45 per hour gives $3 of capacity per eligible question. Covering $500 of monthly running costs requires 167 such questions. Financial ROI depends on how the team uses those hours.
Jump to a section
Pilot worksheet
Would your question volume cover the cost?
Use your own timing and cost estimates. Include the time it takes to open the citation or finish an unanswered question manually.
The example combines $325 per month for the first department with $175 in assumed OpenAI usage and staff review time. You supply the API key and pay OpenAI directly. See LuxonLink pricing.
See the worked example and formulas below.
Monthly estimate
- Staff time
- Capacity value after running costs
- Operating break-even
- Time to recover setup cost
These estimates value time at the rate you enter. Count cash savings separately using actual avoided expenses.
Start with a task that has a clear finish, such as finding who approves an employee’s time off. Stop the timer when a reviewer can point to the supporting passage and act. Stopping when the first sentence appears leaves citation checking and escalation out of the estimate.
Measure both workflows
Choose comparable questions from actual work. Record the start time, finish time, whether someone had to ask a colleague, and whether the answer passed a source check. Alternate which method reviewers try first. Someone who has just read the handbook may remember the answer on a second attempt.
Count unsuccessful assisted questions, including the time needed to finish them manually. Only multiply by the volume of tasks the pilot could realistically handle. The citation field test illustrates a quality check you can run before counting time saved.
Work through a monthly example
These figures are illustrative assumptions. Suppose a team handles 300 eligible questions per month. The current workflow averages six minutes. The assisted workflow averages two minutes including citation review and fallback handling. The team values one staff hour at $45. For this example, it budgets $500 per month in total costs: the LuxonLink base of $325 per month for the first department plus an assumed $175 for OpenAI usage and staff review time. The department supplies its own OpenAI API key and pays OpenAI directly. First-department setup is $1,500, available as two $750 payments at configuration and launch, with 5 GB of storage included. Compare monthly and annual billing and optional storage.
- Capacity recovered. 300 × (6 − 2) ÷ 60 = 20 staff hours per month.
- Value at the assumed rate. 20 × $45 = $900 per month.
- Monthly value after running costs. $900 − $500 = $400 if the team can use all the recovered capacity at that value.
- Operating break-even. Each eligible question frees four minutes worth $3. Divide $500 by $3 and round up to 167 questions per month, before setup.
- Illustrative payback. $1,500 ÷ $400 = 3.75 months under the same assumptions.
At 150 eligible questions, the same four-minute improvement gives ten hours, worth $450 on this basis. The monthly value falls $50 short of running costs. The volume changes the decision even when answer quality stays the same.
Separate capacity from cash
Record where the 20 hours go. For a cash-return calculation, count documented savings such as reduced paid overtime or avoided external spend. If staff use those hours to clear a backlog, report hours and completed work. A salary that stays the same is not itself cash saved.
For a fixed evaluation period, calculate financial ROI as (realized financial benefits − total incremental costs) ÷ total incremental costs. Include setup, maintenance, model usage, and time spent reviewing sources. Keep the original assumptions beside the measured result. This example is a calculation template, not a customer result or a Xillix price quote.