Net return / wk
$27,283
$31,796 displaced, $359 spent
Net return / yr
$1.42M
52 weeks at current run rate
Hours saved / wk
513
Successful calls only
Capacity returned
13.7 FTE
At a 37.5-hour week
ROI
7.0×
Returned per $1 of tokens
Deflection rate
98.5%
Completed without a human
The arithmetic
Successful calls only, converted into the discrete human tasks they actually displace, multiplied by the minutes each task takes, multiplied by the loaded hourly rate of the person who would otherwise have done it, minus everything the business pays to get that return: the metered agent cost and what Sentry itself costs.
tasks displaced×minutes per task×$62 / hour, loaded−agent cost $359−Sentry $4,154=$27,283 / week
- Only successful calls count. A failed call creates work rather than removing it: somebody has to pick it up and finish it by hand. Counting it as a saving would be counting the same hour twice. This is also why a quality regression shows up here as lost value, not just as a lower percentage on a chart.
- Not every call is a human task. Most agent calls are the agent's own internal steps: retries, follow-up turns, tool round-trips. Counting all 26,430 weekly calls as avoided human work is how a 600× return gets manufactured. We count the 3,271 discrete tasks they genuinely displace, at a rate agreed with you per agent and revisited each quarter.
- Sentry's own cost is in the denominator. Comparing labour against a token bill alone produces an absurd multiple, because tokens are nearly free. What you pay us ($18,000 / month) is part of what it costs to get this return, so it is subtracted here rather than hidden.
- Agent cost is metered, not modelled. It is the tokens the platform observed this agent burn, priced at what you actually pay.
- There is no vendor multiplier. Nothing on this page is inflated by a productivity coefficient. Every figure traces back to a call the platform saw.