About this calculator (V2 update)
V2 credit: the false positive cost and portfolio view were both raised as missing in V1 by Ricardo Vieira-Gomes (Co-Founder & Executive Director at ET Armadillo, AI Transformation in Operations) on the launch post. Both critiques were exactly right — they're now built in.
V1 of this calculator answered one question: is this single queue making or losing money? Useful, but the question most ops leaders actually face is harder.
First, accuracy alone misses the cost of wrong cancellations. When an agent blocks a legitimate order, you don't just fail to save the cost of that order — you also lose the lifetime value of the customer who never comes back. V2 adds false positive rate and average customer LTV as inputs, then subtracts that cost from gross savings to give you net ROI.
Second, ops leaders don't run a single queue. They run multiple in parallel — different case types, different markets, different products. The right decision isn't "is queue X profitable?" but "given my next dollar (or next agent), where does it generate the most savings?" V2 portfolio mode lets you model up to 3 queues side by side and surfaces the queue with the highest per-agent net ROI, which is where the marginal agent goes.
Real-world impact
Illustrative scenarios drawn from operator practice. Numbers are realistic order-of-magnitude estimates, not measurements from any specific deployment.
Case 1: Fraud review queue running 3 months in the red
SetupAn 8-agent manual review team at a LATAM marketplace screens 200 flagged orders per day at $30 average order value.
ProblemNet ROI looked positive on paper at 18% catch rate, but false-positive cost was quietly burning roughly $1,400 per day in lost legitimate customer LTV.
Tool surfacedPortfolio mode flagged FP cost at $4.50 per agent-day against gross savings of $3.20, recommending root-cause work on the rule set rather than more headcount.
OutcomeDecision rules rewritten, FP rate dropped from 6% to 2.5%, net ROI flipped to +0.8x, around $280K saved annually vs the "add 2 more agents" plan that was already approved.
Case 2: VP picking between three queues for one new hire budget
SetupA US consumer fintech VP of Ops has budget for exactly one new analyst across three queues (account takeover, disputes, KYC review).
ProblemEach queue manager was lobbying with their own spreadsheet, decisions were political, and the wrong placement would waste roughly $95K fully loaded for the year.
Tool surfacedPortfolio comparison ranked marginal ROI per queue: KYC at 2.1x, ATO at 1.4x, disputes at 0.6x because dispute auto-decision coverage was already high.
OutcomeHire went to KYC, the dispute queue got a process review instead, combined coverage gain came out at +$140K vs the politically favored allocation.
The formula (V2)
true positives = orders × accuracy / 100
false positives = orders × FP rate / 100
gross savings = true positives × avg savings per cancellation
FP cost = false positives × avg customer LTV
net savings = gross savings − FP cost
net ROI = net savings ÷ daily cost
Tier classification (V2 net basis)
Net ROI ≥ 1.0 — Scale
Each $1 in agent cost generates $1+ in net savings after FP cost. Add headcount, expand case mix.
0.5 ≤ Net ROI < 1.0 — Optimize
Marginal. Tune one lever before scaling. Often FP rate or accuracy is the bottleneck.
Net ROI < 0.5 — Root cause
Negative or thin. Adding agents amplifies loss. Investigate FP rate, rule precision, or whether the queue belongs at all.
About the FP cost model
The model is intentionally simple: FP cost = false positives × avg LTV. Real-world LTV impact is more complex (depends on how customers react to a wrongful block — some come back, some don't, some leave negative reviews that affect others). For a back-of-envelope decision, this is good enough. For production, you'd want to validate the LTV assumption against your actual cohort data.
About the portfolio "marginal agent" recommendation
The portfolio mode ranks queues by per-agent net ROI and surfaces the highest as "deploy next agent here." This assumes the marginal agent has roughly the same productivity as the current team. In practice, diminishing returns exist (the 50th agent is less productive than the 5th), but at the scale most ops teams operate, linear assumptions hold over short horizons. Use the recommendation as a starting point, not a final answer.