What each number is
Coverage, durable resolution, net automation and the bridge from the vendor claim are arithmetic on your inputs. Eligibility, resolution, repeat rate and the escalation premium are assumptions about the bot; the opening profile is labelled heuristic and none reaches Planning-grade until a document or observed data backs it. Net savings are a conditional forecast: freed capacity becomes cash only through the capacity action you select, while operating cost and the escalation premium are cash out in full.
Formulas
Coverage = eligible demand ÷ total demand. Apparent resolution = resolved ÷ AI-involved conversations. Net automation = durable resolutions ÷ total demand
They are never shown as one another. The gap between apparent resolution and net automation is where most overstatement starts.
Apparent resolution × (1 − repeat and false resolution rate)
A resolution that comes back was a deferral.
Contacts × eligible share × durable resolution
Everything else the bot touched reaches a human: immediate escalations and returns.
Marginal cost × contacts × eligible × (durable × action share − (1 − durable) × escalation premium) − operating cost
Operating cost is platform, QA, tuning hours × rate and knowledge hours × rate. Marginal cost, when not supplied, is 60% of loaded and disclosed.
Contacts × apparent resolution × loaded cost
The naive slide. The bridge walks from it to net savings in six steps that sum exactly: eligibility gap, repeats, loaded to marginal, capacity not converted, escalation premium, operating cost.
(premium + operating cost ÷ (marginal × contacts × eligible)) ÷ (action share + premium) ÷ (1 − repeat rate)
The apparent resolution at which net savings reach zero. Above 100% it never breaks even.
1 − durable resolution at break-even ÷ apparent resolution
The highest repeat rate the program can carry and still break even.
Month m earns steady savings × min(1, m ÷ ramp months) − operating cost; year one starts at minus the one-time cost
Payback is the first month cumulative net turns positive. Year one at full run rate is shown beside it, so the cost of the ramp is visible.
Rules and bands
The upside case lifts resolution 1.2 times (capped at 100%) and halves repeats (0.5).
A better floor or a lower price could turn it.
Knowledge coverage and intent scope are the constraint.
A pilot earns the evidence, and a thin margin should cost the vendor, through the contract.
Put the resolution rate in the contract with a remedy.
Display only.
The verdict is a property of the answer and never caps a confidence axis. A bot at or under $0.01 of operating cost per attempted conversation holds completeness Directional.
Every constant and where it comes from
Default so the tool opens on a runnable case. A volume still at this value grades evidence Directional.
Default so the tool opens on a runnable case. Loaded cost moves the vendor claim and moves net savings by exactly zero, so it reaches no confidence axis.
Default so the tool opens on a runnable case. Zero means not supplied. The tool then derives marginal cost from loaded and grades cost evidence Directional.
Default so the tool opens on a runnable case. An eligibility still at this value grades evidence Directional.
Default so the tool opens on a runnable case. Ramp length when the ramp is on. It moves Year 1 and payback only.
Default so the tool opens on a runnable case. Assumption set A, the graded set. A resolution rate still at this value grades evidence Directional.
Default so the tool opens on a runnable case. Assumption set A, the graded set. A repeat rate still at this value grades evidence Directional.
Default so the tool opens on a runnable case. Assumption set A, the graded set. No single published figure exists. An escalation premium still at this value grades evidence Directional.
Default so the tool opens on a runnable case. Assumption set A, the graded set. Ships at zero and is disclosed as optimistic while it stands.
Default so the tool opens on a runnable case. Assumption set A, the graded set. A platform fee still at this value grades cost evidence Directional.
Default so the tool opens on a runnable case. Assumption set A, the graded set. Bot quality assurance cost.
Default so the tool opens on a runnable case. Assumption set A, the graded set. Bot tuning effort.
Default so the tool opens on a runnable case. Assumption set A, the graded set. Bot tuning labor rate.
Default so the tool opens on a runnable case. Assumption set A, the graded set. Knowledge maintenance effort.
Default so the tool opens on a runnable case. Assumption set A, the graded set. Knowledge maintenance labor rate.
Default so the tool opens on a runnable case. Assumption set B, shown for comparison and never graded. A resolution rate still at this value grades evidence Directional.
Default so the tool opens on a runnable case. Assumption set B, shown for comparison and never graded. A repeat rate still at this value grades evidence Directional.
Default so the tool opens on a runnable case. Assumption set B, shown for comparison and never graded. No single published figure exists. An escalation premium still at this value grades evidence Directional.
Default so the tool opens on a runnable case. Assumption set B, shown for comparison and never graded. Ships at zero and is disclosed as optimistic while it stands.
Default so the tool opens on a runnable case. Assumption set B, shown for comparison and never graded. A platform fee still at this value grades cost evidence Directional.
Default so the tool opens on a runnable case. Assumption set B, shown for comparison and never graded. Bot quality assurance cost.
Default so the tool opens on a runnable case. Assumption set B, shown for comparison and never graded. Bot tuning effort.
Default so the tool opens on a runnable case. Assumption set B, shown for comparison and never graded. Bot tuning labor rate.
Default so the tool opens on a runnable case. Assumption set B, shown for comparison and never graded. Knowledge maintenance effort.
Default so the tool opens on a runnable case. Assumption set B, shown for comparison and never graded. Knowledge maintenance labor rate.
Marginal cost derived from loaded when none is supplied. Disclosed on the page and grades cost evidence Directional.
Sensitivity band printed around net savings when the resolution rate is an internal estimate. Display only. It reaches no confidence axis.
Sensitivity band when the resolution rate comes from vendor marketing. Display only.
Sensitivity band when the resolution rate comes from a proposal or SOW. Display only.
Sensitivity band when the resolution rate is a contracted floor with a remedy. Display only.
Sensitivity band when the resolution rate was observed in the user's own environment. Display only.
Upside case resolution, capped at 100 percent. Feeds the verdict's upside test only. It reaches no confidence axis.
Upside case repeat rate. Feeds the verdict's upside test only.
Conservative scenario eligibility. Shown in the scenario table, and since method 1.2 a Proceed needs the Conservative scenario to break even. It reaches no confidence axis.
Conservative scenario resolution. Shown in the scenario table and feeds the verdict's downside test. It reaches no confidence axis.
Conservative scenario repeat rate, capped at 100 percent. Shown in the scenario table and feeds the verdict's downside test. It reaches no confidence axis.
Marginal cost at or above this share of loaded usually means loaded cost was entered twice. Disclosed, and holds completeness Directional.
Eligibility at or above this is rare outside narrow scopes. Disclosed on the page. Framing only.
Apparent resolution at or above this is uncommon outside narrow FAQ scopes. Disclosed on the page. Framing only.
Eligibility below this routes the verdict to fixing the foundation. A property of the answer. It reaches no confidence axis by doctrine.
Operating cost at or below one cent per attempted conversation makes the program look costless and drives break-even toward zero. Set below the channel fee line because this cost is a flat monthly spend spread across volume, and a real program at scale runs a few cents. Holds completeness Directional.
Worked example
Computed by the tool's own engine at its opening case: assumption set A, a 6-month ramp, the resolution rate sourced to an internal estimate, no marginal cost supplied, and avoided hiring as the capacity action (75% of freed capacity). The tool itself opens with no action chosen, which realizes $0.
Demand and cost: 80,000 contacts a month, $7.00 loaded; marginal not supplied
Bot: 55% eligible, 65% apparent resolution, 18% repeat, 25% escalation premium
Operating cost: $8,000 platform, $2,000 QA, 40 tuning hours at $65, 20 knowledge hours at $55; no one-time cost
60% of $7.00 = $4.20, disclosed
65% × (1 − 18%) = 53.3% of 44,000 attempted = 23,452 durable; 20,548 reach a human
coverage 55%, apparent 65%, net automation 29.3% of total demand
$52,298 converted less $13,700 operating cost = $38,598 a month; the escalation premium takes $21,575
$364,000 a month on the slide; $38,598 modelled
39.5% apparent resolution; repeats tolerable to 50.1%
$332,435 over the 6-month ramp; payback month 3
Run a bounded pilot: positive on an internal estimate
What this tool cannot tell you
- Every bot rate is usually a claim until your own pilot measures it. The evidence source you pick decides how far the grade can go, and nothing here reaches Finance-grade because no document is inspected.
- The escalation premium has no single published figure. The page shows net savings with it at zero and at double, so you can see how much of the answer rests on it.
- Eligibility is a property of your demand; the vendor does not set it. The AI Readiness Diagnostic looks at what limits it.
- Savings are valued at marginal cost. Loaded cost appears only in the vendor claim, because a deflected contact does not refund fixed cost.