01
Task performance
- Output quality
- How often the AI output can be used without a major rewrite.
- Oversight need
- Human review time needed for each hour of AI-run work.
- Latency
- Whether the system responds fast enough for the task.
Methodology
How Wagecore estimates AI cost for a role.
See the task inputs, cost assumptions, task classes, confidence on core estimates, and the limits of the result.
What you get
What a Wagecard shows.
A Wagecard compares modeled AI run cost with market pay, shows what AI can do in each task and where it needs human review, and records the assumptions and methodology version used.
Financial analyst · US Tier 2 · mid-career
Augmentation territoryEconomic substitution exposure
20/ 100
0high confidence100
Operational AI cost
$4,681/mo
Includes human review, retries, and error cost.
Market pay (p50)
$86k/yr
Median market pay for the selected region and experience level.
Four task classes
Replaceable0%
AI-augmented67%
Human-led, AI-assisted17%
Human-critical17%
Illustrative example. Values change with the role, tasks, hours, location, and experience.
Exposure estimate
A 0–100 summary of how much of the selected task mix may be economical to run with AI. Read it with the task breakdown and qualitative confidence, not on its own.
Operational AI cost
The modeled cost of AI for the selected work, including review time, retries, integration, and expected error cost.
The task mix
Each selected task is placed in one of four classes so you can see where AI may replace, assist, or add little value.
- Confidence on core estimates
- Modeled AI run cost
- Market pay
- What AI can do in each task
Inputs used
What affects the estimate.
The model combines output quality, review time, reliability, error cost, integration work, AI operating cost, and the value of human context.
02
Reliability and error cost
- Reliability
- How often the task needs a retry or human escalation.
- Hallucination risk
- How often the output can be plausible but wrong.
- Error cost
- The financial, regulatory, or reputational cost of a mistake.
03
Operating cost
- Integration overhead
- The work needed to prepare context, monitor output, and maintain the workflow.
- Unit economics fit
- Whether the modeled AI cost is lower than the market cost of the same work.
04
Human context
- Value of human context
- Whether trust, ambiguity, accountability, persuasion, or context make a person more reliable or cost-effective.
Details and limits
Check each part of the method.
Open the sections you need to review task classes, reliability, financial scenarios, transition costs, assumptions, sources, and versioning.
How each task is classified
High substitution
Replaceable
AI can complete the task with little human review at a lower modeled cost.
Augmentation territory
AI-augmented
AI can do much of the task, but a person still needs to check the result, handle exceptions, and remain accountable for the outcome.
Hybrid optimal
Human-led, AI-assisted
A person leads the task while AI helps with drafting, search, summarization, or other supporting work.
Human-critical work
Human-critical
A person remains central because trust, regulation, accountability, or relationships make AI less useful or more costly.
How reliability changes the result
Task capability asks whether AI can produce an acceptable result. Reliability asks how often it does so without a retry, correction, or escalation. Wagecore uses both because review time and failures change the cost.
In the current matrix, 38 task-model cells score capability ≥ 75, while 5 score reliability ≥ 80. That gap is why the cost model includes human review and retries.
A Wagecard shows the hours-weighted capability and reliability estimates for the selected tasks, plus the gap between them. It also shows how much selected work falls into the capability ≥ 75 and reliability < 80 group.
Capability gap
- 38
- task-model cells with capability ≥ 75
- 5
- task-model cells with reliability ≥ 80
Klarna later said it was adding more human support after quality concerns in customer service.
Financial scenarios: NPV, IRR, and payback
Investment View calculates three common finance measures from the salary, transition cost, discount rate, and other assumptions shown in the scenario. These are projections, not guaranteed returns.
NPV
5-year
Discounted scenario savings minus the initial transition cost. A positive result means the selected assumptions produce value at the chosen discount rate.
IRR
Internal rate of return
The annual return implied by the scenario. Compare it with the organization's own hurdle rate before making a decision.
Payback
Period
The time it takes for modeled cumulative savings to recover the transition cost under the selected assumptions.
Investment View shows a five-year scenario and the assumptions used. The discount rate defaults to 10% and can be changed on Pro. The scenario does not include option value, strategic redeployment value, or terminal value after Year 5.
Transition cost & sequencing
The financial scenario also estimates how hard the work is to change. This affects transition cost, but it does not change the task capability or classification estimates.
Switching cost
How the work setup changes transition cost
The scenario uses different transition-cost assumptions for outsourced and internal work. Review and replace those assumptions with your organization's actual costs before using the projection.
Capability trajectory
Current case and optional growth scenario
The base scenario holds the current capability estimate flat. An optional scenario increases the AI-handled share over time while leaving human-critical work unchanged. Both results depend on the assumptions shown.
Work setup belongs in the financial scenario because it describes the organization, not the task. The task estimates remain unchanged.
Why reviewer pay matters
In the representative model shown here, human review accounts for most of the monthly AI run cost. Tokens, orchestration, retries, and integration make up the rest. The loaded hourly cost of the reviewer therefore has a large effect on the result.
Assumed oversight wage
$55/hourloaded
The default uses a US knowledge-work median base rate multiplied by roughly 1.3 for benefits and overhead. Replace it with your own loaded reviewer cost when available because it directly affects the modeled AI run cost.
Qualitative confidence on core estimates reflects uncertainty in this assumption and in the capability matrix. Treat the cost as a modeled estimate and check the reviewer-wage input before using it.
What outside research says about deployment cost
These sources show why an AI cost estimate needs more than model capability and token price. Review time, failures, integration, and the cost of mistakes can change whether a task is worth automating.
Nvidia VP of Applied Deep Learning
April 2026“For my team, the cost of compute is far beyond the costs of the employees.”
Fortune ↗MIT CSAIL
2024 study“AI automation economically viable in only 23% of vision-primary roles at current cost structures.”
Study ↗BCG
2025“Only 5% of companies are capturing AI value at scale; ~60% report no material value despite investment.”
BCG ↗Klarna + Uber
2025–2026“Klarna later added more human support after quality concerns. Uber said its 2026 AI coding budget was used in four months.”
Wagecore uses this evidence to model current task economics. It does not treat a capability demo as proof that changing the work will save money.
How the capability matrix is reviewed and versioned
One evaluator applies the same documented rubric across the matrix. Each published change receives a methodology version so a Wagecard can record which inputs were used.
Documented inputs and formula
This page documents the rubric, cost model, and task classes. Wagecards record the methodology version and show qualitative confidence on core estimates.
Evidence-gated updates
The matrix changes when enough new evidence passes the quality checks. Each published change receives a version; there is no fixed calendar promise.
Confidence on core estimates
Qualitative confidence applies to the core modeled estimates. It is not a guarantee and does not apply to every displayed figure.
Consistent scoring rubric
The same rubric is applied across task-model cells. Version notes record material changes.
Build a Wagecard from your own work.
Choose your tasks and hours. See modeled AI run cost, market pay, task classifications, and confidence on the core estimates.