AINTERNALS · AI TCO CALCULATOR

Every input, explained.

Understand what to enter, why it matters, and where to find reliable assumptions. A practical guide to all 33 fields, with examples and common mistakes.

Calculator 0.4.0 · Cost model 1.0 · Examples are illustrative

A consistent scenario starts here.

Define one workflow task and use that definition throughout. Enter all money in one currency. Replace illustrative defaults with measured usage, verified prices and realistic budgets. Start with General & success, then work through the remaining sections.

To use Price Watch, choose a model and billing options in the calculator, confirm a single-request task and apply the rates. Linked pricing requires USD for every cost. Matching uses your token counts locally; it does not send assumptions to the pricing service. For multi-call tasks or mixed request sizes, use a verified blended manual rate.

The human-only comparison is optional. Saving or printing results does not require an account. Your scenario stays in browser memory until you export it; refreshing resets it.

General & success

Define the workload and its outcomes.

6 fields

Currency

Currency label

What it means: The currency used to label every monetary input and result.
Why it is included: Keeps all costs in one consistent unit so the estimate is readable.
What to enter: Choose the currency in which you have already expressed all costs. Linked Price Watch rates require USD. Convert every non-USD assumption separately before linking; the calculator never converts existing amounts.
Where to find it: Your budgeting currency and any currency conversions you have performed separately.
Illustrative example: Choose INR when all prices, salaries and expenses have been entered in INR.
What it influences: Changes monetary labels only; it does not change the numerical estimate or perform an exchange-rate conversion.
Common mistake: Do not change the label and assume USD prices have been converted into another currency.
When zero or off is appropriate: Choose a supported currency; this field cannot be disabled.

Annual volume growth

% per year

What it means: The expected yearly increase in attempted tasks after Year 1.
Why it is included: A growing workload can change future operating costs and the number of successful outcomes.
What to enter: Enter a yearly percentage between 0 and 100. Use one assumption for both later years.
Where to find it: Workload forecasts, adoption plans or historical demand trends.
Illustrative example: Enter 20 for an illustrative expectation of 20% annual workload growth.
What it influences: Affects later-year task volumes, usage-driven expenses and costs marked as scaling with volume.
Common mistake: This is growth in attempts, not an improvement in task quality or success rates.
When zero or off is appropriate: Enter 0 to keep annual attempted volume unchanged.

Annual operating cost inflation

% per year

What it means: The expected yearly increase in people, rework and ongoing operating expenses.
Why it is included: Captures cost pressure beyond the amount of work performed.
What to enter: Enter a yearly percentage between 0 and 100 for the costs covered by this assumption.
Where to find it: Internal salary plans, vendor budgets and finance-approved operating assumptions.
Illustrative example: Enter 5 for an illustrative yearly operating-cost increase of 5%.
What it influences: Affects later-year labor, rework, monthly operating expenses and enabled human-only costs. It does not change token rates or other per-attempt tool charges.
Common mistake: Do not use this field for token-price changes; those have a separate input.
When zero or off is appropriate: Enter 0 to hold the covered unit costs steady before any workload scaling.

Monthly attempted AI tasks

Attempts per month

What it means: The average number of AI workflow tasks started each month in Year 1.
Why it is included: Establishes the workload served by the proposed solution.
What to enter: Enter a nonnegative whole number of workflow attempts per month, including attempts that fail.
Where to find it: Workflow logs, business activity records or a measured pilot.
Illustrative example: Enter 10000 for an illustrative workload of ten thousand attempted tasks per month.
What it influences: Affects annual task volumes, usage costs, review, rework and an enabled human-only comparison.
Common mistake: Define a task consistently. An attempt is not necessarily one API call; one workflow may make several calls.
When zero or off is appropriate: Enter 0 for a no-workload scenario. Fixed and upfront expenses can still remain.

First-pass completion rate

% of all attempts

What it means: The share of attempted tasks completed successfully before remediation.
Why it is included: Distinguishes reliable initial outcomes from tasks needing extra work.
What to enter: Enter a measured percentage between 0 and 100 using a consistent quality standard.
Where to find it: Evaluation results, quality checks and production task outcomes.
Illustrative example: Enter 90 if 90 out of every 100 attempted tasks meet the required standard initially.
What it influences: Affects first-pass outcomes, initial failures, rework expense and successful-task costs.
Common mistake: Do not count a task as first-pass successful if it required remediation to meet the standard.
When zero or off is appropriate: Enter 0 if no tasks succeed initially; 100 means no initial failures.

Failed-task recovery rate

% of initial failures

What it means: The share of initially failed tasks that become successful after remediation.
Why it is included: Accounts for useful outcomes recovered from initial failures.
What to enter: Enter a percentage between 0 and 100 of initially failed tasks, not of all attempts.
Where to find it: Remediation logs and post-rework quality checks.
Illustrative example: Enter 80 if 80 out of every 100 initially failed tasks are eventually recovered.
What it influences: Changes recovered successes, unresolved failures and successful-task costs. It does not automatically reduce the rework budget.
Common mistake: Rework is charged on all initial failures, even those that remain unresolved. Recovery is not a cost-saving switch.
When zero or off is appropriate: Enter 0 if no initial failures are recovered; 100 means all are recovered.

Model & pricing

Usage, rates and the pricing reference.

7 fields

Input tokens per attempt

Tokens per attempt

What it means: The average billed input tokens used by one attempted workflow task.
Why it is included: Captures the input-side model usage associated with your workload.
What to enter: Enter a nonnegative whole-number average covering the calls within a task.
Where to find it: Provider usage logs and traces from representative workflows.
Illustrative example: Enter 2000 for an illustrative average of two thousand billed input tokens per attempt.
What it influences: Affects model input expense and the overall cost estimate.
Common mistake: Do not measure just the user prompt if system messages, context or other calls are billed as part of the task. Avoid duplicating remediation usage included in rework. When linking Price Watch, this value must represent one billed request in a consistent tier. For multi-call workflows or mixed request sizes, use verified blended manual rates.
When zero or off is appropriate: Enter 0 if this scenario has no billed input-token usage.

Output tokens per attempt

Tokens per attempt

What it means: The average billed output tokens generated for one attempted workflow task.
Why it is included: Captures the output-side model usage, which may have a different rate from input usage.
What to enter: Enter a nonnegative whole-number average across representative tasks.
Where to find it: Provider usage reports and workflow traces.
Illustrative example: Enter 500 for an illustrative average of five hundred output tokens per attempt.
What it influences: Affects model output expense and the overall cost estimate.
Common mistake: Use billed usage for the chosen model and tier, not only the visible text length. Keep remediation usage consistent with rework scope. Linked Price Watch matching checks this count against the selected model output and context limits.
When zero or off is appropriate: Enter 0 if no output tokens are billed.

Input price per 1M tokens

Currency per 1M tokens

What it means: The price of one million billed input tokens for the selected pricing assumptions.
Why it is included: Turns input usage into an estimated model expense.
What to enter: Enter a nonnegative rate in the same currency as every other monetary input. Alternatively, apply a model from Price Watch to populate and link the matching rate.
Where to find it: Your provider's official pricing, contract or invoice, including the applicable tier.
Illustrative example: Enter 2 for an illustrative rate of 2 currency units per million input tokens. This is not a verified quote.
What it influences: Affects input-model cost. The annual token-price assumption can change this rate in later years. For Moonshot models with required cache writes, the linked effective input rate includes those charges and is labeled accordingly.
Common mistake: Check billing tier, context bands and discounts. A single average rate may not represent every provider pricing rule. Linked rates are read-only until you switch to manual pricing.
When zero or off is appropriate: Enter 0 only if input usage is free for the scenario; record the reason in the pricing reference.

Output price per 1M tokens

Currency per 1M tokens

What it means: The price of one million billed output tokens for the selected pricing assumptions.
Why it is included: Keeps output pricing distinct from input pricing.
What to enter: Enter a nonnegative rate in your selected scenario currency. Applying a Price Watch model fills the matching published output rate automatically.
Where to find it: Official provider pricing, a contract or an invoice for the relevant model and tier.
Illustrative example: Enter 10 for an illustrative rate of 10 currency units per million output tokens. This is not a verified quote.
What it influences: Affects output-model cost and later-year token expense.
Common mistake: Do not use the input-token price here unless the provider actually charges the same rate. Linked rates are read-only until you switch to manual pricing.
When zero or off is appropriate: Enter 0 only when output tokens carry no charge in the scenario.

Other API/tools cost per attempt

Currency per attempt

What it means: Other usage-based API or tool charges associated with one attempted task.
Why it is included: Includes workflow services beyond the model's input and output token bill.
What to enter: Enter an average nonnegative amount per attempt in your scenario currency.
Where to find it: Invoices for search, retrieval, document processing, external tools or other metered services.
Illustrative example: Enter 0.01 for an illustrative additional tool expense of one hundredth of a currency unit per attempt.
What it influences: Affects recurring usage costs as attempted volume changes. Token-price changes and operating inflation do not alter this entered unit rate.
Common mistake: Keep monthly subscriptions in software and licenses. Do not repeat token charges or remediation expenses already included elsewhere.
When zero or off is appropriate: Enter 0 if there are no additional per-attempt tool charges.

Annual token price change

% per year

What it means: The yearly change in input and output token prices after Year 1.
Why it is included: Lets you explore token-price uncertainty separately from general operating inflation.
What to enter: Enter a percentage greater than negative 100. Negative values represent price decreases; positive values represent increases.
Where to find it: A documented scenario assumption, contract schedule or pricing sensitivity analysis.
Illustrative example: Enter -20 for an illustrative annual token-price decline of 20%. This is a scenario, not a forecast.
What it influences: Affects later-year input and output token expense only. It does not change other tools, workload or staffing costs.
Common mistake: Do not apply provider price expectations to every workflow expense, or treat an illustrative decline as guaranteed. Even with linked Price Watch prices, this remains your future scenario assumption; a current snapshot is not a three-year forecast.
When zero or off is appropriate: Enter 0 to hold token prices steady across all three years.

Pricing reference

Reference text

What it means: A note identifying the provider, model, pricing basis and date checked.
Why it is included: Makes the assumptions understandable when you revisit or share a scenario.
What to enter: Enter a short reference with the official pricing link, model, billing tier and verification date, using at most 2000 characters. Applying Price Watch records the model, billing scope, rates, fetch time and official source automatically.
Where to find it: Official pricing pages, your contract and the records used to verify the entered rates.
Illustrative example: Example: provider and model name; standard tier; official pricing URL; checked on your chosen date.
What it influences: Adds context to saved scenarios. It does not verify prices, fetch live rates or change numerical results. Exported scenarios retain these rates and the reference; imports do not silently replace them with newer prices.
Common mistake: Do not label prices verified unless you actually checked the relevant source. Avoid placing private credentials or account details in this note.
When zero or off is appropriate: Leave a clear illustrative or unverified note when rates have not been checked; the field can also be left empty.

Operating & scaling

People and systems behind every task.

11 fields

Cloud infrastructure per month

Currency per month

What it means: The monthly infrastructure expense supporting the workflow at Year 1 workload.
Why it is included: Accounts for hosting, compute, storage and supporting infrastructure beyond the API bill.
What to enter: Enter an average nonnegative monthly amount attributable to this workflow.
Where to find it: Cloud invoices, infrastructure budgets or an agreed share of shared services.
Illustrative example: Enter 500 for an illustrative monthly infrastructure budget of 500 currency units.
What it influences: Affects recurring operating costs. Inflation applies in later years; the adjacent scaling choice determines whether volume growth also applies.
Common mistake: Do not include a provider's token bill twice. Allocate shared infrastructure consistently rather than assigning the whole company bill.
When zero or off is appropriate: Enter 0 if no separate infrastructure expense applies.

Cloud scales with volume?

Yes or No

What it means: Whether infrastructure expense rises with task volume in addition to inflation.
Why it is included: Distinguishes a workload-dependent infrastructure budget from a stable one.
What to enter: Choose Yes if the infrastructure budget is expected to grow with demand; otherwise choose No.
Where to find it: Capacity plans, cloud usage trends and engineering estimates.
Illustrative example: Choose Yes for a service where hosting resources grow with traffic; No for adequate fixed capacity over the modeled range.
What it influences: Changes later-year infrastructure expense. It does not affect Year 1 infrastructure cost.
Common mistake: Scaling here assumes a smooth relationship with demand; stepped capacity purchases or volume discounts are not modeled separately.
When zero or off is appropriate: No removes workload scaling for this expense, but operating inflation still applies.

Monitoring & security per month

Currency per month

What it means: The ongoing monthly cost of observing and protecting the workflow.
Why it is included: Includes the operational visibility and security needed to run the service.
What to enter: Enter an average nonnegative monthly monitoring and security amount.
Where to find it: Observability invoices, security-tool costs and operating budgets.
Illustrative example: Enter 200 for an illustrative recurring budget of 200 currency units per month.
What it influences: Adds ongoing monitoring and security expense, with later-year inflation and optional demand scaling.
Common mistake: Keep initial security assessments and setup in one-time security and compliance; avoid duplicating general software licenses.
When zero or off is appropriate: Enter 0 if no separate recurring monitoring or security expense applies.

Monitoring scales with volume?

Yes or No

What it means: Whether recurring monitoring and security expense grows with attempted volume.
Why it is included: Allows usage-priced logging or security services to respond to workload growth.
What to enter: Choose Yes for demand-sensitive costs, or No for a stable subscription or budget.
Where to find it: Vendor billing terms and expected log, trace or event volumes.
Illustrative example: Choose Yes for monitoring billed by growing event volume; No for a fixed plan with adequate capacity.
What it influences: Changes later-year monitoring expense in addition to operating inflation.
Common mistake: Do not select Yes simply because the service is important; base the choice on how its cost changes with demand.
When zero or off is appropriate: No keeps workload growth out of this expense; inflation still applies.

Software & licenses per month

Currency per month

What it means: The monthly cost of software subscriptions and licenses used by the workflow.
Why it is included: Captures recurring software expenses outside separately entered API and infrastructure costs.
What to enter: Enter an average nonnegative monthly amount attributable to the workflow.
Where to find it: License inventories, subscription invoices and procurement records.
Illustrative example: Enter 300 for an illustrative software budget of 300 currency units per month.
What it influences: Adds recurring license expense with inflation and optional workload scaling.
Common mistake: Exclude subscriptions already included in monitoring, infrastructure or another expense category.
When zero or off is appropriate: Enter 0 if there are no separate software or license expenses.

Licenses scale with volume?

Yes or No

What it means: Whether the software and license budget increases as workload grows.
Why it is included: Reflects additional seats, usage bands or licenses needed with increased demand.
What to enter: Choose Yes when the total budget is expected to grow with workload; otherwise choose No.
Where to find it: Contract terms, seat plans and procurement forecasts.
Illustrative example: Choose No for an unchanged fixed subscription; Yes if workload growth requires a proportionately larger license budget.
What it influences: Changes later-year software expense; Year 1 remains the entered base amount.
Common mistake: License bands and seat counts can change in steps. This simple choice does not model individual contract tiers.
When zero or off is appropriate: No disables volume scaling only; operating inflation still applies.

Human review minutes per attempt

Minutes per attempt

What it means: The average human review time allocated to every attempted workflow task.
Why it is included: Makes the people cost of checking AI work visible.
What to enter: Enter nonnegative average minutes across all attempts, including tasks receiving no review.
Where to find it: Time studies, sampled review queues and pilot observations.
Illustrative example: If half the tasks need two minutes of routine review, enter 1 minute averaged across all attempts.
What it influences: Affects human-review operating expense and total ownership cost.
Common mistake: Keep incremental remediation time in rework if already budgeted there. Do not enter the time for reviewed tasks alone when only some tasks are reviewed.
When zero or off is appropriate: Enter 0 if the workflow has no routine human review.

Fully loaded reviewer hourly cost

Currency per hour

What it means: The fully loaded hourly cost of the people reviewing AI work.
Why it is included: Values review time using more than the base salary alone.
What to enter: Enter a nonnegative hourly amount in your scenario currency, including applicable employer overhead.
Where to find it: Finance-approved labor rates, staffing budgets or contractor rates.
Illustrative example: Enter 30 for an illustrative fully loaded review rate of 30 currency units per hour.
What it influences: Affects review expense; the operating inflation assumption also affects later-year review costs.
Common mistake: Do not enter annual salary or add the same employer overhead again in another cost category.
When zero or off is appropriate: Enter 0 only when reviewer time is deliberately assigned no cost in this scenario.

Rework cost per initially failed task

Currency per initial failure

What it means: The incremental cost of handling one initially failed task, whether recovered or not.
Why it is included: Captures remediation effort that the normal task budget does not cover.
What to enter: Enter an average nonnegative amount per initial failure, covering the agreed scope of extra processing and human effort.
Where to find it: Remediation logs, retry usage, support records and observed task-level costs.
Illustrative example: Enter 1.5 for an illustrative average remediation expense of 1.5 currency units per initially failed task.
What it influences: Affects recurring rework expense. Initial failure levels matter; recovery alone does not remove the charge.
Common mistake: Do not charge only unresolved failures. Avoid counting extra tokens or review here if those same costs are already included in routine usage or review.
When zero or off is appropriate: Enter 0 only if initial failures require no incremental remediation expense in the scenario.

Support & maintenance per month

Currency per month

What it means: The recurring monthly cost of keeping the workflow running and maintained.
Why it is included: Includes operational support, maintenance and incident handling outside other categories.
What to enter: Enter an average nonnegative monthly amount attributable to the workflow.
Where to find it: Support contracts, maintenance budgets and allocated engineering time.
Illustrative example: Enter 1000 for an illustrative support budget of one thousand currency units per month.
What it influences: Adds ongoing support expense, with inflation and optional workload scaling.
Common mistake: Separate routine maintenance from initial implementation. Do not include routine review or per-failure remediation again.
When zero or off is appropriate: Enter 0 if no separate ongoing support expense is expected.

Support scales with volume?

Yes or No

What it means: Whether the support and maintenance budget grows with attempted task volume.
Why it is included: Reflects a support burden that may increase as adoption expands.
What to enter: Choose Yes for a demand-dependent budget, or No for a stable recurring amount.
Where to find it: Support workload trends, service agreements and staffing plans.
Illustrative example: Choose Yes if a growing workload needs proportionately more support; No if the existing service agreement covers the modeled demand.
What it influences: Changes later-year support expense in addition to operating inflation.
Common mistake: This is a simple workload-scaling assumption, not a detailed staffing or incident forecast.
When zero or off is appropriate: No removes workload scaling for support; inflation still applies.

Upfront & financial

Setup costs and financial assumptions.

6 fields

One-time implementation

One-time currency amount

What it means: The one-time cost of designing, integrating and launching the workflow.
Why it is included: Recognizes the initial investment before routine operations begin.
What to enter: Enter a nonnegative project-start amount in the selected currency.
Where to find it: Implementation plans, vendor proposals and allocated delivery effort.
Illustrative example: Enter 15000 for an illustrative initial implementation budget.
What it influences: Adds to initial investment and Year 1 total cost, not annual operating expense.
Common mistake: Do not repeat recurring maintenance, data preparation, training or compliance costs already entered in their own categories.
When zero or off is appropriate: Enter 0 if implementation requires no additional investment in the scenario.

One-time data preparation

One-time currency amount

What it means: The one-time cost of preparing the data needed to launch the workflow.
Why it is included: Captures initial collection, cleaning, labeling, migration or organization of data.
What to enter: Enter a nonnegative setup amount attributable to this workflow.
Where to find it: Data preparation plans, labeling quotes and project budgets.
Illustrative example: Enter 5000 for an illustrative initial data-preparation budget.
What it influences: Adds to upfront investment and Year 1 total cost.
Common mistake: Do not repeat work already included in implementation. Ongoing data expenses belong in an appropriate recurring category.
When zero or off is appropriate: Enter 0 if the necessary data is ready and there is no additional setup cost.

One-time training & change

One-time currency amount

What it means: The one-time investment in user training and organizational adoption.
Why it is included: Recognizes that introducing the workflow can require more than technical deployment.
What to enter: Enter a nonnegative project-start budget for training and change activities.
Where to find it: Training plans, enablement budgets and change-management estimates.
Illustrative example: Enter 3000 for an illustrative launch training and adoption budget.
What it influences: Adds to upfront investment and Year 1 total cost.
Common mistake: This means people training and change management. Do not interpret it automatically as model-training compute or duplicate ongoing support.
When zero or off is appropriate: Enter 0 if no additional training or adoption investment is needed.

One-time security & compliance

One-time currency amount

What it means: The one-time expense of security and compliance preparation for launch.
Why it is included: Accounts for initial assessments, controls and approval work.
What to enter: Enter a nonnegative upfront amount attributable to the proposed workflow.
Where to find it: Security reviews, audit estimates, compliance plans and vendor proposals.
Illustrative example: Enter 2000 for an illustrative initial security and compliance budget.
What it influences: Adds to upfront investment and Year 1 total cost.
Common mistake: Keep ongoing security subscriptions in monitoring and security. This field does not automatically cover all future regulatory impacts.
When zero or off is appropriate: Enter 0 if no additional initial security or compliance work is required.

Operating contingency rate

% of recurring costs

What it means: The reserve added to recurring operating expenses for uncertainty.
Why it is included: Provides an explicit allowance for operating costs that differ from the base assumptions.
What to enter: Enter a percentage between 0 and 100 based on the uncertainty you want to explore.
Where to find it: Finance policies, project risk assessments and operating experience.
Illustrative example: Enter 10 for an illustrative operating contingency of 10%.
What it influences: Raises the operating-cost allowance. It does not add a reserve to one-time setup expenses.
Common mistake: Do not use this as a guarantee that every omitted cost is covered, or repeat a reserve already built into each expense.
When zero or off is appropriate: Enter 0 to show the operating estimate without an additional contingency allowance.

Annual discount rate

% per year

What it means: The annual rate used to express future operating expenses in today's terms.
Why it is included: Helps compare costs occurring at different times on a consistent present-value basis.
What to enter: Enter a percentage between 0 and 100 appropriate to your scenario.
Where to find it: Your organization's finance team or approved investment-appraisal assumptions.
Illustrative example: Enter 10 for an illustrative annual discount rate of 10%; this is not a recommended rate.
What it influences: Changes present-value results, not undiscounted totals, usage or initial setup expenses.
Common mistake: Do not confuse this with operating inflation or a provider discount. Upfront spending is paid at the start and is not discounted.
When zero or off is appropriate: Enter 0 when you want no reduction of future expenses for timing.

Human-only comparison

Optional costs for equivalent manual work.

3 fields

Human-only minutes per task

Minutes per equivalent task

What it means: The time needed for a person to complete an equivalent task without the AI workflow.
Why it is included: Enables an optional comparison against a genuinely comparable human-only approach.
What to enter: Enter nonnegative average minutes per equivalent task. Leave at 0 to keep the comparison off.
Where to find it: Measured manual workflows and comparable quality checks.
Illustrative example: Enter 10 for an illustrative manual task taking ten minutes, after checking equivalent scope and quality.
What it influences: Enables human-only costs and cost-advantage results. It never changes the AI workflow's own cost estimate.
Common mistake: A faster manual task is not comparable if the quality, scope or outcome differs. The comparison uses the same attempted workload.
When zero or off is appropriate: Enter 0 to disable the complete human-only comparison, even if other manual cost fields contain values.

Human-only hourly cost

Currency per hour

What it means: The fully loaded hourly labor cost for the human-only alternative.
Why it is included: Values the manual work independently of the AI-reviewer role.
What to enter: Enter a nonnegative hourly amount in the same currency as the AI scenario.
Where to find it: Finance-approved staffing rates or relevant contractor agreements.
Illustrative example: Enter 30 for an illustrative manual labor rate of 30 currency units per hour.
What it influences: Affects human-only costs and cost advantages only when the comparison is enabled. It does not change AI costs.
Common mistake: Do not assume the manual worker and AI reviewer have the same rate unless that matches the workflow.
When zero or off is appropriate: Enter 0 only if manual labor is deliberately assigned no cost. This field has no effect while manual minutes remain 0.

Human-only overhead per task

Currency per manual task

What it means: Other direct expenses incurred for each equivalent human-only task.
Why it is included: Includes manual-work costs beyond the fully loaded labor rate.
What to enter: Enter a nonnegative amount per task, excluding costs already in the hourly rate.
Where to find it: Manual-process budgets, consumables records and attributable service charges.
Illustrative example: Enter 0.5 for an illustrative additional manual-task expense of half a currency unit.
What it influences: Affects enabled human-only costs and comparison results, not the AI estimate.
Common mistake: Do not add employer overhead again if the hourly rate already includes it. Only include costs attributable to the comparable manual task.
When zero or off is appropriate: Enter 0 when there are no additional direct manual-task expenses.

Reading your results

Setup, ongoing costs and totals

Upfront investment is the one-time setup budget, included in Year 1 only. Operating costs cover running the workflow. Annual totals include the costs for that year; cumulative totals include all years up to that point.

Present value

Present value expresses future expenses in today's terms. It helps compare costs occurring at different times. It is separate from the undiscounted amount you expect to spend, and a lower present-value figure does not mean the bill has been reduced.

Successful-task costs

These describe the cost of useful outcomes, including recovered tasks. Unresolved failures are excluded from successful-task counts. A zero unit cost alongside zero attempts or successes is not a meaningful cost estimate.

Positive and negative cost advantage

A positive advantage means modeled AI costs are lower than the human-only alternative. A negative advantage means modeled AI costs are higher. Compare equivalent scope, quality and outcomes. Neither result establishes ROI or guarantees cash savings.

Annual versus three-year comparison

Annual advantages describe a single year. Three-year advantages describe the full period. The three-year present-value comparison includes initial setup and discounts future operating expenses for both alternatives consistently.

Before you decide

What is excluded

Taxes, working capital, opportunity cost, carbon, regulatory impacts and other externalities are not automatically included. Revenue and ROI are outside this calculator's scope.

Check before deciding

Validate actual token usage, review times, completion and recovery rates, remediation costs and scaling assumptions. If comparing with human-only work, confirm equivalent outcomes.

Privacy

Your assumptions are processed in your browser and kept in memory. The calculator loads public model-pricing data from this site, whose pricing service retrieves provider documents. Request sizes, costs, scenarios and results are not sent to that service or to providers. No calculator analytics or session replay is enabled.

Save the assumptions behind the result

Use Export scenario to keep a JSON file you can import later. Print the calculator to share the assumptions and results; print this guide separately for the full field reference. Files can contain sensitive business assumptions, so store and share them carefully.

Something doesn’t look right?

Report a calculation discrepancy, unclear guidance, or any other issue to sriram@sriramadvisory.com.

Include the affected field or result, what you expected, and what you observed. Avoid sharing sensitive financial assumptions or private scenario files.