Monthly manual cost minus monthly agent-assisted cost
Manual cost is current workflow time multiplied by the loaded cost of people doing it. Agent-assisted cost includes assessment, setup, software, integration, management, and review time.
AI agent ROI calculator
How to Calculate AI Agent ROI Before You Build One
A practical ROI worksheet for New Zealand businesses considering AI agents
To calculate AI agent ROI, estimate manual workflow cost, subtract assessment, setup, tool, management, and review costs, then adjust for risk, quality, and human approval burden.
An AI agent is worth assessing when the workflow is frequent, measurable, commercially relevant, and safe to pilot with approval. Decide whether one workflow deserves a pilot, more preparation, simpler automation, or no build.
If the numbers are unclear, start with the $1,000 AI Agent Assessment. If budget is the first question, read AI Agent Cost NZ alongside this worksheet.
AI agent ROI calculator: the simple formula
Use this plain-English formula before you build:
Add the decision filter
A positive estimate is not enough. The workflow must be safe, useful, reviewable, and connected to a real business result. Time saved with customer, privacy, or data-quality risk may still be the wrong first build.
Step 1: Name the workflow, not the tool
ROI starts with a job, not a software category.
Good candidates: sales prep, CRM hygiene, reporting, content operations, website checks
Good candidates are recurring tasks with inputs and reviewable outputs: lead follow-up preparation, CRM cleanup, weekly management reporting, content briefs, website checks, proposal research, and knowledge summaries. For more ideas, use AI Workflow Automation Examples for NZ Businesses.
Weak candidates: rare, vague, high-risk, or poorly documented work
Weak candidates are hard to measure, happen rarely, depend on undocumented judgement, or require risky action without review. If the workflow cannot be described in one sentence, it is not ready for ROI calculation.
Step 2: Estimate current manual cost
Manual cost shows whether the workflow matters.
Time per task
Estimate how long one cycle takes now. Include preparation, searching, checking, handoff, correction, and follow-up. Do not count only typing time.
Frequency per month
Multiply task time by monthly frequency. A 20-minute task done 80 times may matter more than a 4-hour task done once.
Who currently does the work
Use a realistic loaded cost for the person doing the work. Owner, manager, salesperson, and admin time carry different opportunity costs. If unknown, use conservative ranges.
Step 3: Estimate agent-assisted operating cost
Agent-assisted cost includes more than the monthly software subscription.
Setup or assessment cost
Include deciding what to build, mapping the workflow, preparing sources, setting permissions, and designing approval gates. AI Agent Agency uses a $1,000 assessment before installation so the build decision is scoped.
Review time
Human review is part of the operating model. If the agent prepares a follow-up, brief, report, or page recommendation, someone approves it before it affects customers, records, money, or public content.
Tool, integration, and management cost
Include model access, automation tools, integrations, maintenance, testing, logs, error review, and ongoing management. Management is what keeps an agent useful after the first demo.
Step 4: Add quality and risk factors
Some value does not show up as hours saved, but it still affects the decision.
Error reduction
An agent may reduce missed steps by preparing a checklist, source note, draft, or exception list. This is valuable only if a human can verify it against trusted sources.
Missed follow-up reduction
In sales and client-service workflows, ROI may come from fewer missed handoffs or stale opportunities. Do not assume revenue uplift. Track whether the agent reliably prepares the next action for review. For sales examples, read AI Lead Follow-Up Automation NZ.
Faster reporting or decision preparation
For management reporting, the value may be a better weekly brief, not an automated decision. The agent can gather approved figures, flag missing data, and prepare notes before a manager decides. See AI Management Reporting Automation for NZ Businesses.
Privacy, permissions, and approval risk
Risk can erase apparent ROI. Decide what the agent may read, draft, update, send, publish, delete, or never touch. Use the AI Risk Register for NZ Businesses Using AI Agents when sensitive data or customer trust is involved.
Bounded access
ROI worksheet for NZ businesses
Use this worksheet before assessment or build.
1. Workflow name
Write the job in one sentence: “Prepare a weekly sales follow-up queue for human review” is stronger than “use AI for sales.”
2. Current monthly manual cost
Estimate task time, monthly frequency, reviewer role, correction time, and handoff time.
3. Expected agent-assisted cost
Estimate assessment, setup, subscription, integration, management, and review time. Stay conservative.
4. Quality and risk adjustment
Name the main quality upside and risk. If the risk cannot be controlled with permissions and approval gates, the workflow should wait.
5. Build, prepare, wait, or reject
Choose a decision, not a fantasy forecast. Build means a narrow pilot is ready. Prepare means data, permissions, or process need cleanup. Wait means the timing is wrong. Reject means the risk or low frequency does not justify the work.
Two practical examples
These examples show how to think, not guaranteed results.
Example: lead follow-up preparation
Manual cost may include reading enquiries, checking CRM history, researching fit, drafting a response, setting a next action, and reminding the owner. An agent-assisted version might prepare context, draft a response, suggest a next step, and stop before sending. Include review time and prospect-trust risk.
Example: weekly management reporting
Manual cost may include exporting data, checking spreadsheets, chasing missing figures, summarising performance, and preparing decisions. An agent-assisted version might gather approved sources, flag gaps, prepare exceptions, and draft a brief. Include source cleanup and human decision review.
What ROI cannot prove before a pilot
An ROI worksheet helps decide whether to assess or pilot a workflow. It does not prove the agent will work.
It cannot prove data quality
The estimate may assume the CRM, files, reports, or SOPs are accurate. A pilot has to test that.
It cannot prove staff adoption
If nobody reviews the prepared work, the agent creates another queue. Assign an owner.
It cannot remove approval risk
Customer messages, pricing, proposals, sensitive CRM updates, publishing, legal or privacy wording, finance decisions, deletion, credentials, billing, and DNS should stay behind approval.
How the AI Agent Assessment turns ROI into a build decision
The AI Agent Assessment reviews one workflow for value, frequency, source readiness, permissions, approval gates, risk, and next step.
Validate the numbers against the workflow
The assessment checks whether the time estimate, frequency, review burden, and risk assumptions are realistic enough for a pilot.
Choose build, prepare, wait, or no build
The outcome may be an agent pilot, simpler automation, data cleanup, policy work, a permissions map, or no build. If the workflow is rule-based, compare AI Automation Assessment NZ before committing to an agent.
Frequently asked questions
How do you calculate ROI for an AI agent?
Calculate current monthly manual cost, subtract expected assessment, setup, software, management, and review costs, then adjust for quality, risk, and approval burden. Treat the result as a planning estimate until tested.
What workflows usually have the clearest AI agent ROI?
The clearest candidates are frequent, measurable, reviewable workflows such as lead follow-up preparation, CRM hygiene, weekly reporting, website checks, content operations, proposal preparation, and knowledge summaries.
Should ROI include human review time?
Yes. Human review time is part of safe AI agent operation. If review takes longer than the manual task, the workflow may need better source material, a narrower scope, simpler automation, or no build yet.
What if the workflow saves time but increases risk?
Do not build it first. Customer, privacy, financial, legal, data, or public-content risk needs stronger permissions, approval gates, and source checks before automation.
When should an NZ business pay for an AI Agent Assessment?
Pay for an assessment when one recurring workflow looks valuable enough to improve, but the business needs help validating ROI, mapping sources, setting permissions, approval gates, and choosing build, prepare, wait, or no build.
Next step
Use this calculator to shortlist one workflow. Then test assumptions before buying tools, commissioning a build, or giving AI access to live systems.
Book the $1,000 AI Agent Assessment to validate the ROI estimate, map approval gates, and decide whether to build, prepare, wait, or reject the automation.