The AI Agent ROI Calculator Cheat Sheet

Example: agent saves 12 hours/month at a £75/hour equivalent value, costs £40/month to run. Monthly ROI = (12 x 75) minus 40 = £860. Annualised that is £10,320. If the build takes 8 hours at £75/hour (£600), payback is under one month.

Divide last month's total revenue by hours worked. That is your actual rate. Using a fantasy rate inflates the ROI and leads to bad build decisions. If you delegate the task to a VA or contractor, use their rate instead.

If a human still reviews, edits, or approves every output, the task is not replaced. It is assisted. Assisted tasks save roughly 40 to 60 percent of the original time, not 100 percent. Recalculate accordingly or the ROI number is fiction.

Get the per-1,000-token rate for your chosen model and multiply by average tokens per run, then by monthly run volume. For high-volume agents this can be £100 to £400/month. For low-volume it might be £3. Know the number before you commit.

Hours to build and test the agent multiplied by your hourly rate equals your upfront cost. A simple agent might be 3 to 5 hours. A complex multi-step workflow might be 20 to 40 hours. Divide this by monthly ROI to get your payback period in months.

Every agent needs checking. Prompts drift, APIs change, outputs degrade. Budget 1 to 3 hours per month per agent for monitoring and fixes. If you outsource this, budget the contractor cost. Agents that look free to run are not free to maintain.

Zapier, Make, n8n, vector databases, memory layers, third-party APIs. These stack up fast. List every tool the agent depends on and allocate a fair share of each subscription cost to the agent. If the agent uses 30 percent of your Make plan, count 30 percent of the bill.

Agents make mistakes. What is the cost of a wrong output reaching a customer, a missed lead, or a corrupted record? Low-stakes tasks (draft a summary) carry near-zero error cost. High-stakes tasks (send a client proposal, update a CRM, process a payment) carry real financial risk. Factor it in.

Repetitive, rule-based tasks with clear inputs and outputs. Examples: routing inbound enquiries, pulling weekly reports, sending follow-up emails on triggers. Build time under 5 hours. These are the easiest wins and the fastest payback.

Tasks that require judgment, varied inputs, or multi-step logic. Examples: qualifying leads, summarising meeting notes with action items, drafting personalised outreach. Build time 10 to 25 hours. Worth it only if the task runs frequently enough to recover the build cost quickly.

End-to-end workflows replacing a human role or team. Examples: a full content production pipeline, an automated sales research and outreach system. Build time 30 to 80 hours. Require a written business case showing 12-month ROI before committing. Do not build on enthusiasm alone.

If the task happens more than 10 times a week, follows a consistent pattern, and you can define what a correct output looks like in one sentence, it is a strong agent candidate. Volume times consistency equals ROI potential.

Anything where a bad output damages a client relationship, a public-facing brand moment, or a negotiation. Agents can draft, but a human must review. Budget the review time into your hours-saved calculation or the ROI is overstated.

If saving the time does not free you to generate more revenue or reduce headcount, the ROI calculation is theoretical. Ask: what will I do with the hours saved? If the answer is vague, the agent is a nice-to-have, not a profit driver.

Running cost is usually under £20/month on a mid-tier model. Payback period is commonly under two weeks. One of the highest ROI builds for service businesses and consultants.

Running cost ranges from £15 to £60/month depending on volume and model. The real value here is speed, not just hours. Leads that get a response in minutes convert at a higher rate than leads that wait 24 hours.

Running cost is usually £10 to £30/month. Human review time still applies. Savings are real but moderate. Best justified as part of a broader content system, not as a standalone build.

Running cost is £20 to £80/month depending on data volume. High ROI when the saved hours belong to a senior person who bills at £75/hour or above. Low ROI when the hours belong to a junior who could be doing something else.

Any agent that pays back its build cost in under 30 days is a priority. These are almost always Tier 1 tasks with high frequency and low complexity. Every week you delay is a week of ROI you are leaving on the table.

Worthwhile, but not urgent enough to drop everything. Schedule the build into a specific week in the next quarter. Do not let it drift to 'eventually' or it will never happen.

Not every long-payback build is wrong. Some are strategic infrastructure that enables other revenue. But they should not compete for time with short-payback builds. Park them, set a review date for 90 days out, and revisit once your faster wins are running.

An agent that needs correcting 30 percent of the time does not save 100 percent of the task time. Measure actual error rate in a test run of 20 to 50 outputs before finalising your hours-saved estimate.

Getting the prompt right for a production agent takes time. For a simple task it might be 2 hours. For a nuanced one it might be 10 to 15 hours of iteration. This belongs in the build cost, not hidden as 'experimentation.'

New agents in production are not fully trusted yet. You will still check outputs more than you will once the agent is bedded in. Budget 50 percent of normal human review time for the first 60 to 90 days. Your true payback period starts after that adjustment period.