Put in one process, your volumes and your costs, and see the capacity it releases, the errors it avoids and how long the investment takes to pay back. Every assumption is visible and yours to change.
Updated 12 October 2026
C4C is an independent, platform neutral automation practice. We help you decide what is worth automating, choose the platform that fits the process, negotiate the licensing and deliver it. About the practice.
Start with one process you know well. Use fully loaded costs and your own estimates of what can really be automated.
Volumes and handling effort as the process runs now.
Planning assumptions. Replace them with your own; exceptions usually reduce what can be automated.
Your own estimates, from proposals or budgets. No vendor prices are built in.
Get a formatted PDF of these results by email, to keep or share with your team. We send it once and do not add you to any list.
Prefer email? Write to hello@c4cgroup.co.uk.
Annual cases are your monthly volume times twelve. Automated cases are that times the share you can automate. Hours released are automated cases times handling minutes times the share of handling time removed, divided by sixty, and valued at your fully loaded hourly cost.
Errors avoided are automated cases times your rework rate times the share of errors avoided, valued at your cost per error. Annual value is the two added together. Net value per year subtracts the platform licence and the run and support cost.
Payback is the one off build cost divided by the monthly net value. Net over three years is three years of net value less the build cost. All figures are before tax and undiscounted.
Automation business cases are often built on borrowed percentages. Here is what stands up, as checked on 12 October 2026, and what does not.
Median hourly earnings for full time UK employees were £19.67 in April 2025, excluding overtime (ONS). For administrative roles, a 2025 Home Office impact assessment took the ONS median of £14.78 an hour, added 22 percent for non wage costs to reach £18.02, and uplifted that to £19.48 in 2026/27 prices (Home Office). That £19.48 is a sound starting point for clerical work, and the default above uses it, rounded.
If you build the figure from payroll instead, employer National Insurance is 15 percent above £5,000 a year (HMRC), and the minimum automatic enrolment employer pension contribution is 3 percent of qualifying earnings between £6,240 and £50,270 (The Pensions Regulator). Do not add those on top of the 22 percent as well, because that uplift already covers non wage costs. Add office, IT and management overheads only if automation will genuinely reduce them.
The widely quoted 60 to 70 percent figures from McKinsey are a single record to report case from 2017 (McKinsey) and a 2023 estimate of the technical potential of generative AI across the economy (McKinsey). Neither is a straight through processing rate for your process. We found no reliable benchmark linking exception rates to automation share, so measure your own exceptions.
MIT Sloan put it plainly in 2026: an AI agent reclaiming 20 percent of someone’s time does not mean a 20 percent labour cost saving (MIT Sloan). Released hours are capacity. They become savings only when the work is redeployed, overtime or agency spend falls, or recruitment is avoided.
Agents can handle more contextual work than rules based automation, and Gartner expects a third of enterprise software applications to include agentic AI by 2028. Gartner also forecast that more than 40 percent of agentic AI projects will be cancelled by the end of 2027, citing escalating cost, unclear business value and weak return (Reuters, reporting Gartner, June 2025). Budget for integration, human review of the cases that need it, guardrails and governance, not just the licence.
The often repeated claim that 30 to 50 percent of RPA projects fail could not be traced to a primary source, so we do not use it. Nor did we find a reliable independent figure for the cost of back office errors or for annual maintenance as a share of build cost. Use your own numbers for those inputs.
Estimate the hours a process takes, the share of cases that can genuinely be automated and how much handling time automation removes, value those hours at a fully loaded cost, add the value of errors avoided, then subtract the platform, run and support costs. Payback is the build cost divided by the monthly net value.
A 2025 Home Office impact assessment used £19.48 an hour in 2026/27 prices for administrative roles, built from the ONS median wage of £14.78 plus 22 percent for non wage costs. Your own payroll figure is better if you have it, but do not add National Insurance and pension on top of an uplift that already includes them.
It depends on the process and your appetite for risk, but many organisations look for payback within the first year or two of running costs. If a process only pays back late in a three year view, the volume, exception rate or running cost usually needs a second look.
Because exceptions are where automation programmes lose their value. A process where 90 percent of cases follow the rules is a very different case from one where half need a person. Headline figures such as 60 to 70 percent describe technical potential, not your process, so measure the exceptions before you commit.
Only if the time is used. As MIT Sloan noted in 2026, reclaiming 20 percent of someone’s time is not a 20 percent labour cost saving. Hours released become savings when work is redeployed, overtime or agency costs fall, or recruitment is avoided. An honest business case counts the hours you will genuinely use.
They widen what can be automated, because agents can handle more contextual work, but they do not make the return more certain. Gartner forecast that more than 40 percent of agentic AI projects will be cancelled by the end of 2027, citing cost, unclear value and weak return. Budget for integration, human review, guardrails and governance.
No. Count outcomes: cycle time, errors, rework, compliance, capacity released and customer impact. The number of bots says little about value, and a programme measured on bots tends to automate the easy things rather than the valuable ones.
No. It has no vendor names or prices built in. You enter your own estimates of build, licence and running costs, so it works for any kind of automation, from process orchestration and low code to robotic process automation and AI agents.