The AI Agent That Assumes: Why Automation Needs to Ask, Not Guess
AI agents are good at filling in gaps quietly, which is exactly the problem. Here's why business automation needs to be built to ask rather than assume, with real examples.
Manual admin cost calculator
Repetitive admin rarely shows up as a line on the accounts, but it is paid for in staff hours every single week. Put in a few numbers and see what one process costs per year, and how much automation could give back.
Estimated annual cost
£0
Adjust the inputs to see the numbers update.
£0 per month·0 hours per week
With your implementation cost
Figures are estimates to help you weigh up a process. Actual savings depend on the process and how it is automated. This is not a quotation.
Could this process be automated?
If the number above made you wince, it is probably worth a look. Tell me how the process works today and I will tell you, honestly, whether it is worth automating and roughly what it would take.
Discuss this processThe costliest admin is rarely one big job. It is the small, repetitive tasks that each take a few minutes and happen constantly: copying an order from an email into a system, re-keying the same customer details into three places, chasing a sign-off, exporting a spreadsheet to build the same report every week. Individually they feel trivial. Added up across a team and a year, they are often someone's entire salary.
The three patterns worth watching for are duplicate data entry, moving information between systems that do not talk to each other, and routine documents or reports produced by hand. All three are usually strong candidates forworkflow automation or a piece of custom software that removes the manual step entirely.
If the number above is significant, the next question is not "which tool" but "what is the process, really". That is where I start: mapping how the work flows today before recommending anything. You can read more on that below, or justtell me about the process.
Insights
AI agents are good at filling in gaps quietly, which is exactly the problem. Here's why business automation needs to be built to ask rather than assume, with real examples.
When you inherit software with no documentation and no developer to ask, the temptation is to rewrite it fast. Here's a safer way to work out what it actually does first.
AI coding agents don't carry memory of past decisions between sessions the way a human developer does. That gap shows up as inconsistency in your codebase, and it matters more than most people realise.
It multiplies the number of people doing the task by how long each occurrence takes and how often it happens, to get total hours per year. That is multiplied by your average employment cost per hour to give an annual cost. The potential saving applies your estimated automation percentage to that figure.
Use a fully loaded figure where you can: gross salary plus employer National Insurance, pension and other on-costs, divided by working hours. A rough rule of thumb is to add around 20 to 30 percent to the base hourly wage. If you are not sure, the base hourly rate still gives a useful, if conservative, estimate.
For repetitive, rules-based admin like copying data between systems, chasing updates or generating routine documents, 60 to 90 percent is often achievable. Tasks that need judgement or handling exceptions will be lower. The slider lets you model a cautious and an optimistic scenario.
No. They are an estimate to help you decide whether a process is worth looking at. Actual savings depend on the specifics of the process and how it is automated. Treat the result as a starting point for a conversation, not a formal projection.
The usual candidates are repetitive data entry, copying information between systems that do not talk to each other, re-keying the same details, chasing approvals and producing routine reports or documents. If a person does it the same way every time, it can usually be automated in part or in full.
Worth a proper look?
Send me a couple of lines about the process and what it involves. You'll get an honest view on whether automating it stacks up, straight from the person who'd build it.
Or email me at josh@thackr.co.uk