The four-hour Monday report is your Priority 1 problem

Start with the recurring cost, not the tool

Pick one process that runs on a schedule — the Monday report, the invoice reconciliation, the weekly inventory count. Write down three numbers: how long it takes end to end, how often it runs, and the fully loaded hourly cost of the people who run it (salary plus benefits and overhead, usually 1.25–1.4x base).

Multiply them. A four-hour report run every Monday by someone costing $60/hour fully loaded is roughly $12,500 a year in direct labour. That number is the ceiling on what automating it can be worth in year one — and it is usually smaller than owners expect.

Then add the costs the spreadsheet hides

Direct labour is the visible part. The larger costs are usually the ones nobody logs:

• Rework: how often does the output get corrected after it ships, and what does the correction cost?
• Latency: what decisions wait on this output, and what does waiting cost in inventory, cash, or a lost deal?
• Key-person risk: if one person owns the process, what happens during their vacation or notice period?
• Opportunity cost: what would that person do with the hours back? If the answer is nothing specific, the savings are soft.

Rank by cost-to-fix, not by pain

The loudest process is rarely the best first project. Score each candidate on annual cost, data readiness (is the input already structured and accessible?), and blast radius if it breaks. High cost, clean data, low blast radius wins — that is your Priority 1.

A process that costs $40,000 a year but depends on scanned PDFs and one person's judgment is a phase-two project. A $12,500 process running off a clean export is a phase-one project you can ship in weeks and use to fund the next one.

Set the threshold before you shop

Decide in advance what payback period you will accept. For small businesses we generally use twelve months on total cost of ownership — build, licences, and the ongoing maintenance that vendors leave off the quote.

Written down first, that threshold makes the buying decision boring. Any tool that cannot clear it gets a no, regardless of the demo.