The two cost lines
Every automation has a build cost and a run cost. Build cost covers diagnosis, design, integration, agents and testing; for a focused first automation (one or two systems, one clear process) this is typically a small fixed-scope project. Multi-system programmes cost more because integration, not AI, is where the hours go.
Run cost has three parts: model usage (pennies to a few dollars per hundred tasks, depending on model choice), infrastructure (usually modest), and maintenance (the honest line most quotes omit: prompts need tuning, APIs change, edge cases surface). Budget maintenance, or the system decays and you will blame AI.
The saving side of the ledger
Count hours, then count cost per hour, then discount for what those hours were actually worth. Data entry is cheap per hour but blocks expensive people downstream; quoting delays cost revenue, not salary. The right comparison is not time saved but outcome changed: faster response, fewer errors, capacity to grow without hiring.
Our ROI calculator walks through this with your numbers, and it deliberately shows pessimistic scenarios too. If an automation only pays back under optimistic assumptions, it is not an automation project yet.
Where budgets go wrong
Three patterns account for most disappointed budgets. First, automating a process that should be deleted: automation multiplies whatever it is given, including waste. Second, underestimating integration: the model is a small part of the work; getting data in and out of your systems is the project. Third, buying tool subscriptions without the implementation: software that needs a full-time person to babysit it is not automation.
- Simplify the process before automating it
- Assume integration is most of the effort, because it is
- Budget maintenance from month one
A sane sequence
Start with one automation that touches money or time weekly, prove it boring and dependable, then expand. Teams that sequence this way end up with three working automations and compounding trust; teams that attempt the grand programme often end up with a cancelled project and organisational scar tissue.