Count the full cost of an AI workflow.
Organise a fictional workflow estimate around usage, preparation, review and upkeep before comparing two approaches.
A synthetic planning exercise. No vendor prices, market trends, research results or customer costs are reported.
Choose the boundary of the estimate
An estimate is easier to examine when it names the task it covers. Preparing a short internal handoff has a different boundary from maintaining a document collection for an entire organisation. State which work enters the comparison, which output counts as complete and which activities sit outside it. Otherwise two apparently comparable totals may describe different amounts of work.
For this fictional exercise, imagine a recurring handoff that requires finding approved notes, preparing a draft and checking unresolved decisions. A usage charge would describe only one input to that workflow. Preparation, review and maintenance may also need attention. Listing those activities does not prove that one approach costs more or less; it makes the assumptions available for discussion.
Keep unlike quantities visible
Separate usage units, people time and recurring upkeep before combining anything. A count of requests cannot be added to a count of minutes. If a planning discussion uses a monetary conversion, identify the conversion assumption and its basis, and keep the original quantities alongside it. Do not hide a disputed assumption inside a polished total.
Review time also needs a consistent definition. Does it include opening the source notes, correcting the draft and resolving conflicting information? Does upkeep include checking whether the material is still appropriate to use? A comparison that includes these activities for one method and omits them for another will be hard to interpret even if its arithmetic is correct.
Test the assumption that changes the decision
Identify which uncertain input has the largest effect on the proposed decision. In a small fictional handoff, that might be the time required to check a difficult draft rather than the number of drafts generated. Try an ordinary case and a case with missing context, keeping both clearly labelled as examples until observations are available. A single convenient case should not become a general expectation.
An estimate should explain what remains unknown and when it will be revisited. Preserve the underlying quantities, the proposed scope and the reason for each assumption so another reviewer can reproduce the comparison. The output is a planning question to investigate, not a quote, a purchasing recommendation or evidence of an operating service. No inputs need to be sent from this website to carry out the educational exercise.