Quick answer
Measure ROI as value delivered divided by total cost, where total cost includes the subscription, any overage, and the cost of blocked time. Value is finished tasks or shipped work, not tokens. If you cannot state the value side, compare plans on cost per completed task instead.
The formula
roi = value_of_completed_work / (subscription + overage + blocked_time_cost)Blocked time is the term teams omit. Two blocked afternoons a month at a loaded hourly rate can exceed the price gap between tiers, which is exactly the decision you are trying to make.
Where the inputs come from
| Input | Source | Confidence |
|---|---|---|
| Subscription | Invoice | Exact |
| Overage | Billing | Exact |
| Completed tasks | Issue tracker | Good |
| Blocked hours | Your own log | Estimate |
| Value per task | Team estimate | Low |
Label the low-confidence inputs. A range with honest assumptions is more persuasive than a single number with none.
Use it for one decision
ROI is not a score to maximise; it is a way to choose between two options. Compare the current plan with one alternative over the same period, and keep the measurement running so the next review starts with data. See cost per developer and cost analysis.
Report it as a range, not a score
ROI estimates carry uncertainty on both sides: the value of a completed task is a judgement, and blocked time is a self-reported log. Present a range with the assumptions attached, and say which input would change the decision. A range of 1.4 to 2.1 with a named risk is more useful than a single 1.8.
Keep the same method between reviews. Changing how you value a task halfway through a year makes every comparison meaningless, and the audience notices before you do. The cost analysis framework gives the fixed structure.
Questions, answered
How do I calculate AI coding subscription ROI?
Divide the value of completed work by the subscription plus overage plus the cost of blocked time. State which inputs are estimates.
What if I cannot measure value?
Compare plans on cost per completed task instead. It is a narrower but more defensible measure.
Is a higher tier worth it?
Only when blocked time costs more than the price gap. Count blocked days for a full cycle before deciding.