In the meeting where an AI project is proposed, the owner of a small business usually listens to the list of capabilities and then asks one question: when does this money come back? Having no answer to that question sinks more projects than any technical objection does.

Calculating return on investment here is harder than buying a machine, but it is not impossible; the costs have a hidden component and the benefits look qualitative at first glance. The work is to turn both sides of that equation into a number written down before the project starts, not after the budget is spent.

The cost side: what usually goes uncounted

• Data preparation: collecting, cleaning and merging records; in most projects this is the largest part of the work.
• Integration: connecting the new tool to the existing finance, inventory or sales system, at a cost tied to the state of those systems.
• Running cost: the subscription or processing charge paid per use, which rises as the scope widens.
• Training and human review: the hours the team spends learning and checking output; this cost falls but never reaches zero.

The benefit side: three separate categories

Do not add the benefits into a single number, because they do not carry the same degree of certainty. Time saved is the most measurable category: hours freed, multiplied by the real hourly cost of that role. Added revenue, such as a faster reply that converts into more sales, is plausible but less certain. The third category is avoided loss: fewer errors, less duplicated work, less scrap; that category holds the most value and receives the least attention, because it has no dedicated line in the financial statements.

From benefit to number: a straightforward method

Three steps are enough. First, measure the current state; without that baseline, any claim of improvement is disputable. Second, convert the monthly saving into money: hours freed multiplied by the hourly cost, and errors avoided by the average cost of an error. Third, derive the break-even point by dividing the setup cost by the net monthly saving, where net means the saving minus the running cost. If that figure runs longer than your planning horizon, shrink the scope of the project rather than enlarging your hopes.

Cautions specific to Iran's economy

Two factors make this calculation different from foreign examples. First, inflation and currency volatility: the cost of foreign-currency subscriptions and some processing services does not move in step with rial revenue, so assess the break-even point against the plausible worst case, not today's rate. Second, a short planning horizon: where multi-year forecasting is difficult, a project that pays for itself within months beats a larger one promising returns in two years, even if its nominal return is lower.

A three-month measurement path

• Month one: choose one process and measure its current state: time, volume, error rate and the cost of the people involved.
• Month two: run the pilot in a limited scope and record every cost, including internal working hours.
• Month three: compare the result with the baseline and calculate the real break-even point; let that figure decide whether to continue or stop.

The mistakes that make the number false

• Having no baseline; an improvement with nothing to compare against is a claim, not a result.
• Booking freed hours as savings when overtime has not fallen, no planned hire was avoided and no new revenue-generating work was done.
• Counting the cost of the tool while ignoring internal working hours, which are often larger.
• Attributing the effect of outside events, such as seasonal market growth, to the project.
• Generalising the result of a pilot that ran with the team's special attention to the whole organisation.
• Leaving maintenance out; a system that is not kept current loses its return gradually.

Frequently asked questions

• What is a reasonable break-even point?
  There is no fixed figure, but a range that matches your planning horizon and cash position is preferable to any attractive nominal return.
• What if the main benefit is qualitative?
  It can usually be translated into a proxy number; higher satisfaction into customer retention rate, and better decisions into error rate.
• When should a project be stopped?
  When the metric you set before starting does not move within the agreed window; stopping early is a correct decision, not a failure.

Takeaway

The return on an AI investment is neither proven by optimism nor refuted by pessimism; it is assessed with a baseline, a complete list of costs, and a metric written down in advance. Start small, count honestly, and scale only what has demonstrably paid for itself.

Glossary

• Return on investment: the ratio of the net benefit of an investment to the cost incurred for it.
• Break-even point: the moment when a project's accumulated benefit equals its cost.
• Total cost of ownership: the sum of setup, running and maintenance costs across a system's life.
• Baseline: a numeric picture of current performance, taken before any change, for later comparison.
• Avoided loss: a cost no longer paid because errors or scrap have been reduced.