Make: budget a pilot by actions, not customer submissions

A customer sends one request. Your business sees one task. But Make may carry out several actions before that request becomes something useful.

That is the number to examine before choosing a plan. Counting submissions alone can make an allowance look more generous than it will be once the work starts.

Start with the credit arithmetic

Make lists 1,000 monthly credits on its Free plan. Its pricing FAQ says most actions consume one credit, while some advanced AI features consume more. One customer submission is therefore not necessarily one credit.

Illustration—not a tested result: four standard one-credit actions × 300 submissions = 1,200 credits. That is already 200 credits above the Free allowance, before any additional consumption.

The assumption matters: exactly four actions, each consuming one credit. This is not a promise about what an AI task will cost. Advanced AI usage, extra steps, retries, and external model charges may change the bill.

A Free allowance can still help you explore a small pilot. Just do not treat “1,000 credits” as “1,000 completed customer requests.” Those are different units.

Count a finished task, not just an arrival

If you want to explore this in Make, choose one narrowly defined customer task. For example, you could propose a pilot that prepares a response for staff review rather than sending an AI answer straight to the customer.

That is a proposed build, not an existing tested template. Before building it, decide what “completed” means: the information arrived, the response was prepared, and a person checked that it was usable.

During a small test, compare the credits consumed with the number of usable completed requests. Include unsuccessful attempts in the consumption total. A request that needs a retry has not become free simply because the first answer was unusable.

Then compare that observed consumption with your expected monthly volume. Keep the four-action illustration separate from your test results. Your actual task may involve fewer actions, more actions, or different AI consumption.

Keep the other costs in view

Credits are one part of the purchase decision. If the task uses a separately billed AI provider, include those charges too. Add the time spent setting it up, checking responses, correcting mistakes, and handling requests that need another attempt.

Before quoting a paid-plan price, confirm the selected credit allowance, billing period, and any upfront annual commitment. This issue deliberately leaves out a precise paid-plan price: a displayed monthly figure is not enough to establish the purchase terms.

For the time comparison, use the same task done manually as your starting point. Compare staff minutes per usable result, including review—not just how quickly the AI produces text.

The result you want is a clearer buying decision: expected consumption, charges, and staff effort for work your business can actually use. Released staff capacity can be valuable, but it is not automatically cash savings.

NYC proposes rules for AI customer service

A customer-facing AI tool can look inexpensive until a customer cannot get help with a refund or a billing problem. That is a different kind of cost: the task remains unresolved even if the automated conversation keeps going.

New York City's consumer agency proposed a rule on October 20 addressing automated customer service. It is a proposal, not a final rule, and it concerns New York City—not a nationwide obligation.

What the proposal would require

Businesses would have to disclose promptly that a customer is dealing with an automated system. They would also have to offer a one-step transfer to a human, or instructions for reaching one if no person is available.

The city's announcement describes fines as starting at $525 under the proposed rule. That figure belongs to the proposal; it should not be read as an already-final requirement.

Read the official NYC announcement for the proposed terms. Do not assume that a tool's features, or the presence of a chatbot, settle how a future rule might apply to your business.

Try the customer experience yourself

Here is one generic practical check—not legal advice: use your own chatbot or phone menu to try to reach a person about a refund or billing problem.

Follow the route a customer would see. Is the human-contact option clear? If nobody is available, do the instructions tell you how to reach someone, or do they send you back into the same automated exchange?

The useful result is concrete: you find out whether someone with a money-related problem can get beyond the automated response. Passing this check is not a legal-compliance determination. It is a way to spot an obvious customer-service obstacle.

That connects directly to the Make budget question. A completed customer task should mean a useful outcome, not merely a submitted form or another generated message. Check the consumption—and check that a person remains reachable.

Quick Win

Before your next Make test, write down three quantities: expected monthly submissions, actions per submission, and credits per action. For the illustration in this issue, those are 300, four, and one.

Multiply them: 1,200 credits. Compare that with the 1,000-credit Free allowance, then mark the total “illustrative—before extra consumption.”

Beside it, leave space for actual test consumption, completed usable requests, any external AI charges, and staff review time. You now have a specific question to answer before buying: does the real task fit the allowance and effort you can afford?