{"course":{"slug":"ai-for-small-business","title":"AI for a Small Business","url":"https://addaly.com/learn/ai-for-small-business","level":"beginner","track":"work"},"lesson":{"slug":"what-a-call-costs","title":"What a call costs, and how a loop becomes a bill","url":"https://addaly.com/learn/ai-for-small-business/what-a-call-costs","position":60,"of":88,"minutes":9,"answer":"Usage pricing is charged per token in and out, so cost scales with how much context you resend rather than with how many tasks you do.","keyIdea":"Usage pricing is charged per token in and out, so cost scales with how much context you resend rather than with how many tasks you do, which is why an automated loop is a financial exposure and a spending cap is not optional."},"body":{"format":"markdown","text":"## The two pricing worlds\n\nA subscription is a flat monthly fee per person, typically USD 20 to USD 30, for a chat window with limits you will rarely hit. Predictable, and covered in module one.\n\nUsage pricing is what you meet the moment you automate. You pay per token, separately for what goes in and what comes out, and there is no ceiling until you build one.\n\nA token is roughly three-quarters of a word in English, and rather less efficient in most other scripts, a point worth knowing if you work in Hindi, Tamil or Arabic, where the same sentence can cost noticeably more tokens than its English translation.\n\n## Work an actual example\n\nSay you classify and draft a reply to each customer enquiry. Per enquiry:\n\n- Facts sheet and instructions sent in: about 900 tokens.\n- The customer's message: about 200 tokens.\n- The drafted reply out: about 300 tokens.\n\nAt the prices of a small, fast model, take USD 0.15 per million tokens in and USD 0.60 per million out as a realistic order of magnitude: that is about 0.00016 USD in and 0.00018 USD out. Roughly USD 0.0003 per enquiry.\n\nThree hundred enquiries a month is about nine US cents. Not a typo. For ordinary small-business volumes, the model is not the expense; your time and the subscription are.\n\nUse a large frontier model instead, at perhaps twenty times the price, and it becomes about USD 2 a month. Still not the expense.\n\n## So where does the money actually go\n\nThree places, and none of them is the per-enquiry cost.\n\n**Resending context.** You pay for the input every single time. That 900-token facts sheet is charged on every one of the three hundred calls. Double the facts sheet and you double the input bill without doing any more work. This is why the length of your standing instructions is a cost decision, not just a clarity one.\n\n**Long conversations.** Module two explained that the whole thread is re-read on every turn. The cost consequence is that a twenty-turn conversation costs far more than twenty single questions: turn twenty pays for turns one to nineteen again. Prompt caching, offered by most providers, cuts the repeated portion substantially and is worth switching on, but the shape of the problem stays.\n\n**Loops.** This is the one that produces the alarming invoice.\n\n## How a loop becomes a bill\n\nAn automation that calls the model, and whose output triggers something that calls the model again, can run without limit. A retry that does not check whether the first attempt succeeded, a rule that files a document and a rule that processes filed documents, an agent that decides to try once more: each is an ordinary bug, and each can run thousands of times overnight.\n\nThe per-call cost is what makes this dangerous rather than merely embarrassing. At a hundredth of a penny per call, nobody notices ten thousand calls until the statement arrives.\n\nThree defences, and set all three before the first automated call:\n\n1. **A hard spending cap on the account.** Every major provider has one. Set it to a number that would annoy you rather than hurt you: for most small businesses USD 20 or USD 50 a month is generous.\n2. **A separate key for each automation.** When one misbehaves you can revoke it without stopping everything else.\n3. **A counter.** Anything automated should refuse to run more than N times an hour, where N is three times your realistic maximum.\n\n## Rate limits, which are not costs\n\nSeparately from money, providers cap requests per minute and tokens per minute. Hit the cap and you get an error, not a bill.\n\nThis matters for batch work: firing two hundred product descriptions at once will fail partway through. The fix is a pause between calls and a retry with an increasing wait, but read the next lesson before writing any retry, because a naive one causes the duplicate problem.\n\n## The honest comparison\n\nFree tiers exist on most providers and are real. They are also rate-limited, sometimes routed to smaller models, and frequently carry different terms about whether your inputs can be used for training. If the work is customer data, read those terms before the price.\n\nThe free path is running a small open model yourself, covered in module one. The per-call cost is then zero and the cost is your hardware and your attention. For a business doing three hundred enquiries a month, paying nine cents to somebody else is usually the better trade; the free path earns its place when confidentiality, not money, is the binding constraint."},"check":{"question":"A shop automates replies at about 0.0003 USD per enquiry and is pleased at how cheap it is, so it adds a fuller facts sheet, a longer instruction block and three worked examples, taking the input from 900 to 4,000 tokens. Volume is unchanged. What happens to the bill, and why?","options":[{"text":"It stays about the same, because output tokens are priced several times higher than input tokens and the output length has not changed.","correct":false,"tempting":"Notices the price asymmetry correctly, then forgets that a four-fold rise in the cheaper item still multiplies that part of the bill."},{"text":"The input portion roughly quadruples, because the whole block is charged again on every single enquiry rather than once.","correct":true},{"text":"It rises slightly on the first call and then flattens, since the provider caches repeated instructions and charges the full rate only once.","correct":false,"tempting":"Caching does reduce the repeated portion, but it is a discount that must be enabled and does not make resent context free."},{"text":"It falls, because better instructions produce shorter and more accurate replies, and output tokens are the expensive half of the transaction.","correct":false,"tempting":"Assumes a quality improvement pays for itself in output length, which is possible but small next to a four-fold input increase charged three hundred times."}]},"prev":{"slug":"fields-not-prose","title":"Getting fields out instead of paragraphs","url":"https://addaly.com/learn/ai-for-small-business/fields-not-prose"},"next":{"slug":"what-an-agent-actually-is","title":"What an agent is underneath, and why ten steps go wrong","url":"https://addaly.com/learn/ai-for-small-business/what-an-agent-actually-is"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'What a call costs, and how a loop becomes a bill', in AI for a Small Business. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-small-business/what-a-call-costs (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"What a call costs, and how a loop becomes a bill","courseTitle":"AI for a Small Business","module":{"index":7,"slug":"letting-it-act","title":"Letting it act: automation you can trust with the keys"},"answer":null,"keyIdea":"Usage pricing is charged per token in and out, so cost scales with how much context you resend rather than with how many tasks you do, which is why an automated loop is a financial exposure and a spending cap is not optional.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"the-two-pricing-worlds","runs":[{"text":"The two pricing worlds"}]},{"t":"p","runs":[{"text":"A subscription is a flat monthly fee per person, typically USD 20 to USD 30, for a chat window with limits you will rarely hit. Predictable, and covered in module one."}]},{"t":"p","runs":[{"text":"Usage pricing is what you meet the moment you automate. You pay per token, separately for what goes in and what comes out, and there is no ceiling until you build one."}]},{"t":"p","runs":[{"text":"A token is roughly three-quarters of a word in English, and rather less efficient in most other scripts, a point worth knowing if you work in Hindi, Tamil or Arabic, where the same sentence can cost noticeably more tokens than its English translation."}]},{"t":"h2","id":"work-an-actual-example","runs":[{"text":"Work an actual example"}]},{"t":"p","runs":[{"text":"Say you classify and draft a reply to each customer enquiry. Per enquiry:"}]},{"t":"ul","items":[[{"text":"Facts sheet and instructions sent in: about 900 tokens."}],[{"text":"The customer’s message: about 200 tokens."}],[{"text":"The drafted reply out: about 300 tokens."}]]},{"t":"p","runs":[{"text":"At the prices of a small, fast model, take USD 0.15 per million tokens in and USD 0.60 per million out as a realistic order of magnitude: that is about 0.00016 USD in and 0.00018 USD out. Roughly USD 0.0003 per enquiry."}]},{"t":"p","runs":[{"text":"Three hundred enquiries a month is about nine US cents. Not a typo. For ordinary small-business volumes, the model is not the expense; your time and the subscription are."}]},{"t":"p","runs":[{"text":"Use a large frontier model instead, at perhaps twenty times the price, and it becomes about USD 2 a month. Still not the expense."}]},{"t":"h2","id":"so-where-does-the-money-actually-go","runs":[{"text":"So where does the money actually go"}]},{"t":"p","runs":[{"text":"Three places, and none of them is the per-enquiry cost."}]},{"t":"p","runs":[{"text":"Resending context.","b":true},{"text":" You pay for the input every single time. That 900-token facts sheet is charged on every one of the three hundred calls. Double the facts sheet and you double the input bill without doing any more work. This is why the length of your standing instructions is a cost decision, not just a clarity one."}]},{"t":"p","runs":[{"text":"Long conversations.","b":true},{"text":" Module two explained that the whole thread is re-read on every turn. The cost consequence is that a twenty-turn conversation costs far more than twenty single questions: turn twenty pays for turns one to nineteen again. Prompt caching, offered by most providers, cuts the repeated portion substantially and is worth switching on, but the shape of the problem stays."}]},{"t":"p","runs":[{"text":"Loops.","b":true},{"text":" This is the one that produces the alarming invoice."}]},{"t":"h2","id":"how-a-loop-becomes-a-bill","runs":[{"text":"How a loop becomes a bill"}]},{"t":"p","runs":[{"text":"An automation that calls the model, and whose output triggers something that calls the model again, can run without limit. A retry that does not check whether the first attempt succeeded, a rule that files a document and a rule that processes filed documents, an agent that decides to try once more: each is an ordinary bug, and each can run thousands of times overnight."}]},{"t":"p","runs":[{"text":"The per-call cost is what makes this dangerous rather than merely embarrassing. At a hundredth of a penny per call, nobody notices ten thousand calls until the statement arrives."}]},{"t":"p","runs":[{"text":"Three defences, and set all three before the first automated call:"}]},{"t":"ol","start":1,"items":[[{"text":"A hard spending cap on the account.","b":true},{"text":" Every major provider has one. Set it to a number that would annoy you rather than hurt you: for most small businesses USD 20 or USD 50 a month is generous."}],[{"text":"A separate key for each automation.","b":true},{"text":" When one misbehaves you can revoke it without stopping everything else."}],[{"text":"A counter.","b":true},{"text":" Anything automated should refuse to run more than N times an hour, where N is three times your realistic maximum."}]]},{"t":"h2","id":"rate-limits-which-are-not-costs","runs":[{"text":"Rate limits, which are not costs"}]},{"t":"p","runs":[{"text":"Separately from money, providers cap requests per minute and tokens per minute. Hit the cap and you get an error, not a bill."}]},{"t":"p","runs":[{"text":"This matters for batch work: firing two hundred product descriptions at once will fail partway through. The fix is a pause between calls and a retry with an increasing wait, but read the next lesson before writing any retry, because a naive one causes the duplicate problem."}]},{"t":"h2","id":"the-honest-comparison","runs":[{"text":"The honest comparison"}]},{"t":"p","runs":[{"text":"Free tiers exist on most providers and are real. They are also rate-limited, sometimes routed to smaller models, and frequently carry different terms about whether your inputs can be used for training. If the work is customer data, read those terms before the price."}]},{"t":"p","runs":[{"text":"The free path is running a small open model yourself, covered in module one. The per-call cost is then zero and the cost is your hardware and your attention. For a business doing three hundred enquiries a month, paying nine cents to somebody else is usually the better trade; the free path earns its place when confidentiality, not money, is the binding constraint."}]}],"check":{"question":[{"text":"A shop automates replies at about 0.0003 USD per enquiry and is pleased at how cheap it is, so it adds a fuller facts sheet, a longer instruction block and three worked examples, taking the input from 900 to 4,000 tokens. Volume is unchanged. What happens to the bill, and why?"}],"options":[{"text":[{"text":"It stays about the same, because output tokens are priced several times higher than input tokens and the output length has not changed."}],"correct":false,"tempting":[{"text":"Notices the price asymmetry correctly, then forgets that a four-fold rise in the cheaper item still multiplies that part of the bill."}]},{"text":[{"text":"The input portion roughly quadruples, because the whole block is charged again on every single enquiry rather than once."}],"correct":true,"tempting":null},{"text":[{"text":"It rises slightly on the first call and then flattens, since the provider caches repeated instructions and charges the full rate only once."}],"correct":false,"tempting":[{"text":"Caching does reduce the repeated portion, but it is a discount that must be enabled and does not make resent context free."}]},{"text":[{"text":"It falls, because better instructions produce shorter and more accurate replies, and output tokens are the expensive half of the transaction."}],"correct":false,"tempting":[{"text":"Assumes a quality improvement pays for itself in output length, which is possible but small next to a four-fold input increase charged three hundred times."}]}],"order":[3,1,2,0],"why":null},"prev":{"slug":"fields-not-prose","title":"Getting fields out instead of paragraphs","minutes":9},"next":{"slug":"what-an-agent-actually-is","title":"What an agent is underneath, and why ten steps go wrong","minutes":10}}}