{"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":"the-handover","title":"Designing the moment it gives up","url":"https://addaly.com/learn/ai-for-small-business/the-handover","position":32,"of":88,"minutes":8,"answer":"The handover is the part of the design that decides whether automation helps, so hand over after two failures, on any money or emotional signal, and instantly on request; and never measure success by deflection.","keyIdea":"The handover is the part of the design that decides whether automation helps, so hand over after two failures, on any money or emotional signal, and instantly on request; and never measure success by deflection, which counts abandonment as a win."},"body":{"format":"markdown","text":"## The failure customers actually remember\n\nNobody minds a bot that cannot answer. Everybody minds a bot that will not let them past.\n\nThe loop is familiar. Three rephrasings of the same non-answer. \"I'm sorry, I didn't understand that.\" No visible way out. The customer leaves, and you never learn that they were trying to give you money.\n\nSo the handover (the moment the system stops and fetches a person) is not an edge case bolted on at the end. It is the part of the design that decides whether the whole thing helps or harms.\n\n## Four triggers, all cheap to implement\n\n**Two failures, then stop.** Not five. If the assistant has not helped after two attempts, it will not help on the third, and each additional attempt costs goodwill that is hard to get back.\n\n**Anything about money or a commitment.** A price for custom work, a discount, a refund, a delivery promise, a change to an existing order. Handover on sight.\n\n**Emotional signal.** Anger, distress, urgency, or the word \"complaint\". A person, immediately. A bot handling a upset customer well is a rare event; a bot handling one badly is the review you will read for two years.\n\n**On request, instantly.** \"Human\", \"person\", \"speak to someone\", \"agent\", always works, at any point, no qualifying questions. Anything else is a dark pattern, and customers recognise it as one.\n\n> Figure: The moment it stops and fetches a person. Steps: 1. Two failures, then stop (Not five. If it has not helped after two attempts it will not help on the third, and each further attempt costs goodwill that is hard to get back.); 2. Money or a commitment (A custom price, a discount, a refund, a delivery promise, a change to an existing order. Handover on sight.); 3. Emotional signal (Anger, distress, urgency, or the word complaint. A person, immediately.); 4. On request, instantly (Human, person, agent, speak to someone. At any point, with no qualifying questions. Anything else is a dark pattern and customers recognise it.); 5. Carry the transcript across (They have already typed the order number. Making them repeat it is the moment they conclude the automation was for your benefit rather than theirs.). Never judge this by deflection rate, which counts the customer who left in frustration as a success. Count how many enquiries became a booking, an order or an answer the customer confirmed.\n\n## The handover itself\n\nTwo rules.\n\n**Carry the conversation across.** The customer has already typed their order number and explained the problem. Making them repeat it is the moment they conclude the automation was for your benefit rather than theirs. Whatever tool you use must pass the transcript to the person.\n\n**Be honest about the wait.** \"Kemi will reply within four hours: we're open until six\" is a promise you can keep. \"Connecting you to an agent\" when nobody is there until Monday is worse than saying nothing, because it converts a mild disappointment into a broken promise.\n\n## Out of hours\n\nMost small businesses are not staffed at 11pm, and the honest version wins:\n\n> *We're closed now. I can answer questions about prices, opening hours and delivery areas. For anything else, leave a message and Kemi will reply by 10am tomorrow.*\n\nThat sets a boundary, delivers value inside it, and makes a keepable promise. Compare it with a bot that pretends to be a full service at midnight and produces an invented answer nobody will honour in the morning.\n\n## The transcript is the diagnosis\n\nEvery handover is a small report on where your system falls short. Read them weekly and sort them into three piles.\n\n**Should have been answered.** A gap in your documents. Write the missing page.\n\n**Correctly escalated.** Working as designed. Note the volume; if this pile is large, the bot is scoped too narrowly, or too broadly.\n\n**Gave up before asking.** The saddest pile and the most useful. Look for the last message before the silence. It is usually the same three or four questions, and they tell you exactly what is missing.\n\n## The measure that matters\n\nNot deflection rate. Every vendor reports it, and it counts conversations that did not reach a human, which includes every customer who left in frustration. Deflection and abandonment look identical in that number.\n\nMeasure instead: **how many enquiries turned into what you wanted**: a booking, an order, an answered question the customer confirmed. That is harder to count and it is the only one that distinguishes a bot that helped from a bot that got in the way.\n\n## Staffing the other side\n\nA handover promise is only as good as the person on the other end. Before you turn anything on, decide three things: who receives handovers, within what time, and what happens when they are on holiday.\n\nIf the answer is \"the owner, whenever she sees it\", say that honestly in the bot's wording rather than implying a support desk. Small businesses are trusted for being small. Pretending to be large is the thing that breaks trust, not the four-hour wait.\n\n## The transcript belongs in the record\n\nWhen a conversation is handed over, the transcript should end up wherever you keep customer history: the helpdesk, the CRM, or an email to yourself.\n\nTwo reasons. The person picking it up has the context. And if a dispute follows, you have what was actually said rather than the customer's recollection of what the bot promised. Given that a bot's statements may bind the business, a record of its statements is not optional bookkeeping."},"check":{"question":"A vendor reports that a shop's bot achieved a 78% deflection rate in its first month. Why is this a weak basis for judging it?","options":[{"text":"Deflection counts every conversation that did not reach a person, so a customer who gave up in frustration is recorded exactly like one whose question was answered.","correct":true},{"text":"One month is too short for a deflection figure to stabilise, so the number will change substantially as the bot handles more varied questions.","correct":false,"tempting":"Questions the sample rather than the metric; a stable 78% measured over a year would be just as uninformative about whether customers were helped."},{"text":"Deflection rates are calculated differently by different vendors, so the figure cannot be compared against any external benchmark or industry standard for support automation.","correct":false,"tempting":"Objects to comparability when the problem is what the metric counts, and a perfectly standardised deflection rate would still merge success with abandonment."},{"text":"A 78% deflection rate is implausibly high for a first month, suggesting the bot is answering questions it should be escalating.","correct":false,"tempting":"Treats a high figure as suspicious on instinct; the number carries no information either way, which is the actual reason it cannot support a judgement."}]},"prev":{"slug":"a-bot-on-your-website","title":"Putting a bot on your website, with your eyes open","url":"https://addaly.com/learn/ai-for-small-business/a-bot-on-your-website"},"next":{"slug":"complaints-and-refunds","title":"The angry message, and what your reply commits you to","url":"https://addaly.com/learn/ai-for-small-business/complaints-and-refunds"},"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) 'Designing the moment it gives up', in AI for a Small Business. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-small-business/the-handover (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":"Designing the moment it gives up","courseTitle":"AI for a Small Business","module":{"index":4,"slug":"customers-and-the-front-desk","title":"Customers, from first message to resolved complaint"},"answer":null,"keyIdea":"The handover is the part of the design that decides whether automation helps, so hand over after two failures, on any money or emotional signal, and instantly on request; and never measure success by deflection, which counts abandonment as a win.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"type":"flow","steps":[{"label":"Two failures, then stop","detail":"Not five. If it has not helped after two attempts it will not help on the third, and each further attempt costs goodwill that is hard to get back."},{"label":"Money or a commitment","detail":"A custom price, a discount, a refund, a delivery promise, a change to an existing order. Handover on sight."},{"label":"Emotional signal","detail":"Anger, distress, urgency, or the word complaint. A person, immediately."},{"label":"On request, instantly","detail":"Human, person, agent, speak to someone. At any point, with no qualifying questions. Anything else is a dark pattern and customers recognise it."},{"label":"Carry the transcript across","detail":"They have already typed the order number. Making them repeat it is the moment they conclude the automation was for your benefit rather than theirs."}],"title":"The moment it stops and fetches a person","caption":"Never judge this by deflection rate, which counts the customer who left in frustration as a success. Count how many enquiries became a booking, an order or an answer the customer confirmed."},"alt":"Figure: The moment it stops and fetches a person. Steps: 1. Two failures, then stop (Not five. If it has not helped after two attempts it will not help on the third, and each further attempt costs goodwill that is hard to get back.); 2. Money or a commitment (A custom price, a discount, a refund, a delivery promise, a change to an existing order. Handover on sight.); 3. Emotional signal (Anger, distress, urgency, or the word complaint. A person, immediately.); 4. On request, instantly (Human, person, agent, speak to someone. At any point, with no qualifying questions. Anything else is a dark pattern and customers recognise it.); 5. Carry the transcript across (They have already typed the order number. Making them repeat it is the moment they conclude the automation was for your benefit rather than theirs.). Never judge this by deflection rate, which counts the customer who left in frustration as a success. Count how many enquiries became a booking, an order or an answer the customer confirmed."}},"blocks":[{"t":"h2","id":"the-failure-customers-actually-remember","runs":[{"text":"The failure customers actually remember"}]},{"t":"p","runs":[{"text":"Nobody minds a bot that cannot answer. Everybody minds a bot that will not let them past."}]},{"t":"p","runs":[{"text":"The loop is familiar. Three rephrasings of the same non-answer. “I’m sorry, I didn’t understand that.” No visible way out. The customer leaves, and you never learn that they were trying to give you money."}]},{"t":"p","runs":[{"text":"So the handover (the moment the system stops and fetches a person) is not an edge case bolted on at the end. It is the part of the design that decides whether the whole thing helps or harms."}]},{"t":"h2","id":"four-triggers-all-cheap-to-implement","runs":[{"text":"Four triggers, all cheap to implement"}]},{"t":"p","runs":[{"text":"Two failures, then stop.","b":true},{"text":" Not five. If the assistant has not helped after two attempts, it will not help on the third, and each additional attempt costs goodwill that is hard to get back."}]},{"t":"p","runs":[{"text":"Anything about money or a commitment.","b":true},{"text":" A price for custom work, a discount, a refund, a delivery promise, a change to an existing order. Handover on sight."}]},{"t":"p","runs":[{"text":"Emotional signal.","b":true},{"text":" Anger, distress, urgency, or the word “complaint”. A person, immediately. A bot handling a upset customer well is a rare event; a bot handling one badly is the review you will read for two years."}]},{"t":"p","runs":[{"text":"On request, instantly.","b":true},{"text":" “Human”, “person”, “speak to someone”, “agent”, always works, at any point, no qualifying questions. Anything else is a dark pattern, and customers recognise it as one."}]},{"t":"h2","id":"the-handover-itself","runs":[{"text":"The handover itself"}]},{"t":"p","runs":[{"text":"Two rules."}]},{"t":"p","runs":[{"text":"Carry the conversation across.","b":true},{"text":" The customer has already typed their order number and explained the problem. Making them repeat it is the moment they conclude the automation was for your benefit rather than theirs. Whatever tool you use must pass the transcript to the person."}]},{"t":"p","runs":[{"text":"Be honest about the wait.","b":true},{"text":" “Kemi will reply within four hours: we’re open until six” is a promise you can keep. “Connecting you to an agent” when nobody is there until Monday is worse than saying nothing, because it converts a mild disappointment into a broken promise."}]},{"t":"h2","id":"out-of-hours","runs":[{"text":"Out of hours"}]},{"t":"p","runs":[{"text":"Most small businesses are not staffed at 11pm, and the honest version wins:"}]},{"t":"quote","runs":[{"text":"We’re closed now. I can answer questions about prices, opening hours and delivery areas. For anything else, leave a message and Kemi will reply by 10am tomorrow.","i":true}]},{"t":"p","runs":[{"text":"That sets a boundary, delivers value inside it, and makes a keepable promise. Compare it with a bot that pretends to be a full service at midnight and produces an invented answer nobody will honour in the morning."}]},{"t":"h2","id":"the-transcript-is-the-diagnosis","runs":[{"text":"The transcript is the diagnosis"}]},{"t":"p","runs":[{"text":"Every handover is a small report on where your system falls short. Read them weekly and sort them into three piles."}]},{"t":"p","runs":[{"text":"Should have been answered.","b":true},{"text":" A gap in your documents. Write the missing page."}]},{"t":"p","runs":[{"text":"Correctly escalated.","b":true},{"text":" Working as designed. Note the volume; if this pile is large, the bot is scoped too narrowly, or too broadly."}]},{"t":"p","runs":[{"text":"Gave up before asking.","b":true},{"text":" The saddest pile and the most useful. Look for the last message before the silence. It is usually the same three or four questions, and they tell you exactly what is missing."}]},{"t":"h2","id":"the-measure-that-matters","runs":[{"text":"The measure that matters"}]},{"t":"p","runs":[{"text":"Not deflection rate. Every vendor reports it, and it counts conversations that did not reach a human, which includes every customer who left in frustration. Deflection and abandonment look identical in that number."}]},{"t":"p","runs":[{"text":"Measure instead: "},{"text":"how many enquiries turned into what you wanted","b":true},{"text":": a booking, an order, an answered question the customer confirmed. That is harder to count and it is the only one that distinguishes a bot that helped from a bot that got in the way."}]},{"t":"h2","id":"staffing-the-other-side","runs":[{"text":"Staffing the other side"}]},{"t":"p","runs":[{"text":"A handover promise is only as good as the person on the other end. Before you turn anything on, decide three things: who receives handovers, within what time, and what happens when they are on holiday."}]},{"t":"p","runs":[{"text":"If the answer is “the owner, whenever she sees it”, say that honestly in the bot’s wording rather than implying a support desk. Small businesses are trusted for being small. Pretending to be large is the thing that breaks trust, not the four-hour wait."}]},{"t":"h2","id":"the-transcript-belongs-in-the-record","runs":[{"text":"The transcript belongs in the record"}]},{"t":"p","runs":[{"text":"When a conversation is handed over, the transcript should end up wherever you keep customer history: the helpdesk, the CRM, or an email to yourself."}]},{"t":"p","runs":[{"text":"Two reasons. The person picking it up has the context. And if a dispute follows, you have what was actually said rather than the customer’s recollection of what the bot promised. Given that a bot’s statements may bind the business, a record of its statements is not optional bookkeeping."}]}],"check":{"question":[{"text":"A vendor reports that a shop’s bot achieved a 78% deflection rate in its first month. Why is this a weak basis for judging it?"}],"options":[{"text":[{"text":"Deflection counts every conversation that did not reach a person, so a customer who gave up in frustration is recorded exactly like one whose question was answered."}],"correct":true,"tempting":null},{"text":[{"text":"One month is too short for a deflection figure to stabilise, so the number will change substantially as the bot handles more varied questions."}],"correct":false,"tempting":[{"text":"Questions the sample rather than the metric; a stable 78% measured over a year would be just as uninformative about whether customers were helped."}]},{"text":[{"text":"Deflection rates are calculated differently by different vendors, so the figure cannot be compared against any external benchmark or industry standard for support automation."}],"correct":false,"tempting":[{"text":"Objects to comparability when the problem is what the metric counts, and a perfectly standardised deflection rate would still merge success with abandonment."}]},{"text":[{"text":"A 78% deflection rate is implausibly high for a first month, suggesting the bot is answering questions it should be escalating."}],"correct":false,"tempting":[{"text":"Treats a high figure as suspicious on instinct; the number carries no information either way, which is the actual reason it cannot support a judgement."}]}],"order":[2,3,0,1],"why":null},"prev":{"slug":"a-bot-on-your-website","title":"Putting a bot on your website, with your eyes open","minutes":10},"next":{"slug":"complaints-and-refunds","title":"The angry message, and what your reply commits you to","minutes":9}}}