Key takeaways<\/h2>\n\n\n\n\n- Lead with hours removed, not revenue attributed.<\/strong> Hours cannot be argued with; attribution can.<\/li>\n\n\n
- Count conversations resolved<\/em> without a person, not deflected.<\/li>\n\n\n
- Time a sample of each question type by hand – twenty minutes of work, and it makes the whole case concrete.<\/li>\n\n\n
- Message fees are usually zero: most small businesses never pass the 1,000 free service messages a month.<\/li>\n\n\n
- Meta counts a human typing in the same category as a bot, so automating does not change the allowance.<\/li>\n\n\n
- Overnight has the cleanest argument, because the alternative is a salary or nothing.<\/li>\n\n\n
- Do not promise headcount reduction. The realistic outcome is more volume on better work.<\/li>\n\n<\/ul>\n\n<\/div>\n\n\n\n
Why Hours Beat Revenue<\/h2>\n\n\n\n
Both are real. Only one survives a finance review intact.<\/p>\n\n\n\n
Attribution Invites the Argument<\/h3>\n\n\n\n
Any revenue claim needs a model: which sales would have happened anyway, how long the attribution window is, whether a click counts. Every one of those is a legitimate question and every one of them moves the conversation away from the saving and towards the methodology.<\/p>\n\n\n\n
You can win that argument and still lose the meeting, because the person reviewing budgets now associates your programme with a debate.<\/p>\n\n\n\n
Hours Are Just Counting<\/h3>\n\n\n\n
Conversations resolved without a person, multiplied by how long each used to take. There is no model, no window, no counterfactual – the work either happened or it did not, and somebody used to do it.<\/p>\n\n\n\n
It also has the advantage of being conservative, which is the right posture for a number you will be held to next quarter.<\/p>\n\n\n\n
Mention Revenue in One Sentence<\/h3>\n\n\n\n
Not never – just not as the justification. \u201cThere is also revenue upside we have not counted\u201d is a strong closing line and a weak opening one.<\/p>\n\n\n\n
If somebody asks for the revenue figure specifically, the honest way to produce it is a holdout comparison, which is the method in recovering lost sales with DMs<\/a>.<\/p>\n\n\n\n![\"A](\"https:\/\/dmpro.brandid.app\/blog\/wp-content\/uploads\/2026\/10\/reduce-support-costs-automation-model.jpg\")
Hours removed, not revenue attributed.<\/figcaption><\/figure>\n\n\n\nWhere the Savings Actually Are<\/h2>\n\n\n\n
Ranked. The one people expect to optimise is last.<\/p>\n\n\n\n
The Repeated Question<\/h3>\n\n\n\n
Order status is usually the largest single line in any support cost model, because it is the highest-volume question with the least judgement in it and the answer already exists in a system you own.<\/p>\n\n\n\n
Removing it removes genuine hours rather than shifting them, which is what makes it the right first build – the mechanics are in looking up an order mid-conversation<\/a>.<\/p>\n\n\n\nOvernight<\/h3>\n\n\n\n
The cleanest argument in the whole subject, because the alternative is not a cheaper person – it is either nobody at all or a night shift. Automation competes against a salary or against reputational damage, and it wins comfortably against both.<\/p>\n\n\n\n
The models and what each costs are in answering DMs outside business hours<\/a>.<\/p>\n\n\n\nFewer Touches per Case<\/h3>\n\n\n\n
\u201cLet me check and come back to you\u201d is two extra exchanges and a second clock. A webhook that returns the real answer in the first reply removes both.<\/p>\n\n\n\n
Moderate in size, easy to measure, and it improves the customer experience at the same time, which is unusual in a cost exercise.<\/p>\n\n\n\n
Message Fees, Which Are Usually Zero<\/h3>\n\n\n\n
The line people expect to optimise and the one that rarely matters. Every business phone number gets 1,000 free service messages a month, and most small businesses never pass it.<\/p>\n\n\n\n
Crucially, Meta counts a reply typed by a person in the same service category as one sent by a bot. Automating therefore does not reduce your allowance consumption – it reduces the hours. The published rates<\/a> matter for forecasting peak months, not for the base case.<\/p>\n\n\n\n![\"Where](\"https:\/\/dmpro.brandid.app\/blog\/wp-content\/uploads\/2026\/10\/reduce-support-costs-automation-where.jpg\")
Rarely in message fees, which is what people expect to optimise.<\/figcaption><\/figure>\n\n\n\nWhat It Costs on the Other Side<\/h2>\n\n\n\n
A case that only has savings in it is not credible. Put the costs in yourself.<\/p>\n\n\n\n
Build and Maintenance Time<\/h3>\n\n\n\n
Someone writes the flows, reads the escalations, and fixes the wording. Realistically a few hours a month after the first fortnight, and it never goes to zero – a flow that nobody maintains drifts, which is the pattern in writing escalation rules<\/a>.<\/p>\n\n\n\nAI, If You Use It<\/h3>\n\n\n\n
Meta’s own Business Agent is priced per token – roughly four to five cents a message – with no free tier and no exemption even inside the free entry point window. A third-party AI, by contrast, counts as an ordinary service message and draws on the free thousand.<\/p>\n\n\n\n
That distinction is worth several hundred a month at moderate volume and almost nobody includes it in a comparison.<\/p>\n\n\n\n
The Cost of Getting It Wrong<\/h3>\n\n\n\n
One confidently wrong automated answer about a refund costs more than a month of saved hours. This belongs in the case as a risk with a mitigation – the escalation rules – rather than as something to leave out and be surprised by.<\/p>\n\n\n\n
A Case That Holds Up<\/h2>\n\n\n\n
Six lines, one page. Present in this order.<\/p>\n\n\n\n
Why Hours Beat Revenue<\/h2>\n\n\n\n
Both are real. Only one survives a finance review intact.<\/p>\n\n\n\n
Attribution Invites the Argument<\/h3>\n\n\n\n
Any revenue claim needs a model: which sales would have happened anyway, how long the attribution window is, whether a click counts. Every one of those is a legitimate question and every one of them moves the conversation away from the saving and towards the methodology.<\/p>\n\n\n\n
You can win that argument and still lose the meeting, because the person reviewing budgets now associates your programme with a debate.<\/p>\n\n\n\n
Hours Are Just Counting<\/h3>\n\n\n\n
Conversations resolved without a person, multiplied by how long each used to take. There is no model, no window, no counterfactual – the work either happened or it did not, and somebody used to do it.<\/p>\n\n\n\n
It also has the advantage of being conservative, which is the right posture for a number you will be held to next quarter.<\/p>\n\n\n\n
Mention Revenue in One Sentence<\/h3>\n\n\n\n
Not never – just not as the justification. \u201cThere is also revenue upside we have not counted\u201d is a strong closing line and a weak opening one.<\/p>\n\n\n\n
If somebody asks for the revenue figure specifically, the honest way to produce it is a holdout comparison, which is the method in recovering lost sales with DMs<\/a>.<\/p>\n\n\n\n Ranked. The one people expect to optimise is last.<\/p>\n\n\n\n Order status is usually the largest single line in any support cost model, because it is the highest-volume question with the least judgement in it and the answer already exists in a system you own.<\/p>\n\n\n\n Removing it removes genuine hours rather than shifting them, which is what makes it the right first build – the mechanics are in looking up an order mid-conversation<\/a>.<\/p>\n\n\n\n The cleanest argument in the whole subject, because the alternative is not a cheaper person – it is either nobody at all or a night shift. Automation competes against a salary or against reputational damage, and it wins comfortably against both.<\/p>\n\n\n\n The models and what each costs are in answering DMs outside business hours<\/a>.<\/p>\n\n\n\n \u201cLet me check and come back to you\u201d is two extra exchanges and a second clock. A webhook that returns the real answer in the first reply removes both.<\/p>\n\n\n\n Moderate in size, easy to measure, and it improves the customer experience at the same time, which is unusual in a cost exercise.<\/p>\n\n\n\n The line people expect to optimise and the one that rarely matters. Every business phone number gets 1,000 free service messages a month, and most small businesses never pass it.<\/p>\n\n\n\n Crucially, Meta counts a reply typed by a person in the same service category as one sent by a bot. Automating therefore does not reduce your allowance consumption – it reduces the hours. The published rates<\/a> matter for forecasting peak months, not for the base case.<\/p>\n\n\n\n A case that only has savings in it is not credible. Put the costs in yourself.<\/p>\n\n\n\n Someone writes the flows, reads the escalations, and fixes the wording. Realistically a few hours a month after the first fortnight, and it never goes to zero – a flow that nobody maintains drifts, which is the pattern in writing escalation rules<\/a>.<\/p>\n\n\n\n Meta’s own Business Agent is priced per token – roughly four to five cents a message – with no free tier and no exemption even inside the free entry point window. A third-party AI, by contrast, counts as an ordinary service message and draws on the free thousand.<\/p>\n\n\n\n That distinction is worth several hundred a month at moderate volume and almost nobody includes it in a comparison.<\/p>\n\n\n\n One confidently wrong automated answer about a refund costs more than a month of saved hours. This belongs in the case as a risk with a mitigation – the escalation rules – rather than as something to leave out and be surprised by.<\/p>\n\n\n\n Six lines, one page. Present in this order.<\/p>\n\n\n\nWhere the Savings Actually Are<\/h2>\n\n\n\n
The Repeated Question<\/h3>\n\n\n\n
Overnight<\/h3>\n\n\n\n
Fewer Touches per Case<\/h3>\n\n\n\n
Message Fees, Which Are Usually Zero<\/h3>\n\n\n\n
What It Costs on the Other Side<\/h2>\n\n\n\n
Build and Maintenance Time<\/h3>\n\n\n\n
AI, If You Use It<\/h3>\n\n\n\n
The Cost of Getting It Wrong<\/h3>\n\n\n\n
A Case That Holds Up<\/h2>\n\n\n\n