Answer
What can go wrong when you automate customer messages?
Loops, dead ends, wrong recipients and broken promises. Almost none of the real failures are the model saying something odd.
Replies with nowhere to go, promises nobody staffed, loops between two automations, and messages sent to a stale contact. The model saying something strange is the least common and most recoverable failure of the set.
Anxiety about automated messaging concentrates on the model producing something embarrassing. That does happen, it is usually recoverable, and it is far from the main risk. The failures that actually damage businesses are plumbing failures — they occur after the message has been sent, and they occur reliably rather than randomly.
The most common is the reply that goes nowhere. An automated message is sent from an address or number that nobody monitors, the customer answers it with a real question, and the answer sits unread. This is worse than never having messaged, because the customer has now been given evidence that the business is contactable and has demonstrated that it is not. Every automated outbound message needs a monitored return path, and the number of businesses that discover this by finding a year of unread replies is not small.
The second is the promise nobody staffed. An acknowledgement that says somebody will call within the hour is a commitment, and it is being made automatically at three in the morning on a Sunday. If the rota does not exist, the automation has converted a customer who was waiting into a customer who was let down at a specific, documented time. The rule is simple and frequently broken: an automated message may only promise things that are true at the hour it is sent.
The third is loops. Two automations that each respond to the other — an out-of-office replying to an acknowledgement, a ticketing system and a messaging tool each creating a record from the other's output — produce message storms that are visible to the customer and hard to stop under pressure. They come from adding a second automation without mapping what the first one does, which is how most small-business stacks grow.
The fourth is the wrong recipient, and it is the one with genuine legal exposure. Messaging a number that has been reassigned, continuing to message somebody who asked to stop, or sending to a list assembled without a basis for contacting it. Consent and opt-out handling are rules about outbound messaging that apply whether the sender is a person or a system, and automation's contribution is that it does the wrong thing at scale and without pausing. An opt-out has to work immediately and across every channel and system that holds the contact, which is harder than it sounds once a business has three tools.
The fifth is quieter and slower: tone drift. Automated sequences accumulate. A follow-up is added, then a reminder, then a re-engagement message, each reasonable alone, and nobody ever reads the whole series as one customer receives it. The result is a business that sounds pushy to its own customers without any individual having decided to be. The cheap defence is to walk the full sequence as a customer once a quarter, which almost nobody does and which surfaces this immediately.
The expensive failures in automated messaging are almost never about what was said; they are about what happened when somebody answered.
Answer Production Engine, Context Theory
Related questions
How do we test an automated sequence before it goes live?
Run yourself through it end to end as a real customer, on a real device, including replying at points the design did not anticipate. Then run the awkward cases: reply with a question, reply with an opt-out, reply after a long delay, reply from a different number. Most of the failures above surface within an hour of doing this, and almost none surface from reviewing the message text in a document.
Should automated messages come from a person's name?
From the business, with a person named as the contact if someone will genuinely respond. Sending automation from an individual's name creates the failure where the customer replies expecting that person and reaches a system, and it puts an employee's name on messages they did not write and may not be able to answer for.
METHOD
Every figure below carries its source and the date it was verified. Nothing on this page is asserted.
The numbers on this page.
| What | Value | Specific to |
|---|---|---|
| Firms that never responded to a web enquiry at all | 23% | Category-wide |
| Average B2B first-response time | 42 hrs | Category-wide |
| Agents who give up after one contact | 44% | Category-wide |
Oldroyd, McElheran & Elkington, "The Short Life of Online Sales Leads", Harvard Business Review (March 2011) · hours · 1.25M inbound leads across 2,241 US firms · verified
Multi-study aggregate · verified
What is specific to this page.
| Kind | Claim | Check it against |
|---|---|---|
| Software | An automated outbound message sent from an unmonitored address or number produces replies that are never read, which is worse than not messaging because the customer has been shown the business is contactable and then that it is not. | The inbox or number each automated message sends from, checked for unread inbound replies and for who is assigned to monitor it. |
| Workflow | An automated acknowledgement that promises a callback within a stated period is a commitment made at whatever hour it fires, so it may only promise what is true at that hour, and an unstaffed promise converts a waiting customer into a documented failure. | The message template's promised response window, compared with the staffing rota for the hours the automation is active. |
| Software | Two automations that each respond to the other's output produce message loops visible to the customer, and they arise from adding a second tool without mapping the triggers of the first, which is how small-business stacks typically grow. | A trigger map of every automation in the stack, checked for any tool whose output is another tool's input. |
| Constraint | Consent and opt-out obligations for outbound messaging apply regardless of whether a person or a system sends, and an opt-out must take effect immediately across every tool holding the contact, which is materially harder once a business runs several tools. | Whether an opt-out recorded in one system propagates to the others, tested by opting out and then triggering each remaining sequence. |
Each row would be wrong on another industry's page. Where a sourced figure exists it is in the table above instead; these are the constraints that shape the work and do not happen to be numbers.
Start with the measurement.
Reading about a benchmark is not the same as knowing your own number. The audit produces yours, measured rather than estimated.
$497 · delivered in 5 business days · credited against month one