Chatbot Handover to Humans: The Test Singapore SMEs Should Run Before Black Friday

The number that decides whether your chatbot survives the festive peak is not how many questions it answers. It is what happens in the 30 seconds after it gives up. Three measurements tell you: the handover rate (what share of chats the bot passes to a person), the handover wait (how long that person takes to reply), and the post-handover drop-off (how many customers leave before the person arrives). If you only track one, track the drop-off. A bot that escalates cleanly at 40 percent is worth far more than one that escalates at 10 percent and loses the other 30 percent in silence.

What counts as a good chatbot handover?

A good handover is boring from the customer side. They ask something the bot cannot do, and within one message they are talking to a person who already knows what they asked. Nothing repeats, nothing restarts.

In practice that means five things are true:

• The bot escalates on the second failed attempt, not the fifth. Three rounds of sorry, I did not understand that is where customers stop being customers.
• The full chat history travels with the handover, so your staff member does not open with can you tell me your order number again.
• Outside office hours the bot says so honestly, with a real time: we reply from 9am Monday. Not we will get back to you shortly.
• There is always a human escape hatch the customer can trigger themselves, on request, without guessing a magic word.
• Someone is actually assigned to the inbox. Most broken handovers are not a software fault at all. The bot passed the chat correctly and nobody was watching.

Which numbers prove the bot is paying for itself?

Budget season is the reason to do this arithmetic now rather than in February. The defensible figure is cost per resolved chat, and you need it for both paths.

Take a customer service staff member on S$3,600 a month. Add employer CPF and you are near S$4,200 fully loaded. Across roughly 168 working hours that is about S$25 an hour, so a five-minute chat costs you about S$2 of staff time. If your team handles 1,000 chats in December, that is around S$2,000 of labour, before any overtime for the Black Friday weekend.

Now the bot path. Add your platform subscription, your messaging fees, and the hours you spend maintaining answers. If the platform costs S$200 a month and the bot fully resolves 600 of those 1,000 chats, the subscription works out near S$0.33 per resolved chat, plus per-conversation messaging charges on WhatsApp, plus setup time amortised over the year. Check your provider rate card for the current per-message or per-conversation price before you quote a number internally, because those rates have changed more than once.

The honest comparison is not S$0.33 against S$2. It is the bot cost plus the cost of the chats it mishandles. One lost order at S$180 wipes out the savings on 90 resolved chats. That is why the drop-off number belongs in the same table as the cost per chat. Without it, your automation case looks great on paper and loses money in the inbox.

Why does the festive peak break a bot that worked all year?

Because the questions change, not just the volume. From Black Friday on 27 November through 12.12 and into the Chinese New Year run-up, customers stop asking about product specifications and start asking about time and money: will it arrive before 25 December, why did my discount code fail, can I return the second one, has my refund gone through.

Those are exactly the questions a bot cannot safely improvise, and they arrive in the same hours your staff are at their thinnest. Volume also changes the maths of a mediocre handover. At 30 chats a day, a 15-minute wait is survivable. At 300 chats a day it becomes a queue, and a queue at 11pm on a Friday is just an abandoned cart with extra steps.

How do you test the handover before 27 November?

Run it as a customer, from a phone that does not belong to anyone on your team. Writing a list of 20 real questions takes an hour. Use your actual last 100 chats to build the list, weighted towards the festive ones.

Score each test on four points: did the bot answer correctly, did it escalate when it could not, how many minutes until a person replied, and did the customer have to repeat anything. Run the set three times: mid-morning on a weekday, 9pm on a weekday, and Sunday afternoon.

Reasonable pass marks for a small team: correct or cleanly escalated on at least 18 of 20, a reply from a person inside 15 minutes during office hours, an honest next-reply time outside them, and zero cases where the customer repeats information. Anything below that, you have four weeks to fix it, which is enough.

What should you fix first if it fails?

In this order, because this is the order that protects revenue:

1. The escalation trigger. Drop it to two failed attempts and add an always-available request a person option.
2. After-hours honesty. Replace the vague holding message with a stated next-reply time, and set it to the truth including weekends.
3. Context transfer. If history does not carry across, this is usually a configuration setting rather than a rebuild.
4. The top ten festive questions. Delivery cutoff dates, return window, refund timing, stock, discount code rules. Write these answers yourself and check them against your actual policy.
5. Logging. If you cannot report handover rate and wait time next month, you cannot defend or cancel the spend in January.

Who covers the handover over Chinese New Year?

Chinese New Year falls on 6 February 2027, and the quiet failure is a bot that keeps promising a reply while the office is shut and suppliers across the region are closed for a fortnight. Decide now who is named on the rota for each day of that window, set the bot to state a real return date rather than shortly, and make sure refund and delivery answers reflect the February lead times rather than the December ones. A handover with nobody on the other end is worse than no bot at all, because the customer waited first.

Frequently asked questions

What is a normal chatbot handover rate for a Singapore SME?

For most SME support inboxes, 30 to 50 percent of chats reaching a person is healthy, and a very low rate is usually a warning rather than a win. It often means the bot is answering questions it should not be answering, or customers cannot find the way out. Judge the rate alongside your drop-off and your refund complaints, never on its own.

Should we turn the chatbot off during Black Friday?

Rarely. The better move is to narrow it. Keep the bot on the high-volume factual questions such as order status, delivery cutoffs and store hours, and route anything touching money, refunds or complaints straight to a person. A narrow bot that escalates fast beats a broad one that argues with customers.

How do we show the chatbot was worth the money?

Put four figures on one page for the budget discussion: chats handled, share fully resolved without a person, cost per resolved chat including messaging fees, and the number of chats abandoned after handover. The first three justify the spend. The fourth keeps you honest about what it is costing you elsewhere.

Doing this in the next four weeks matters more than doing it perfectly. The test takes an afternoon, the top five fixes take a day, and both get much harder once the orders are landing.