A customer types "your charge on my card is wrong and I need this fixed today." If your chatbot responds with a cheerful link to the billing FAQ, you've made things worse. Some questions want a bot. Some want a person who can actually do something. Knowing the difference is most of the job.
The two failure modes you're balancing
There are two ways to get this wrong, and they pull in opposite directions.
Automate too little and your team drowns in questions a bot could have answered in one second, while customers wait in a queue for "what are your hours?" Automate too much and you trap someone with a real problem behind a wall of canned replies, watching them rephrase the same desperate question five times.
The goal isn't maximum automation. It's putting each question in front of whatever can resolve it fastest. Sometimes that's the bot. Sometimes it's a human. A good setup makes the switch quietly, before the customer has to fight for it.
Questions a chatbot should own
Bots are excellent at the stuff you answer over and over. These share a few traits: the answer lives in your content, it doesn't change based on who's asking, and getting it slightly personalized doesn't matter.
- Hours, location, parking, and holiday closures
- Pricing, plan differences, and what's included
- Order status and tracking, when connected to your store
- Return windows, shipping times, and warranty basics
- "Do you offer X?" and "how does Y work?" product questions
- Simple how-to and setup steps pulled from your docs
These questions are boring, high-volume, and time-sensitive in the sense that people want them answered now, not in four hours. A bot answering instantly is a genuine upgrade over a form and a wait.
Signals it's time for a human
Escalation isn't a failure. It's the feature that keeps automation from backfiring. Watch for these triggers and route the conversation to a person the moment they appear:
- Money in dispute. Wrong charges, refunds gone missing, billing errors. People want accountability, not a link.
- Account access and security. Locked out, suspected fraud, "someone changed my password." Sensitive and often urgent.
- Emotion. Words like "furious," "unacceptable," "I've been trying for days," or plain profanity. A calm human de-escalates; a bot inflames.
- Repetition. The same question rephrased two or three times means retrieval isn't finding the answer. Stop guessing and hand off.
- Anything legal, medical, or safety-related. These are not places to let a model improvise.
- High-value or complex sales. A prospect asking about a custom quote or enterprise terms deserves a real conversation.
A useful rule: if a wrong answer would cost real money, damage trust, or put someone at risk, a human takes it.
A quick reference
| Question type | Let the bot answer | Escalate to a human |
|---|---|---|
| "What time do you close?" | Yes | — |
| "Where's my order?" | Yes, with tracking | If lost or late |
| "I was double-charged" | — | Yes |
| "This is unacceptable" | — | Yes |
| "Can you set up custom pricing?" | Collect details | Yes, for the quote |
| "How do I reset my widget?" | Yes, from docs | If steps fail |
A handoff done right
Here's the difference a smooth handoff makes. Picture Brightleaf Studio, a small design agency running a chatbot on its contact page.
A visitor writes: "I need to cancel a project I paid a deposit on and I'm not sure I'll get my money back." That's money, emotion, and a policy question stacked together. The bot shouldn't wing it.
Instead it replies: "That's something our team should handle directly so we get it right. I'm connecting you now. While I do, can I grab the best email to reach you in case we miss each other?" It captures the email, tags the chat as billing-related, and passes the full transcript to a human. When someone picks it up, they already see the deposit, the project, and the worry. The customer never re-explains a thing.
Compare that to "Sorry, I can't help with cancellations." Same limitation, wildly different experience. The first keeps the relationship. The second ends it.
In SpideyChat you can set these escalation rules and a fallback so that when the bot hits its limit, or a keyword like "refund" shows up, it collects contact details and routes the conversation instead of dead-ending it.
Match the rules to who's actually online
Escalation shouldn't behave the same at 2pm on a Tuesday and 2am on a Sunday. When your team is around, a handoff can mean a live reply within minutes, so the bot can offer it freely. When nobody's working, promising a human "right now" sets up a letdown the moment the customer waits and hears nothing.
Tie the behavior to your hours instead. During staffed times, the bot escalates to live chat and says someone's joining. Off hours, it still recognizes the same triggers, but rather than implying instant help it captures the details and sets an honest expectation: "Our team's offline right now, but I've flagged this and we'll email you first thing tomorrow." Same triggers, different promise, and no broken one.
That small distinction saves you the worst review a support bot can earn, the one where a customer was told help was coming and then sat in silence for nine hours.
Design the escape hatch, not just the answers
Most teams spend all their energy on what the bot says and none on what happens when it can't help. Flip some of that attention.
Make "talk to a human" visible at all times, not buried. Set expectations honestly: if your team replies within a few hours, say so rather than implying someone's standing by. When nobody's available, capture an email and promise a follow-up you'll actually keep. And read your escalated chats weekly. They're a map of where your content has holes and where your bot is overreaching.
One more thing worth building in: let people skip straight to a human if they want to. A small number always will, and forcing them through a bot first only breeds resentment. Giving them the choice costs you almost nothing and buys a lot of goodwill.
Get this balance right and the bot stops feeling like a barrier. It becomes the fast lane for simple questions and a smart receptionist for hard ones, quietly deciding which is which so your team spends its hours where they count.