Chatbot vs Live Chat: Which One Your Site Actually Needs
Updated
The choice is not bot or human. It is which questions deserve which, and most sites get the split backwards.
Both tools solve a real problem - a visitor with a question and no patience for a contact form. They fail in opposite directions. Live chat is excellent when someone is there and useless at 11pm. A bot is always available and infuriating when it cannot answer.
Choose a bot when
- Your questions repeat. If eight intents cover most of your volume, a grounded bot handles them faster than a person could type.
- Your traffic is global or out of hours, so live coverage would mean hiring for demand you cannot yet justify.
- The answers are factual: pricing, availability, specifications, order status, opening hours.
Choose live chat when
- Deal values are high enough that a human conversation pays for itself many times over.
- The questions are consultative - someone is deciding what they need, not confirming a known fact.
- Volume is genuinely low. Below a few conversations a day, automation is a project without a payback.
The hybrid that works
Bot answers first, always disclosed, with a visible 'talk to a human' control that never disappears. During business hours it hands over live; outside them it books a callback with a concrete time. The bot's transcript travels with the handover, so nobody repeats themselves. This structure captures the availability of automation and the judgement of a person, and it is barely more work than either alone.
- 8 intents that usually cover most volume
- <60s response expectation on web chat
- 1 click to reach a human, always visible
What each one actually costs
The pricing pages compare badly because the two tools spend money in different places. A bot costs a build, then very little per conversation. Live chat costs almost nothing to install, then a salary.
- Bot: the build, which is mostly the work of collecting real questions and writing grounded answers, plus a small per-conversation model cost. It does not grow with your opening hours.
- Live chat: seat licences, which are the visible part, and staffed hours, which are the real part. One agent handles two or three concurrent conversations before quality drops.
- The cost nobody counts on the live side is interruption. A team answering chat between other work does both worse, and the switching cost lands on the tasks nobody is measuring.
- The cost nobody counts on the bot side is maintenance. Answers go stale, and an unmaintained bot confidently quotes last year's pricing.
Where to put it, and where not to
- On pages where questions block a decision: pricing, delivery, specifications, booking. That is where an answer converts.
- Not as an interruption three seconds after arrival. A popup that talks before the visitor has read anything is a bounce generator, and on mobile it covers the content they came for.
- Trigger proactively only on intent: time on a pricing page, a second visit, a scroll to the terms. Those are questions waiting to be asked.
- Keep it out of the checkout, unless it can actually resolve payment problems. Distraction at the moment of purchase costs more than the answer is worth.
Writing a bot that does not trap people
- Build the answers from real transcripts, not from your FAQ page, and ground them in documents you actually maintain. The FAQ is what you wish people asked.
- Answer first, then offer the next step. Leading with a menu tells the visitor they are talking to a machine that has not listened.
- Two low-confidence turns is the limit. After that, hand over or take a message with a promised time, and mean the time.
- Keep the human control visible at all times, not hidden behind a fallback. Visitors who can see the exit use it far less than the ones who cannot.
- Say what it cannot do, early. A bot that admits it does not handle refunds keeps its credibility on everything else.
The measurements that keep it honest
Resolution rate on its own is a trap, because the fastest way to raise it is to make leaving difficult. Pair it with satisfaction on resolved conversations, and watch three numbers underneath: how long it takes to reach a person when someone asks, how many conversations end without a reply from either side, and how many chats turn into an enquiry your sales team recognises. A bot that resolves less but escalates cleanly beats one that resolves more and irritates everyone it could not help.
Nights, weekends and the promise you make
Most of the damage a chat widget does happens outside working hours, when it looks exactly as available as it does at midday. State the hours in the widget itself, and make the out-of-hours path concrete: a message taken, a specific time for the reply, and a confirmation the visitor can see. Vague reassurance costs you the enquiry twice, once when nobody answers and again when the reply arrives after the decision was made. If you cannot promise a real time, promise a slower one you will keep. The businesses that win on chat are not the ones with the longest coverage, they are the ones whose widget never says anything that turns out to be untrue.
Frequently asked questions
Do chat widgets hurt page speed?
Most do, meaningfully - they are frequently the largest third-party script on a marketing site. Load the widget after first interaction or on scroll rather than immediately, and measure the difference before and after.
What conversion lift should I expect from chat?
Highly variable, and often overstated by vendors. The reliable gain is on complex or high-value products where questions block the purchase. On simple, well-explained offers, chat frequently changes very little.
Should the bot answer from our documents?
If your questions are factual and written down, yes. Grounding the answers in your own pages, policies and specifications is the difference between a bot that helps and one that invents a delivery time. The requirement is that the source material is current and singular. Two contradictory versions of a policy in the corpus produce an assistant that picks one, and the one it picks will be the one you did not want quoted.
How many people do we need to staff live chat?
Count concurrent conversations, not daily volume. One person handles two or three at once while doing nothing else, so the question is how many arrive in your busiest hour rather than how many arrive in a day. Below a handful of conversations a day, staff it informally and skip the software. Above that, either commit to real coverage during published hours or let a bot take the first turn, because a live chat widget that nobody answers is worse than no widget at all.
More on this topic: Artificial Intelligence.
Keep reading
Want this built for your business? See what we do.