Most small teams need a help desk (or a well-run shared inbox) first, an AI chatbot second, and staffed live chat only if someone can reply within about two minutes during opening hours. Add the help desk once two or more people answer customers; add a chatbot once repeat questions pass roughly 30 a week.
The three aren't really rivals. A help desk is where conversations are kept and assigned, live chat is a channel humans answer in real time, and an AI chatbot is an answerer that works without anyone present. Most products now sell all three in one bundle, so the practical question is which layer to switch on first, and whether you can staff it.
What each layer does that the other two can't
| Help desk | Live chat | AI chatbot | |
|---|---|---|---|
| Its core job | Collect email, forms and chats into one queue, with owners, statuses and history | Let a visitor talk to a person on your website while they're still there | Answer common questions instantly from your own content, day or night |
| Who answers | Your team, in their own time | Your team, within minutes | The software, handing over when it can't |
| Right when | Two or more people share customer contact, or anything gets lost | Someone is free to watch it during stated hours; buyers decide quickly | Many questions repeat and have written answers |
| Fails when | Nobody triages the queue, or volume is a handful a week | The widget says "online" and nobody replies | Answers aren't written down, or handovers go nowhere |
| Typical small-team cost | Free for tiny teams; roughly $19-$55 per user a month for paid plans | Often free or bundled; the real cost is staff attention | Metered: per resolution, per conversation or per credit |
One consequence is worth spelling out. A chatbot and live chat both sit on your website, but they solve opposite problems. Live chat helps when a person is available and the visitor needs judgement. A chatbot helps when nobody is available and the visitor needs a fact. The help desk is the back office both of them should hand work into.
Five questions that settle it for your team
Answer these with real numbers from the last four weeks, not impressions. Count enquiries across email, forms, website chat and social messages; phone calls matter too, but they're a separate decision.
- How many written customer contacts a week? Under about 15, a shared inbox with labels does the job. From 15 to 60, a help desk starts paying for its tidiness. Above 60, you want a help desk and some automation.
- What share are repeat questions with a written answer? Take 40 recent messages and mark each one that your website or an existing template already answers. Under 25%, a chatbot has little to do. Over 40%, it has a clear job.
- How many people answer customers? One person can run from an inbox. The moment two people reply, you need assignment and history, or you'll get two replies to one customer and none to another.
- Can anyone reply within two minutes during stated hours? If not, don't offer live chat. Offer a chatbot with a form fallback, and say honestly when a human will reply.
- Does a conversation need to be found again months later? Warranty claims, disputes and repeat customers all need searchable history. That's a help-desk job, not a chat widget's.
A decision table for small teams
| Your situation | Start with | Add later |
|---|---|---|
| One person, under 15 written enquiries a week | Shared inbox with labels and saved replies | A chatbot only if the same five questions dominate |
| Two to five people, 15-60 a week, things occasionally missed | Help desk | AI chatbot when repeat questions pass about 30 a week |
| Busy website, quick buying decisions, someone at a desk all day | Live chat inside a help desk | A chatbot for out-of-hours and the first reply |
| Most enquiries arrive evenings and weekends | AI chatbot with a form fallback | Help desk once handovers exceed what one inbox can hold |
| Complex, project-based conversations; few new enquiries | Help desk or shared inbox only | Probably nothing; AI drafting inside the inbox helps more |
Three small firms, three different answers
A two-person painter and decorator
Illustrative numbers: about 12 written enquiries a week, most arriving as a form with a photo of a room, plus the odd message asking whether they do exterior work. The owner quotes in the evening. Repeat questions are few ("do you do wallpaper?", "how far ahead are you booked?"), perhaps three a week.
A help desk would add a login and a monthly bill to a problem that barely exists, and live chat would sit unanswered while both of them are up ladders. The right answer is a shared inbox with three labels (new quote, booked, chasing) and two or three saved replies, plus a short FAQ on the website covering wallpaper, exterior work and lead times. Cost: nothing beyond the email plan. Revisit if enquiries double.
A five-person architect practice
Around eight new enquiries a week, but long email threads with existing clients, consultants and planners. Nearly nothing repeats: every enquiry is about a specific site and brief. Conversations matter years later when a question comes up about a design decision.
A chatbot has little to answer beyond "do you take on small extensions?" and "what are your fees like?", both of which a clear services page handles. What the practice lacks is a reliable way to see who owns each new enquiry and whether it had a reply. A small help desk (or a shared mailbox with assignment) for the enquiries address only, with project correspondence staying in normal email, fixes that. AI drafting inside the inbox, which several help desks include, is more useful to them than any bot.
A ten-person HVAC installer
Roughly 150 written contacts a week across email, website forms and chat: service bookings, breakdown reports, "when is my engineer arriving?", warranty questions and quote requests. When the office manager tagged 40 messages, 22 were questions with an answer that already existed somewhere (service prices, what's covered by the plan, what to check before calling out an engineer). Two people staff the office from 8 to 5.
This firm needs all three layers, in order. First a help desk, so breakdown reports get assigned and nothing sits in someone's personal inbox. Then an AI chatbot trained on the service-plan terms and price list, answering the repeat 55% out of hours and giving the first reply in hours. Live chat last, visible only 8 to 5 when the office can actually respond, with the bot taking over outside those hours. The costs are worked through below.
What each combination costs at list price
Prices below were checked on vendor pricing pages in September 2026. Help desks price per user; chatbots are mostly metered by results or conversations, which makes their cost move with volume.
| Setup | Example tools and list prices | Illustrative month for the HVAC installer (3 office users) |
|---|---|---|
| Free starting point | Help Scout Free (up to 5 users, one inbox); Crisp Free (2 seats, website chat, no AI) | $0, but no AI answers and limited automation |
| Help desk plus pay-per-answer AI | Help Scout Standard at $25 per user a month plus AI Answers at $0.75 per resolution | $75 in seats plus, say, 100 resolutions at $0.75 = $75, so about $150 |
| Flat workspace bundle | Crisp Essentials at $95 a month per workspace, including 10 seats, chatbot and knowledge base, with AI credits Crisp puts at around 450 automated conversations | About $95 until AI use outgrows the included credits |
| Full suite | Zendesk Suite Team at $55 per agent a month billed yearly, AI agents charged per automated resolution; Intercom from $29 per seat with Fin at $0.99 per outcome | Zendesk: $165 in seats plus resolutions. Intercom Essential: $87 in seats plus 100 outcomes at $0.99 = $99, so about $186 |
Two cost traps are common. Per-resolution pricing looks cheap until the bot gets good: a chatbot resolving 400 conversations a month at $0.99 costs $396, which may still be worth it but deserves a line in the budget. And bundles that include "AI credits" can run out mid-month, at which point the bot either stops or starts charging. Check which happens on the plan you pick. For a line-by-line payback sum, see what a website chatbot costs and whether it pays off.
The ways each layer lets a small team down
Live chat that says online when nobody is. The most common failure, and the most visible. A kitchen-and-bathroom showroom switched on a chat widget with "Typically replies in a few minutes" and forgot it on a busy Saturday. Visitors left messages expecting a reply; the first one went out on Monday. That's worse than having no widget, because it made a promise. If you run live chat, set business hours in the tool so the widget changes state automatically, and write the offline message honestly:
We're not at our desks right now. Leave your question and email, and we'll reply by 10am on the next working day. For a heating breakdown, call the number at the top of this page.
A chatbot with nowhere to hand over. The bot works, customers ask things it can't answer, and the handover creates an email to the owner's personal address or a form submission nobody watches. Symptom: complaints that "your chat said someone would get back to me". The bot's handover must create an assigned item in the help desk, or at minimum a labelled message in a shared inbox that someone checks twice a day. The rules for when to pass over are in when a chatbot should hand over to a human.
A help desk bought for three messages a day. It's tidy but adds a login, a monthly bill and a new place to check, and staff quietly go back to answering from their inboxes. Symptom: tickets marked open for weeks that were in fact answered by email. If volume doesn't justify it, a shared inbox with labels is fine.
Bot and humans giving different answers. The chatbot quotes the service price from the website; the office quotes last month's figure from memory. Customers notice. One source of truth for prices and policies, which both the bot and the saved replies draw from, prevents it. Preparing that content is covered in training an AI chatbot on your FAQs, policies and prices.
How one enquiry moves between the layers
It helps to trace two real-looking enquiries through the HVAC installer's finished setup, because the layers only earn their cost when the hand-offs work.
At 7.40pm on a weekday, a customer opens the website chat. The office is closed, so the chatbot answers. An illustrative exchange:
Customer: our air con unit is making a clicking noise and not cooling. we're on the service plan
Bot: Sorry to hear that. I can't diagnose faults, but I can get this to the team
first thing. Two quick checks first: is the unit's circuit breaker on, and has
the remote been set to cool rather than fan? If both are fine, please share
your name, address and service plan number.
Customer: both fine. [name, address, plan number]
Bot: Thanks. I've logged this as a breakdown on your service plan. The office opens
at 8am and breakdowns are called back first. Your plan covers the call-out;
parts are quoted before any work.Behind that reply, the bot created a help-desk ticket tagged "breakdown" and "service plan", marked high priority, with the transcript attached. At 8.02am the office manager assigns it to an engineer. Nobody retyped anything, and the customer was told the truth about timing. What you'd check in that transcript: the bot quoted plan cover correctly from the plan terms, it didn't promise an arrival time, and the safety checks it suggested are ones the firm is happy for customers to do.
At 2.15pm the next day, a different visitor asks about replacing an old system in a four-bedroom house. The office is open and live chat is showing as available, so the question goes straight to a person, who asks three qualifying questions and books a survey visit while the visitor is still on the page. A bot could have answered "yes, we install new systems", but the sale depended on a human asking the right questions quickly. That's the division of labour in one day: the bot takes the repeatable and the out-of-hours, the person takes the judgement, and the help desk keeps both in one record.
A two-week trial that shows which layer you're missing
Before buying anything, run a two-week tally. It costs nothing and replaces opinion with counts. Create a simple sheet with one row per written enquiry and these columns:
Date | Channel (email/form/chat/social) | Arrived in hours? (Y/N)
Repeat question with a written answer? (Y/N) | Who replied | Hours to first reply
Needed a site visit or judgement? (Y/N) | Was it missed or answered twice? (Y/N)A filled-in fortnight for the painter and decorator might read: 24 enquiries, 19 by form, 15 outside working hours, 5 repeat questions, all replied to by the owner, median 14 hours to first reply, 1 missed. That says the gap is response time out of hours, not triage or volume, and the cheapest fix is a clear auto-reply on the form stating when quotes go out, plus the website FAQ. No new tool.
The same fortnight for the HVAC installer might read: 290 enquiries, 118 outside office hours, 160 repeat questions, 9 missed and 6 answered twice. Missed and double answers point at a help desk; the repeat count and out-of-hours share point at a chatbot. Both have clear jobs.
When to add the next layer
Once the first layer runs, these signals tell you the next one has earned its place. They're my working thresholds rather than industry rules; adjust them to your margins.
- Shared inbox to help desk: two or more missed or double-answered messages a fortnight, or a second person starts replying regularly.
- Help desk to AI chatbot: more than 30 repeat questions a week, or more than a third of enquiries arrive when nobody's working.
- Adding live chat: someone is reliably at a desk during set hours, and your sales depend on catching visitors before they leave, as with urgent repairs or quick product decisions.
- Adding AI triage inside the help desk: the queue passes about 100 a week and sorting it takes someone the first hour of every day. AI ticket triage covers tagging and routing once you're there.
A useful rule when choosing the product: pick the help desk first and check that its chatbot and live chat are good enough, rather than picking a chatbot and bolting a help desk on. The help desk is where you'll spend your working day; the bot is a feature you can change later. If you'd like to compare specific products, the tutorial on AI customer support software for small teams goes through them one by one.
Further reads
- Freshdesk vs Zendesk for a Small Support Team Using AI — Two help desks compared once you've chosen that layer.
- How to Measure Whether Your AI Chatbot Is Actually Working — Check a chatbot is earning its place once added.
- AI Agent vs Chatbot vs Automation: Which Does Your Business Need? — A related distinction people mix up with this one.
- AI Receptionist or Chatbot: Which Suits a Cleaning Company? — The phone-versus-chat version of the same decision.
- Best AI Chatbots for Small Business Websites — Chatbot options if that's the layer you need next.
- AI Email Triage for Shared Inboxes: Sales, Support, and Invoices — Run a shared inbox well before buying a help desk.
- Should You Build or Buy an AI Chatbot for Customer Service? — Buy, assemble or build: 12-month costs, the volume where building wins, and a removals firm worked through the decision.
- AI Customer Service for Small Businesses: What to Automate First — Rank your customer messages by volume and risk, then automate in order: sorting, drafted replies, proactive messages, FAQ answers, and only then actions.
- Tidio vs Intercom: Which AI Chat Tool Suits a Small Business? — Tidio includes ten seats and sells AI in conversation blocks; Intercom bills per seat and per resolved outcome. The sums below show when each wins.
- What Is the Cheapest Way to Handle Enquiries Out of Hours? — Five ways to cover evening enquiries, priced at three volumes, with a charity shop's donation questions and the billing small print on 'resolved'.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Help Scout, Zendesk, Freshdesk, Crisp and Intercom pricing pages; Tidio pricing page.