chatbots
By Saurav | Founder of saavos | Building in public toward $10k MRR
[!TLDR] Live chat puts a real human on every conversation — available business hours, $5–$25 per interaction, unmatched on hard cases. AI chatbots handle the 40–70% of inbound that's repeat questions, 24/7, at fractions of a cent per reply, but fail on judgment and emotional calls. Neither replaces the other. Five hybrid patterns show where to draw the line.
A live chat tool (Intercom, Tidio, Crisp, LiveChat) puts a real person on the other end of the chat box. They're online during your support hours, they handle one conversation at a time (or two if they're fast), and they cost you their hourly wage plus the seat license.
An AI chatbot (saavos, Chatbase, Botsonic) trains on your content and replies 24/7 without a human in the loop. The marginal cost per reply is fractions of a cent. It handles unlimited concurrent conversations. It does not get tired, does not call in sick, and does not need to be onboarded to a new product launch — it re-indexes the launch page automatically.
The honest framing is not "which one is better" but "which one is better for which conversations." Live chat is great at the conversations a chatbot loses. A chatbot is great at the conversations a human shouldn't be wasting time on.
| Dimension | Live chat | AI chatbot |
|---|---|---|
| Available hours | Business hours | 24/7 |
| Median first-response time | 1–10 min (during hours) | 1–3 sec |
| Median first-response time (off hours) | Hours, often next day | 1–3 sec |
| Cost per conversation | $5–$25 (human time) | $0.001–$0.05 (LLM cost) |
| Concurrent conversations | 1–2 per agent | Unlimited |
| Setup time | 1 day (account + onboarding) | 5 min – 1 day |
| Monthly cost (small SMB) | $39–$79 per seat | $0–$49 total |
| Handles factual product questions | ✅ With training | ✅ Native |
| Handles refunds / account changes | ✅ | ❌ (route to human) |
| Handles emotional support, churn | ✅ | ❌ |
| Captures leads via real conversation | ✅ | ⚠️ Rules-based only |
| Maintenance overhead | Hiring + scheduling | Source updates + prompt |
The gap that matters most for SMBs: off-hours response. A live chat tool is dark from 6pm to 9am and on weekends. A chatbot answers the same fast at 2am. If half your traffic visits when your team is offline, a chatbot is the only response option that exists at all.
1. High-touch sales for considered purchases. If your product costs $500+ and the buyer is comparing alternatives, the conversation is part of the sales motion. A human who knows the product well closes deals a chatbot will never close, because the visitor is buying confidence as much as features.
2. Account-state changes a chatbot cannot perform. Cancellations, refunds, password resets that require admin override, billing escalations. These require both system access and judgment. A chatbot's correct behavior is to route the conversation to a human, not to attempt the action.
3. Emotional or trust-sensitive moments. A customer who's just been charged twice, a user whose data didn't import correctly, anyone in churn-risk territory. The chatbot can answer the literal question and lose the customer; a human can answer the literal question, apologize, comp the next month, and keep the customer.
If the conversations on your team are mostly these three categories, lean live-chat-heavy. The chatbot still has a role — answering "what's your pricing?" while the rep is finishing another conversation — but it's the supporting cast.
1. Your inbound is mostly repeat questions. Order status, "how do I cancel," "what's in the Pro plan," shipping policy, integration support. Roughly 40–70% of SMB support volume is the same dozen questions on rotation. A trained chatbot deflects these instantly without wasting human time.
2. You don't have a support team yet. Solo founders, two-person teams, side projects. Hiring a live chat agent or staffing a rotation is not a feasible 2026 move at $0–$50k MRR. A chatbot covers the floor at $19–$49/month.
3. Traffic is highly off-hours. International audiences, weekend ecommerce, late-night SaaS evaluators. A chatbot is the only support surface that's actually present when the visitor is.
4. Pre-purchase questions are the bottleneck. "Does it integrate with X?" "What's the difference between Starter and Pro?" "Do you support [our use case]?" A chatbot trained on your docs, integrations index, and pricing page closes the information gap instantly. A human typing "let me check" loses the visitor while they go check.
5. You want a 24/7 first responder, not a 24/7 resolver. The chatbot's job is to give the visitor a real answer to easy questions and a calibrated handoff for hard ones. If your fallback is "I'll route you to Marina, who replies within 4 hours," visitors are happy with that — they're not happy with a generic "no one is online."
If your conversations are mostly in these five categories, the chatbot is doing the heavy lifting and the live chat is the safety net for the cases the bot routes out.
The honest answer for ~80% of SMBs in 2026 is both, layered, not either. Five hybrid patterns that work:
Pattern 1 — Bot first, human on miss. The default. Chatbot handles every inbound; the fallback message offers email or a meeting link. Cheap, simple, scales to any volume. Right when the team is small (1–3 people). The evaluation checklist emphasizes fallback-message configurability for exactly this reason — it's where the hybrid lives or dies.
Pattern 2 — Bot first, live takeover during hours. Same as pattern 1, but during business hours a human can break in and take over the conversation in real time when the bot routes a hard case. Most modern live chat tools (Tidio, Crisp, Intercom) support a "bot → human" handoff with full transcript context.
Pattern 3 — Bot for FAQs, live chat as a separate widget. Two widgets, two surfaces. Bot launcher on docs and pricing pages; live chat launcher only inside the logged-in app. Right when the bot would dilute the live chat brand on high-touch pages.
Pattern 4 — Live chat first during hours, bot off-hours. Same widget, schedule-based behavior. Visitors during 9–6 see "we're online, ask away;" visitors at 11pm see "our team is offline — try our chatbot, or leave a message." Right when live chat conversion is your moat but you don't want to lose off-hours visitors.
Pattern 5 — Bot deflects, human owns escalation tier. The bot handles tier-1 (FAQs, factual questions). When it can't, it routes to a shared inbox or live chat queue with the conversation transcript attached. Right when ticket volume is high enough that you're staffing real support but most tickets are still repetitive.
The wrong pattern: pretending you can replace live chat entirely with a chatbot because a vendor's demo showed 80% deflection. The remaining 20% is where your hardest customer relationships live, and a "sorry, I can't help" reply burns those relationships fast. The reduce-tickets playbook covers fallback design in detail — it's the most important part of any hybrid.
A typical SMB with 200 inbound conversations per month:
Live chat alone (1 agent, 40 hrs/week):
AI chatbot alone:
Hybrid (bot first + human on miss):
The hybrid cuts cost per conversation by ~57% vs live-chat-alone, while still preserving human handling for the cases that need it. The cost gap closes further as volume grows — the bot scales linearly with subscription tier; live chat scales linearly with headcount.
A 2026 product category — Intercom Fin, Tidio Lyro, Drift Conversational AI, Crisp's MagicReply — that bolts AI onto a live chat tool. The pitch: "the AI handles tier-1, your humans handle tier-2, all in one inbox."
The honest framing: these are great if you already use live chat heavily. The AI features are usually a bolt-on rather than a re-architecture, and the per-resolution pricing (~$0.99 per AI-resolved conversation on Intercom Fin, similar on Drift) gets expensive fast at small scale. Below ~500 conversations per month, a standalone chatbot at $19/month bundled into a separate live chat at $39/month is almost always cheaper than an "AI live chat" at $99+/month.
Above ~1,000 conversations per month, the integrated tools start to make sense — single inbox, single seat, single training surface. Run the math at your actual volume before signing anything.
For a deeper criteria walkthrough on the AI side specifically, the 12-question evaluation rubric applies whether the AI is standalone or part of a live chat suite.
A few decisions a chatbot does not let you skip:
If you're an SMB in 2026 deciding between live chat and a chatbot:
Start free on saavos — paste your URL, get a chatbot in 5 minutes, plug it in next to whatever live chat you're already running, no credit card required. See our pricing for what each tier handles in monthly volume.
Get the next post in your inbox
Honest writing on building, embedding, and shipping AI chatbots. No spam. Unsubscribe anytime.
Live chat puts a real human on the other end of the chat — available during business hours, $5–$25 per conversation in human time, unmatched on hard cases. An AI chatbot trains on your content and replies 24/7 at $0.001–$0.05 per conversation, handles unlimited concurrent sessions, and resolves the 40–70% of inbound that is repeat questions. Neither replaces the other; the right move for ~80% of SMBs is both, layered, with the bot as first responder and the human as fallback for cases the bot routes out.
Three scenarios where live chat is the right primary tool: (1) high-touch sales for considered purchases over ~$500, where the conversation is part of the sales motion; (2) account-state changes the chatbot cannot perform — cancellations, refunds, password resets requiring admin override; (3) emotional or trust-sensitive moments — billing disputes, churn-risk customers, data-import failures. If your conversations are mostly these three categories, lean live-chat-heavy and use the chatbot as a supporting cast.
For a typical SMB with 200 conversations per month: live chat alone runs ~$1,539/month (one part-time agent at ~$1,500 + tool seat at $39) for a per-conversation cost of $7.70 with zero off-hours coverage. AI chatbot alone runs $19–$49/month total with 100% off-hours coverage but ~50% of conversations get out-of-scope replies. The hybrid (bot first, human on miss) runs ~$660–$690/month — chatbot at $19–$49 plus live chat seat at $39 plus ~$600 of human time on the 30–40% the bot routes out — at $3.30 per conversation, a 57% cost cut vs live chat alone with full off-hours coverage.
Almost always no. The 20–30% of conversations a chatbot cannot handle is exactly where your hardest customer relationships live — refund requests, churn-risk users, complex pre-sales for high-ticket purchases. Replacing live chat entirely with a chatbot saves money on day one and burns customer relationships over the next quarter. The right move is layering: chatbot handles tier-1 (FAQs, factual questions, off-hours), human handles tier-2 (judgment, empathy, account-state changes). The chatbot does not replace the human; it frees the human to spend time on cases that need them.
Five hybrid patterns work in practice: (1) bot first, human on miss via fallback message — simplest, scales to any volume, right for 1–3 person teams; (2) bot first, live takeover during hours — humans break in to take over when the bot routes a hard case; (3) bot for FAQs, separate live chat in the logged-in app — different surfaces, different jobs; (4) live chat first during business hours, bot off-hours — schedule-based same widget; (5) bot deflects tier-1, human owns escalation tier with transcript handoff. Pick based on team size and where your hard cases come from.
These bolt AI features onto a live chat tool so tier-1 and tier-2 share an inbox. Below ~500 conversations per month, a standalone chatbot at $19/month plus separate live chat at $39/month is almost always cheaper than an integrated "AI live chat" tool at $99+/month — especially the per-resolution pricing models like Intercom Fin at ~$0.99/resolution. Above ~1,000 conversations per month, the integrated tools start to make sense for the single-inbox simplicity. Run the math at your actual volume before signing.
Builds tools for solopreneurs and small SaaS teams who don't have an afternoon to spare.
Paste your URL. Train your bot. Drop one script tag. No credit card.