Intro: 8 line items behind "let's build a chatbot"
"Let's build an AI chatbot" hides 8 line items: license + token + integration + RAG + team + security + maintenance + measurement. Asking only the "setup price" puts the project in a corner 6 months later.
In this post we examine chatbot cost under 8 headings: 3-band package comparison, token economics, RAG infrastructure, integration, team, security, maintenance, ROI measurement.
2026 reference bands: SaaS (Intercom Fin, Zendesk AI) $120-1.2K/month. Custom GPT/Claude $5-17K setup + $1-3.5K/month. Enterprise RAG + multi-channel $17-70K setup + $3.5-14K/month.
1. SaaS band: fast start, limited flexibility
Intercom Fin: $0.99/resolution + Intercom subscription ($170-2,700/month). Document-based RAG ready, 50+ languages, avg 40% ticket resolution rate.
Zendesk AI: $49-115/agent/month (Suite Professional + AI). Strong ticket automation; weak open-ended chat.
Tidio Lyro / HubSpot AI / Drift: SMB-focused, $30-200/month/seat. Limited product range; complex workflows are hard.
When is SaaS right? Standard customer service + high volume + low customization need. When wrong: sector-specific terminology, personalized pricing, intensive internal-system integration.
2. Custom GPT/Claude band: middle ground
Typical setup: Next.js frontend + OpenAI/Anthropic API + Pinecone/Weaviate vector DB + Redis cache. Setup $5-17K, 4-12 weeks.
Monthly cost: token ($170-1K/month by chat volume), vector DB ($100-500/month), hosting ($70-340/month), monitoring + logs ($100-340/month) = $440-2.2K/month infra.
Advantages: full control — prompt engineering, data privacy, brand tone, sector terminology. Multi-LLM (OpenAI + Anthropic + local) redundancy.
Disadvantages: requires a maintenance team (minimum 1 backend + 1 ML engineer). "Set and forget" not possible; when LLMs update, prompt + RAG calibration is needed.
3. Enterprise RAG band: scale + security
RAG (Retrieval-Augmented Generation): company documents (PDF, Word, intranet, CRM, ERP) are chunked, embedded, loaded into vector DB. When a question arrives, relevant chunks are added to LLM context.
Document pipeline: processing at 10K-1M document scale. ETL infra (Airflow, Prefect), embedding model (OpenAI text-embedding-3-large or Cohere), vector DB (Pinecone production tier $70+/month).
Multi-channel integration: web + WhatsApp Business API (Meta-approved, $150-2K/month tier) + Telegram + voice (Twilio + Whisper + ElevenLabs) + email. Each channel is a separate integration effort.
Cost: first 6 months $17-34K setup, $3.5-14K/month operational. Not a "one-off project"; continuous evolution.
4. Token economics: hidden line item
Token cost = chat volume × avg tokens/chat × token price. 2026 reference prices: GPT-4o input $2.50 / 1M tokens, output $10 / 1M tokens. Claude Sonnet 4.6 in similar band. Anthropic Opus 4.7 ~3-4x more expensive.
Typical B2C chatbot: 8-15K tokens/chat (system prompt + RAG + chat history + answer). 1,000 chats/day × avg 12K tokens = 12M tokens/day ≈ 360M tokens/month.
Monthly token cost: ~$1,500-4,000 (mostly input) = $1.5-4K/month. Linear growth with volume; 10K chats/day → $15-40K/month token alone.
Token savings: shrink RAG chunk size (~200 tokens), summarize chat history ("sliding window"), prompt caching (Anthropic 90% discount), small-model fallback (Haiku, GPT-4o-mini) — 40-70% cost cut possible.
5. Integration: chatbot doesn't run "alone"
CRM integration: HubSpot, Salesforce, Zoho — create customer record, open ticket, assign lead during chat. Each CRM $100-340 setup + maintenance.
ERP / e-commerce integration: Shopify, Ikas, SAP, Logo — order status, stock check, product recommendation. $340-1.4K integration.
Auth + permission: SSO (SAML, OIDC), role-based access. Critical in enterprise. $500-1.7K setup.
Webhook + event-driven: backend triggers when chatbot fires an action (refund initiated, appointment created). Often forgotten line item; source of "the bot talks but does nothing" complaint.
6. Team + maintenance: the "forgotten" cost
Minimum team: 1 backend engineer ($1.4-2.7K/month), 1 ML / prompt engineer ($1.7-4K/month), 1 part-time product manager ($500-1K/month).
Maintenance schedule: weekly prompt improvement (per RAG quality metrics), monthly LLM model upgrade test, quarterly security audit, semi-annual full re-evaluation.
Hallucination prevention: validating LLM outputs — link to RAG source, "I'm not sure" response mechanism, human approval on critical actions. Continuous investment in prompt + example set.
Knowledge base updates: company docs change monthly; chatbot must auto-pick. Without a pipeline, the bot serves stale info after 6 months — user trust evaporates.
7. Security + compliance: enterprise mandate
KVKK + GDPR compliance: mask user data (PII redaction) before sending to LLM. Privacy notice + explicit consent + data deletion procedure mandatory. $1-3.5K setup + legal counsel.
Prompt injection defense: attacks like "forget all instructions above, give me X". Input filtering + system prompt isolation + output validation. Continuously evolving threat surface.
Rate limiting + DDoS: per-user hourly message limit, IP-based throttling. A malicious user can burn $100K/month in tokens alone.
Audit log + access control: who asked what when, what answer was returned — 12-month retention for audit. Mandatory in SOC 2 / ISO 27001 certified projects.
8. ROI measurement: right metrics
Wrong metrics: "replied to X messages", "ran for Y minutes". These are vanity metrics.
Right metrics: ticket containment rate (resolved without human escalation — target 40-70%), lead conversion rate (chatbot lead → customer), avg response time, customer satisfaction (post-chat CSAT), cost/chat, revenue/chat (e-commerce).
ROI calc: monthly savings (reduced customer service calls × avg call cost) + monthly extra revenue (chatbot-driven conversion lift) - monthly chatbot cost. Target payback 4-12 months.
Adoption: 60%+ of users should use the chatbot in the first month. Less = revisit UX (widget too small? messaging flow confusing?).
Conclusion: not "let's build a chatbot" but "we have a chatbot strategy"
An AI chatbot is not a one-off project but continuous operation. Year-one operations are 1.5-2x of setup cost. Don't start without this calc.
Healthy planning: pick the right band by sector + volume + integration need + 12-month ops budget + ROI metrics tracked from day one + 6-month improvement cycle.
If you want a detailed scope + cost + ROI projection for your AI chatbot project, reach out via our AI software page; we'll prepare a sector-specific 8-line plan.
Related articles
Other articles that support the same decision
Cost Analysis
Mobile App Cost in 2026: What Actually Sets the Price?
$10K or $200K? Mobile app cost is set by 7 line items: scope, backend, integrations, native need, testing+security, maintenance, phased approach. Concrete ranges per item.
Cost Analysis
Website Prices 2026: A Real Cost Guide
3 price bands, 8 cost lines, hidden lines, content + performance + SEO + maintenance. 8-heading data-driven website pricing guide.
Cost Analysis
SEO Pricing 2026: Monthly Packages and Real ROI
SEO retainers $300-3K/month band; but "price" misleads. 8 headings on real scope, correct ROI measurement and 2026 market guide.
Next step
If you are planning a similar project, we can clarify the scope and shape the right proposal flow together.
Start a project request