The gap between "chatbot in theory" and "chatbot that actually works in our industry" is significant. Here are real use cases, by vertical, that deliver measurable results — and the failure modes to avoid.
E-commerce
E-commerce is the highest-ROI environment for chatbots, for three reasons: high contact volume, predictable question types, and customers who already expect digital self-service.
What works well:
- Order tracking: "Where is my order?" is the single most common e-commerce support query. A chatbot that integrates with your OMS and returns accurate status in two messages deflects 30–50% of all contacts.
- Returns and refunds: Automated returns initiation — collect order number, confirm item, generate return label — reduces agent time and customer wait.
- Product recommendation: A chatbot trained on your catalogue can suggest products based on customer input ("I need a gift for a 7-year-old who likes science") more effectively than a search bar.
- Abandoned cart recovery: Proactive chatbot engagement for users who add to cart and don't check out — with a discount trigger — converts 5–15% of recoverable sessions in well-implemented setups.
What doesn't work: Complex complaints about damaged goods, nuanced refund disputes, and any conversation that requires reading tone. These still need a human.
SaaS and software
SaaS support chatbots work well for onboarding, troubleshooting common errors, and routing bug reports — but only when trained on comprehensive documentation.
- Onboarding flows: Guided first-use walkthroughs triggered when a user reaches a feature for the first time. More effective than tooltips because they're conversational.
- Error disambiguation: A chatbot that can identify which error code a user is describing and point to the correct resolution article deflects significant support volume.
- Upgrade triggers: A chatbot that detects a usage limit approaching and proactively offers an upgrade path performs better than a static upgrade button.
Healthcare (non-clinical)
Healthcare chatbots work well in administrative contexts and badly in clinical ones.
- Appointment booking: "Book a GP appointment for next week" is a well-defined, automatable task. Healthcare providers using chatbots for appointment management see 20–40% of bookings completed without staff involvement.
- Symptom triage (limited): Pre-consultation intake — collect symptoms, duration, severity — structures the information before a clinician sees it. This speeds up consultations, not replaces them.
- Post-discharge follow-up: Automated check-in messages with structured response capture reduce readmission for certain chronic conditions.
Hard limits: Any chatbot providing diagnostic advice or treatment recommendations creates regulatory and liability risk. Even the best current LLMs are not safe for clinical decision support without human oversight at every step.
Legal services
Legal chatbots are most useful for intake — qualifying a prospective client before attorney time is involved — and for document Q&A within clearly scoped areas.
- Client intake: Gather case type, jurisdiction, timeline, and key facts before the first consultation. Law firms using intake chatbots report 40% shorter initial consultation times.
- Document Q&A: A chatbot trained on a specific contract or legal agreement ("What does clause 12 mean?") is genuinely useful for clients reviewing documents.
- Statute navigation: Large firms use internal chatbots to help lawyers quickly find relevant precedent within curated internal knowledge bases.
What to avoid: Consumer-facing chatbots giving legal advice on complex personal situations. The risk of incorrect advice creating client harm — and liability — is high.
Financial services
- Account enquiries: Balance checks, recent transactions, and statement requests are high-volume, low-complexity, and well-suited to automation.
- Fraud alerts: Automated proactive outreach ("We noticed unusual activity. Was this you?") reduces fraud losses when customers can confirm or deny in seconds.
- Mortgage pre-qualification: Initial income, expenditure, and credit profile collection before an advisor call. This converts prospects who would otherwise bounce at the effort barrier.
Real estate and property
- Property enquiries: "What is the asking price of this property?", "Is this still available?", "Can I book a viewing?" — all automatable.
- Tenant management: Maintenance request logging, rent payment reminders, and lease renewal prompts via chatbot reduce property manager workload.
For all of these use cases, the build-vs-buy decision is critical. Read our guide on building vs buying a chatbot to understand which approach fits your industry and scale, and use our Chatbot ROI Calculator to model the business case before investing.