We’re bringing you 5 more common failures in chatbot and voicebot development we see among our customers and partners, and how to handle them.
Failure number 6️⃣
❌ No way to connect to a live agent
We can automate up to 70-80% of customer queries thanks to Feedyou chatbots and voicebots. For the remaining 20-30% we should be always able to handover the conversation to a live agent.
When to handover:
- Virtual assistant doesn’t understand
- Virtual assistant doesn’t have the knowledge to answer correctly
- Customer asks for it
- Customer isn’t satisfied with the answer
How to handover:
- In real-time (smart routing based on context we have)
- Fallback via email / phone callback / JIRA ticket if there is no live agent available
- Providing customer with contact details / opening hours
- Virtual assistant takes back the conversation if a live agent doesn’t connect
Failure number 7️⃣
❌ No fallback when no live agents are available
Your chatbot and voicebot solutions should deal with:
- Business / customer support opening hours
- Live agent availability
- Queues and peak sessions
Tips to improve:
- Connect to a live agent during opening hours only
- Disallow the live agent handover if no live agent is available
- Switch to a Smart IVR / voicebot to solve, or at least qualify, the customer need
- 1-minute waiting room – offer fallback to asynchronous channels (email, phone callback after some time once you’re available / JIRA ticket / SMS / WhatsApp etc.)
- 2-minute waiting room – fallback to asynchronous channels by default
Failure number 8️⃣
❌ Connecting wrong channels
- Chatbot on a website with no traffic
- Voicebot for handling just a few calls a day
- Connecting other channels just because it’s trendy or your competitors have it
Analyze your customer support touch points and start with those channels where you have the most traffic. And remember, all channels should be connected into one seamless omnichannel experience.
❌ Connecting no internal systems
Only about 30% of the questions are of a general nature. A significant part of them are specific requests, for example, about the status of your bank account, making a payment, finding specific information in the knowledge base, asking for example where your shipment is, whether you can change the delivery date and time, etc.
To solve such requests, integration with your internal systems are crucial, i.e.:
- ATS / HR systems
- CRM / ERP systems
- Knowledge databases / Google Sheets / Office 365
- Omnichannel / live chat platforms
- Telephony networks
- Ticketing tools
- Analytical platforms
- 3rd party NLP/NLU providers
- Generative AI / LLM providers
- Integration platforms such as Make, Twilio, Zappier, Superface etc.
Failure number 9️⃣
❌ Missing small talk
Your chatbot and voicebot should always answer:
- Greetings (Hi, Hello, Good morning, Goodbye, Ciaoo, etc.)
- Confirmations (Thanks, Thank you, Nothing else, etc.)
- I don’t understand
- I need help
- I want to talk to a live agent
- Supported languages
People often test chatbots and voicebots:
- Commands like “Test”, “Start over”, etc.
- Gibberish – expressions like “kjefajerfsd”
- What’s the weather in …
- Do you love me?
- Tell me a joke
- Are you a chatbot or voicebot
- What is your name
If you’re able to response to these general questions, you will:
- Make the conversation more natural
- Look smarter than 99% of solutions on the market
- Deal with 99% of teases (it’s a matter of just a few phrases)
Failure number 🔟
❌ Design it, launch it and leave it
There should be a product owner dealing with further training:
- Optimizing the conversational flow based on data
- Adding answers to new questions
- Deleting pointless questions
- Detecting problems
- Updating information
- Reporting outcomes
- Evaluating customer experience (NPS)
- Evaluating business impact (ROI)
2 hours a week is the minimum to keep your chatbot or voicebot in shape.
Therefore our Feedyou Platform and all virtual assistant solutions are provided as Software as a Service (SaaS). Not only a technology license, but also dedicated support from our team of Conversational Designers, Bot Developers and Account Managers are always provided hand in hand.
Extra failure at the very end
❌ No Generative AI involved
No Large Language Model (LLM) such as GPT involved. You have to keep training
everything tediously from scratch.