Chatbots are HOT. Gartner predicts chatbots will power 85% of customer service interactions by next year, and 31% of marketers believe that virtual personal assistants will have more impact than any other AI technology.
Analyzing over 20 years of live chat transcripts, LivePerson found 69% of chats and calls could be handled quickly by bots, and Nielsen found 56% of online shoppers say they prefer to resolve issues through messaging apps than call customer service.
If you want to join the 80% of executives that are using or planning to use chatbots by 2020, read on to learn what to look for in an ecommerce chatbot solution.
A key differentiator between chatbot tools is whether they’re rules-based scripting or leverage learning algorithms and natural language processing.
Chatbots are programmed with conversation trees, a mapping of “bot says / user says” dialogs. Responses can be closed (predetermined) or open (type a question/response), or a mix of both.
Image: Chatteron.io Demo
Scripted scenarios are great for common customer support inquiries that can easily be served by quick answers or links to shipping and return policies, or similar content. Of course, this limitation may leave customers unsatisfied. Some rules-based chatbot tools allow you to transfer a user to a live agent if their questions can’t be resolved by the bot.
More nuanced chat requires natural language processing to infer intent and to map open responses to bot replies. A chatbot using NLP (natural language processing) looks at a user’s utterance, parse out entities to infer intent. It then matches intent to predetermined intents you create, such as “showGifts.”
Utterance is what the user says. For example, “I’m looking for gift ideas for a 3 year old boy.”
Intent is what it sounds! It’s what your user most likely wants to know or do.
Entities are data buckets that include keywords and phrases with similar characteristics that modify user intent. For example, “gift ideas,” “gift suggestions,” “gift recommendations,” “gifts,” “best gifts for,” “presents for,” “present recommendations.” And “3 year old boy,” “three year old boy” “3 yr old,” “toddler boy,” et cetera.
Natural language programming can match intent to pre-built responses more flexibly than rules-based bots. But NLP without a machine learning component falls short of truly being “AI-driven.” AI chatbots will recognize patterns and optimize itself based on user interaction and feedback, but often require human training to fine-tune their algorithms.
* Chatbot solutions that use NLP but don’t include training capabilities are not considered AI chatbots in this guide.
Whether you choose a rules-based, NLP-enabled or full-on AI chatbot solution, be prepared to invest time in mapping out and scripting conversation trees, building sets of entities and defining intents. Many chatbots offer pre-built templates to guide your conversation, but still require contextual input from you.
Also, keep in mind good AI requires more training before release than people think. Natural language processing doesn’t just “know everything” out of the box.
Minimum viable chatbot
If your goal is simply to make live help available 24/7, relieve chat agents of the 70% of questions that can be handled by a chatbot, or configure light product recommendation dialogs, you’ll be covered by any business-user friendly, self-service tool. Most support Facebook Messenger and/or your own site widget, and many use natural language processing to match responses to intent.
Facebook Messenger marketing
If chat is one part of a larger Facebook marketing strategy, including cart recovery, automated sequencing (remarketing) and ads, look for platforms that include these tools.
If you see conversational commerce as a key initiative, you want more flexibility to train your AI, deploy to multiple messaging platforms, devices and touchpoints, and own your code. Look for a build framework that jives with your platform, cloud services and IT’s preferred programming languages.
You may opt to host your own chatbot, or leverage third-party messenger applications such as Facebook Messenger, WhatsApp, WeChat, Telegram, KiK or Skype. You may also want to jump into voice commerce, integrating with Alexa or Google Assistant.
Why use Facebook Messenger chatbots
First-movers like 1-800-Flowers, ASOS, Sephora and Nike have all embraced Facebook Messenger, thanks to its high user adoption rates and in-chat purchase capabilities (63% of shoppers are willing to buy through SMS). Messenger chatbots support natvie payment through Facebook Pay (US only) or Stripe.
An added benefit — whether a visitor engages with your Facebook chatbot or not, they can be retargeted through Messenger with offer codes, cart recovery messages, back-in-stock alerts and other notifications (with up to 80% open rate). Some Facebook chatbot vendors described below allow you to build and manage segmented “automated sequencing” campaigns in addition to building chatbot scripts.
Why use your own chatbot
To ensure all-inclusivity, consider using your own chatbot. You can use multiple bots, but keep in mind the impact on performance when running multiple scripts. You may choose to leverage Facebook Pixels for Messenger retargeting without using a Messenger bot as your on-site concierge.
Why explore voice commerce
While only 2% of Alexa owners made a voice purchase in 2018, voice is a touchpoint within a broader shopping journey. 47% of smart speaker owners (around 25 million in the US) use voice assistants for product search and research, and 43% use them to make shopping lists.
If you want to get in early on this action (like Walmart’s Voice Order), consider building Alexa Skills or Google Assistant applications.
Regardless if you build-your-bot or buy-your-bot, you should know your must-have features before you go bot-shopping. Make note of any of the following that apply to your project:
Many tools offer goodies above and beyond the above features. We’ve reviewed 14 of the top chatbot solutions for ecommerce to help you build your short list.
One of the most popular options, Chatfuel claims 46% of Facebook Messenger chatbots run on its platform, including Levi’s, Adidas and T-Mobile. Free for up to 500,000 monthly active users, Chatfuel requires no coding skills and has inbuilt NLP (via Facebook), analytics and supports automated sequencing campaigns.
Best for: Merchants that want a mostly free Facebook Messenger chatbot integrated with remarketing
MobileMonkey is positioned as a “Facebook Messenger Marketing Platform” — a suite of tools within which Messenger chat is only one capability. As such, it only supports Facebook chatbots and enables automated sequencing and retargeting. MobileMonkey has an agency partner network that can build and manage your ads and bots for you if you prefer to outsource.
MobileMonkey has a drag-and-drop dialog builder, premade dialog templates and supports live chat handoff through its Zapier integration.
Best for: Merchants who want to build and manage Facebook Messenger ads in addition to a chatbot (and are willing to pay for these additional capabilities), and/or with to leverage an agency to build and run Messenger campaigns
Octane.ai is Facebook Messenger bots and remarketing for Shopify merchants. It supports unique third-party integrations such as Yotpo and Klavio to add review context to product recommendations and targeted offers (including direct link to read reviews).
Best for: Shopify merchants that want to leverage Facebook retargeting
Chatteron allows you to build chatbot logic (no coding required) and publish it to Facebook Messenger, Workplace (Facebook’s work collaboration tool) or your own chatbot. Unlike other Facebook Messenger solutions, Chatteron lacks remarketing features and some of the handy drag-and-drop functionality of other builders. However, it comes with over 20 pre-built templates with intent, entities and conversation flow, and leverages machine learning. Chatteron allows you to try-before-you-buy with its live demo.
Best for: Merchants that want to build and host their own AI-driven chatbot without coding skills
ManyChat supports Facebook Messenger chatbots and retargeting. It provides similar capabilities to its competitors, but supports file attachments, allows you to preview recipients’ names in a group blast and provides analytics for subscribes and unsubscribes. ManyChat also integrates with Google’s Dialogflow framework (reviewed below) for AI and voice capabilities.
Best for: Merchants who want a Facebook Messenger chatbot with simple remarketing capabilities and live chat takeover, or who want to add Messenger chatbots to a Dialogflow project.
Botsify lets you build dialog for Facebook, your own chatbot or Alexa. Its voice features differentiate it from competitors, however Botsify lacks Facebook remarketing and campaign tools, and appears to focus more on customer service dialogs than commerce. Botsify supports team collaboration and workflows, enterprise Single Sign On and prototype preview across devices.
Best for: Merchants who want to support Messenger or their own chatbot and explore voice for customer service, but can live without merchandising and retargeting features
Ada is unique among the “buy bots” in that it’s a no-IT-required chatbot builder that’s not focused on Facebook Messenger (you can publish to Messenger, with WeChat and WhatsApp coming soon). Ada’s sweet spot is supporting use cases that require collection of personal information within chat, such as secure customer data (with PCI, GDPR and PIPEDA compliance), and can provide replies based on customers’ orders, bookings, accounts, passwords and more. You can also segment and target visitors with proactive chat based on visitor attributes or on-site behavior.
Best for: Telecom, travel, B2B and industries where you might need to collect personal information within chat, provide service related to accounts, plans and bookings, and leverage proactive chat.
Dialogflow (formerly API.ai) is a Google product that leverages its speech-to-text, natural language processing and machine learning technologies. Running on Google’s Cloud Platform, Dialogflow projects can scale to hundreds of millions of users, and integrate with Google Assistant, Alexa, Facebook Messenger, Twitter, Telegram, KiK, Viber, Skype, Slack and more. It also supports integration with Internet of Things such as wearables, car speakers and other smart devices.
While users say publishing to Facebook is difficult with Dialogflow (you have to build this yourself), integrations with ManyChat and Botsociety (below) bridge this gap.
Dialogflow lacks pre-built integrations with live help, email, reviews and other third party services, but can be custom built. Developers can use Dialogflow by making direct REST over HTTP requests, and use client libraries in C#, Go, Java, Node.js, Ruby, Python and PHP.
Best for: Merchants with developer resources and bespoke, omni-touchpoint use cases including voice (with Google Assistant / Google Home most important)
Lex, the AI/conversational technology behind Alexa, is now open to developers to build Facebook, KiK, Slack or Twillo chatbots (or integrate with Alexa Skills Kit). Naturally, Lex includes speech capabilities and automatic speech recognition (ASR) for speech-to-text. Lex also has pre-built connectors to Salesforce, HubSpot, Marketo, Zendesk and Quickbooks.
Thanks to Amazon’s ecommerce context, Lex has one of the most robust libraries of product-related subjects and entities, meaning less manual setup. It also have a few ready-to-use bots and supports AI training.
For projects that go beyond website chatbots such as in-store kiosks and Internet of Things, you can leverage Amazon Web Services and third-party APIs, SDKs and services. Pay-as-you-go pricing makes it easy to experiment, and you’re only charged per text or speech request.
Best for: Merchants who want to build a bot with voice capabilities, leverage Amazon’s ecommerce entity libraries, or build Alexa Skills
Microsoft Bot Framework works with Azure Cognitive Services to support complex, AI-driven use cases including adaptive cards (display real-time, contextual information on orders, bookings, etc), image and voice recognition. If you want chat that listens and speaks (even in a “branded voice”) and can identify speakers’ voices and faces, this bot’s for you. Despite these advanced capabilities you can try it for free, and pay only for what you need.
Microsoft’s own LUIS (Language Understanding Intelligence Service) is used to configure your own business logic with advanced NLP and AI training capabilities. The Framework allows you to use multiple data sources (yes, Big Data!) and integrate with any channel and touchpoint.
However, Microsoft’s solution is developer-heavy. Business users have no tooling to customize dialog, meaning IT must be committed to not just the initial build, but to the entire lifecycle of the bot.
Best for: Enterprise merchants with IT budget that see conversational commerce as a strategic initiative and use the Microsoft development environment
Pandorabots is a hosted framework to build rules-based chatbots. An AI “OG,” Pandorabots has been around since 2008, with over 1M community users worldwide. It’s based on Artificial Intelligence Markup Language (AIML) which it claims is “easy for anyone to learn, even if you aren’t a programmer.” It also offers its own professional services. For developers, SDKs are available for Java, Rube, Go, Node.js, Python and PHP.
Pandorabots lacks a major feature of other frameworks — machine learning. However, it positions this as a benefit, as machine learning tools tend to experience performance lag the more intents it keeps.
Best for: Merchants that want to own their own bot code and extend over time, but don’t mind sacrificing machine learning capabilities for fast performance. While not impossible for business users to learn, AIML is more difficult to use than hosted tools with drag-and-drop capabilities.
Botsociety is a business-friendly design tool that lets business teams build live prototypes for Facebook Messenger, WhatsApp, Google Assistant and more with drag-and-drop and collaborative tools — then hand off to developers to customize code. Programmers can use their frameworks of choice as Botsociety works with Rasa, Dialogflow, Microsoft Bot Framework/LUIS, Amazon Lex and more.
Lacks (must build with your own framework):
Best for: Merchants that want the “best of both worlds” — to empower business to design and collaborate without losing the ability to own and extend code. Botsociety makes it quicker to publish to popular messaging platforms without building from scratch.
Botsociety is pre-launch as of this writing, but you can request early access.
Built on a neural network of 1 Billion Wikipedia words, Watson’s AI powers 1-800-Flowers’ GWYN (Gifts When You Need) concierge and the North Face’s voice search. Watson features an “empathy analyzer” to predict personality through text (aka sentiment analysis).
While it’s an enterprise framework, it ships with a visual dialog editor for non-techies to build simple flows. SDKs are available for Java, Node.js, .NET, Python, Ruby and other popular languages, as well as Android and Salesforce. While there are a few pre-built dialog templates, expect to build most bots from scratch.
Best for: Enterprise merchants who want to build an AI-driven chatbot or other interactive, conversational experience.
Pypstream is a build framework that serves as a conversational commerce microservice. It’s API-driven, and integrates with any web service using REST, SOAP, or legacy APIs.
While Pypestream isn’t primarily focused on retail, it has some very appealing features for travel, insurance and finance that can apply to B2C and B2B commerce scenarios. Its main differentiator is encryption and privacy that supports entry and retrieval of passwords, orders and other sensitive information, and file upload which can be used for advanced support and AI-driven onboarding and verification of new accounts.
For travel companies such as cruise lines, Pypestream can support in-chat bookings, activities, concierge services, loyalty points and rewards, dining vouchers, discount codes, personalized guest experiences and travel insurance. Hotels can integrate guest services, late checkout, room service, upgrades, extended stays, loyalty points and rewards, amenities, bookings, locations, concierge services and other discounts. Travelers can also spend points within chat.
Pypestream’s AI maintains context throughout a chat history, which is useful for personalized experiences. It can also trigger outbound SMS notifications via event-based broadcasts.
Based on the selected use cases for automation, Pypestream will extract relevant data from APIs to authenticate users, and can even trigger outbound SMS notifications via event-based broadcasts.
Best for: Enterprise merchants that want conversational commerce as a microservice that can connect to any system or device, and handle sensitive account information with encryption
Want more info on chatbots? Check out our other articles:
The post How to choose an ecommerce chatbot (14 solutions reviewed) appeared first on Get Elastic Ecommerce Blog.