Chatbots: can they talk the talk?
It is fast becoming a cliché to say that chatbots are the new apps. The average phone user uses just 3–5 apps per day, Facebook Messenger has over 1bn users, and the comfort of communicating with businesses in the same way as you would with a friend is undeniable. How much of the hype is justified? I’d like to use this post to explore where I think chatbots won’t disappoint.
1. There are different types of chatbots
Chatbots are pieces of software that provide automated responses to user interactions in a conversational interface. Interacting with chatbots doesn’t always entail typing — it often involves buttons, images and payments which can make interactions more efficient. Chatbots work in four ways: through simple click navigation, recognising keywords, identifying structures within phrases, and through unrestricted natural language as if the user is talking to a human — in order of increasing complexity.
2. Why now?
a) Strides in AI
In recent years there have been significant developments in the area of deep learning, in particular in convolutional neural networks and recurrent neural networks which have drastically improved accuracy and opened up new applications for automation, enabling chatbots to recognise user inputs (image, texts, speech) far more accurately than in the past:
WeChat, the popular Chinese messaging app, has earned a position as a kind of blueprint for the new conversational paradigm many expect to see in the West, with messaging as the platform for all kinds of internet activity from payments (e.g. buying a coffee from Starbucks and picking it up in store) to bookings (e.g. scheduling a doctor’s appointment), removing the need to download bulky applications.
c) New distribution channels
With this realisation has come a massive expansion in distribution channels for chatbots. The most obvious of these is the opening of Facebook Messenger to bots in early 2016 — a clear attempt by Facebook to stake its claim to this new platform. Google Assistant, Slack and Amazon’s Alexa are other notable distribution channels, and the latter two have even gone so far as to fund the growth of chatbots with dedicated venture funds.
3. Chatbot companies are prone to several challenges
There is a genuine risk that chatbots may be overfunded as a sector. According to Job Pal, 2016’s chatbot funding was around $360M, with the average chatbot seed round 2.5 times larger than the average US seed round.
Start-ups looking to capitalise on conversational interactions have to deal with a wildly competitive young market: Messenger already has over 30,000 bots and 34,000 bot developers, Microsoft has an estimated 45,000 developers using the Microsoft Bot Framework. On a conceptual level, it’s still unclear how the landscape will develop and which use cases will dominate. A key trend is the emergence of “master bots” such as Google Assistant, Amazon’s Alexa, and Siri, which are well positioned to become the powerhouses of the new conversational paradigm, interpreting user requests and forwarding them onto subordinate bots.
4. Look to niches
From an investor’s perspective, then, the most interesting land grab will occur in more specialised use cases, where
- Even rudimentary AI can still be effective
- A new, younger demographic expects the same ease of interaction in these niches as they experience with their friends on Facebook, Whatsapp, etc.
- Synchronicity is important
- Markets are large and anachronistic
5. Customer service is the ideal play for chatbots at this stage
Customer service chatbot businesses fulfill all these criteria.
- A large chunk of customer service queries can be easily automated, because they are simple, text-based problems. Context (Name, reference number, and so on) is more efficiently provided through text conversations than through calls. In some of our larger portfolio companies, as much as 40% of calls to customer service are simple, common enquiries that could be efficiently dealt with, on a 24/7 basis, by a simple keyword matching bot; for complex inquiries, the chatbot passes the customer onto a human. This translates into a massive and easy efficiency gain in customer service.
- This would also provide more value to the customer: according to a study by Twilio, 90% of consumers want to use messaging to communicate with businesses, but under half of businesses are actually equipped to do so.
- Unlike in other verticals such as recruiting or HR, synchronicity is really important to good customer service. If I’m looking for candidates for a new job in my company, I’m unlikely to demand an instant response. But if I want to know why my dinner hasn’t arrived yet, or why I’ve been overcharged for a purchase, chances are I want to know why immediately.
- The customer services market is massive: Radiant Insights, Inc. estimates the outsourced customer services market will be worth nearly $85bn by 2020. It’s also old-fashioned, relying far too much on manual human interaction.
Lastly, leveraging chatbots for customer service offers businesses a great opportunity to build up a relationship — a chat history — with their customers, allowing them to understand and model customer profiles better. In turn, this can be used to improve topline, by anticipating and encouraging customer purchasing habits, e.g. Odeon messages you letting you know that a new film starring your favourite actor is coming out and prompting you to buy a ticket. Messaging can thus act as a hub where sales and customer service converge.
6. There are several different models for CS chatbot companies
There are 3 key parts to the CS stack:
- AI — ranges from basic triage, with rules set by the user, to complex machine learning algorithms with high data requirements
- Customer messaging service — lets the customer interact with customer service, e.g. Facebook Messenger, Slack
- Helpdesk — lets customer service interact with customers, e.g. Zendesk
Different CS chatbot companies are targeting different parts of the stack. There are three main approaches:
- Platform — create a new platform which integrates the existing helpdesk, messaging services, and AI functionality. This allows a more tailormade approach to automated customer support as the platform is designed specifically with automation in mind
- Tech enabler — treats the chatbot functionality as a new tool for developers; aims to simplify the bot creation process
- Integration — treat the AI itself as an integration into the existing helpdesk, so that the chatbot functionality occurs within the current platform. Easier to onboard new clients, but vulnerable to being made redundant by equivalent features rolled out within the existing helpdesk
7. Business models
Most companies in this vertical aim in the long-term to operate a SaaS model, charging a monthly fee per agent using the product, or per customer conversation. A common approach in the meantime is targeting enterprises on a bespoke basis, and using the large contracts to help finance the development of a more scalable product.
There is also a separate question of who the target market is, however. For companies favouring machine learning and deep learning techniques, enterprise customers are the only real options because they have large enough datasets for the algorithms to run effectively. When the “AI” involved is less data hungry because it is trained on FAQs, ongoing customer service interactions, or supplemented with tree-like flows, this opens up the customer pool to smaller businesses with smaller bodies of historical chat data.
The most serious risk to start-ups looking to automate customer service is that the incumbent helpdesk and CRM firms innovate into this space themselves. Salesforce has signalled that it intends to do exactly this with its Einstein product and so does Zendesk with Zendesk Automatic Answers. Both products are far from ready, however, and it is unclear whether incumbents are agile enough to beef up their AI capabilities before new entrants.
A more fundamental objection might be that since customer service chatbots are concerned with cost reduction, they will not be a focus for topline-obsessed companies, meaning that making sales will be difficult. As sales and customer services converge, however, this point will become decreasingly relevant, and the need to offer customers conversations for commerce and support will become imperative.
9. Will 2017 be the year of the customer service chatbot?
Customer service chatbots will become integral to customer service; the question for most B2C firms is not whether to use CS chatbots, but when. At their current stage, many chatbots are rudimentary and fail to live up to the hype. But with the sector increasingly well-funded and customer service chatbot start-ups rapidly completing successful pilots, customer service is primed to really bring chatbots into the conversation.