Recently, there has been a tremendous amount of hype around chatbots, such that it has become difficult to filter the signal from the noise. Before you choose your chatbot vendor, it’s important to understand the pros and cons to a variety of technical approaches to the very hard, unsolved AI problem that is Natural Language Understanding, and what differentiates Pandorabots.
Pandorabots has been in business since 2008. Our visionary founders are “Original Gangsters” when it comes to chatbots, and the platform has survived several hype cycles (yes, bots existed prior to 2015 when Facebook opened the Messenger platform - in fact, the technology originated back in the 1960s). Over a quarter of a million developers and business - including top global brands - have registered for and used our platform since inception. It is a proven, market-tested, enterprise-grade solution that works in production.
Back in 2015, Pandorabots was one of only a few chatbot platforms. Today, there are hundreds, but that number has peaked (along with the hype cycle) and will increasingly decrease, as startups fail to monetize or raise their next round, and get acquihired or shutter their offering. Building on a startup that isn’t market tested is very risky!
Keep in mind that building a robust conversational agent is a long term investment. Even if you have a ton of training data such that machine learning methods may be effective, machine learning (and deep learning and other AI buzz words) aren’t a silver bullet when it comes to Natural Language Processing. It is important to deploy a chatbot early, collect data, and constantly improve the system - and for this reason, it is important to select a vendor that has been around and will be around for a while.
Pandorabots offers the widest possible range of solutions, from the free, DIY (a.k.a. build it yourself) platform offering to fully turnkey chatbot and application development (available to larger enterprises). This allows everyone, from individual hobbyists to big brands, to build a successful chatbot strategy based on internal resources available and outsource as needed.
AIML is a flexible, extensible open standard with a large community backing beyond Pandorabots, though we have historically promoted and supported its development and our platform is the most popular implementation of the latest version of the standard, AIML 2.0. We estimate that there are between 500,000 - 1,000,000 AIML developers worldwide, including the 250,000 on the Pandorabots platform, who also have a very international makeup, with robust communities in the US, Brazil and LatAm, India, Japan, and Europe. AIML also underlies some major platform offerings from very large organizations, and various startups.
There are many, many advantages to working with an open standard, chief among them that the code is something you, the creator, own and can run independently of our platform. Most importantly: the system is open - not a black box.
Machine learning based systems and APIs (like Watson, api.ai, wit.ai, etc.) are black boxes. In theory, the user can provide sample inputs, and the correct response, and the system can then identify inputs that are similar to the sample inputs and learn that these map to the same response. In reality, these systems are not effective without a large amount of training data, and it is not possible to debug the system when it returns the wrong answer. These systems also typically fail to provide any dialog management components. The art of conversation is far more complicated than distilling a single utterance down to it’s single intent (e.g., I want to book a flight from New York to LA) and determining what action to take within your application (Search Flights).
When it comes to these systems, you, the user, do not own your data, and you do not own the code for your chatbot. Your data is being fed into a third party system that dictates what input is returned, and why. From the perspective of a large brand, returning a “learned” input from a machine rather than trained copywriters on staff would be disastrous! (Think: Microsoft Tay.) But the even bigger disaster is not owning your own destiny when it comes to your data and code.
These ML-based systems have another problem: performance. Above approximately ~500 defined intents, the systems begin to slow way down. By contrast, the response time for bots hosted on Pandorabots is always around ~300 milliseconds, even for bots that have ~300,000 intents defined. For finite domains, a botmaster may not need to define over 500 intents, but generally speaking robust chatbots have a lot more intents defined, and that number should only increase over time.
The downside to Scripting Languages like AIML is that you do have to script a lot of input/output pairs (“rules”). Fortunately Pandorabots provides base content in the form of libraries that spare you from reinventing the rules for common chitchat, and other tools that streamline bot development and make maintaining your bot easy and fun. We also have a number of machine learning tools available as part of our Professional Services offering that help streamline the bot development process if your organization has decent, usable datasets - please get in touch (email firstname.lastname@example.org) to learn more.
Platforms that provide a drag-and-drop or flowchart-like interface (GUI) are pretty awesome insofar as they enable hobbyists to get up and running with a Hello World chatbot in “five minutes” and easily publish it to a messaging channel. While great for prototyping and indie developers or small businesses with limited resources and in-house developers, most of these platforms are not production-ready solutions capable of generating a real business result. GUIs are quite restrictive and have the same black box problem as ML-based systems. At the end of the day, you can’t see, download, alter, or even own your code.
Ultimately, you can’t build anything worthwhile in five minutes. Building a good bot is hard. Bot building is an iterative process that requires constant updating and tuning based on what people are saying to your bot, as you build up real client data over time. For that reason, a CMS or dashboard that allows the botmaster to review logs and make frequent updates is critical. While the current trend may be toward button-based, rigid decision-tree like bots because they are easier to build, buttons are very much a stop gap en route to solving the unsolved problem: AI’s ability to understand human language. With voice computing very much on the horizon, it’s important to start investing in supporting actual natural language inputs/outputs in your chatbot app.
Most GUIs are abstraction layers on top of scripting languages. In many cases, that language is AIML! There are also a number of forks of AIML or languages that have been inspired as an “answer” to AIML including: Rivescript, Chatscript, Superscript, Watson Dialog (now deprecated), and more. Most chatbot platforms on the market today that aren’t offering an NLP API (machine-learning based) are abstraction layers on top of one of these scripting languages or a proprietary fork. While Pandorabots is committed to making building bots as easy as possible (without sacrificing quality!) and will invest in further development of our own abstraction layer, we will always expose the layer below - the code - to you, the botmaster, so you can own and extend your project.
Building a chatbot on the Pandorabots platform is 100% free. When you are ready to deploy your chatbot, i.e., make it available to the public, we require users to have a valid credit card on file, however you can still enjoy up to 1,000 chatbot interactions for free monthly. Above 1,000 interactions in a given month, a tiny fee of $0.0025 will be charged per interaction. By default, a max monthly cap of 100,000 interactions is in place to guard against egregious charges in the event of unexpected popularity or spam. This monthly cap can be adjusted upon request. If your chatbot will exceed 100,000 interactions per month, you likely qualify for the Enterprise tier, which provides a flat fee factoring in your unique business needs.
This pricing model is consistent with industry standards for cloud services, and, to the best of our knowledge, competitive with or significantly cheaper than other offerings.
When it doubt, you can always fallback on customer case studies. After all, in some sense, the underlying technology matters a lot less than what it can enable - the business result! In fact, before selecting your vendor, we recommend that you always ask for case studies. Most so-called chatbot platforms will have a couple of name-brand clients they can reference, but fewer have referenceable customers for whom they have generated real business results. You can find some key customers and case studies on the Pandorabots homepage, with additional material available on request (email email@example.com).