This post was originally published by Patrick Meyer at Medium [AI]
Coming from my active watch on the conversational assistant market, I give you some statistics and information on the almost 550 solutions I studied. As a conclusion, I will give you my view about the future.
Source: photo by Patrick Meyer
The event still seems to feel everything in the minds of those who followed it so much was it striking. At Google’s annual conference on May 8, 2018, the company will present its new generation of assistant: Google Duplex. It is a truer-than-life assistant. For once, a human did not call a robot but an assistant called a human (a hairdresser) to make an appointment instead of a person. The generated voice was so close to the human timbre and its variations that it was almost impossible to tell whether it was a real automation or a specially made set-up to simulate what such an assistant could be . The whole dialogue was identical to an everyday conversation, quite standard, with pauses and intonations. In retrospect, the aim of this demonstration was to create a buzz about a project rather than to present an already available product. But it did suggest the potential future of support solutions. In any case, it showed the path Google wants to take.
It is difficult to know how much the market for conversational assistants will grow in the next few years. The values given by analysts sometimes vary from simple to double. By averaging the values of the figures announced by the analysts, we can envisage a market valued at approximately 11.05 billion dollars in 2024 and 44.58 billion in 2027, i.e. a CAGR (Compound Annual Growth Rate) of 26%. All analysts agree to indicate an exponential growth with most certainly a levelling off at some point in time.
Solutions account for by year of creation of the company. It is important to note that the year of creation (on the abscissa of the graph below) does not always correspond to the year of availability of the solution. Many companies have often had an existence in a related field before creating their own conversational solution. This makes it difficult (unless you spend a lot of time searching when the company communicated about its solution) to know exactly when the solution was created.
However, let us go back a few years to fully understand where we are starting. It is commonly accepted that the oldest solution helping you to create your own conversational assistant (also called “chatbot”, contraction of “chatterbot”) is called A.L.I.C.E. This acronym stands for Artificial Linguistic Internet Computer Entity. It is an application developed by William Wallace since 1995 to simulate a conversation between a human and a machine . It allows the execution of symbolic rules consisting in testing the presence of keywords in the user’s sentence. This mode of operation is very simple to use and is sufficient to cover many use cases. On the other hand, it quickly finds its limits as soon as the conversation becomes more complex and the vocabulary becomes very large.
The principle of conversational assistant was really popularized to the public on October 4, 2011 during the presentation of the new iPhone 4S integrating a revolutionary application for the public called Siri. It is important to see that many consumer innovations have been taken up in companies since employees are very often users of this type of solution in their personal lives and expect companies to provide the same service (see Enterprise Social Network, Instant Messaging, powerful search engines, etc.) It was therefore natural that assistants would also arrive in companies someday.
To understand the genesis of Siri, we have to go back 8 years, to May 2003 when DARPA launched its project called CALO. On the Calo site of the SRI in 2006 is written :
The Defense Advanced Research Projects Agency (DARPA), under its Perceptive Assistant that Learns (PAL) program, has awarded SRI the first two phases of a five-year contract to develop an enduring personalized cognitive assistant. DARPA expects the PAL program to generate innovative ideas that result in new science, new and fundamental approaches to current problems, and new algorithms and tools, and to yield new technology of significant value to the military.
The team has named its new project CALO, for Cognitive Assistant that Learns and Organizes.
The name was inspired by the Latin word “calonis”, which means “servant of the soldier”. The goal of the project is to create cognitive software systems, i.e. systems that can reason, learn from experience, be told what to do, explain what they do, reflect on their experience and react strongly to surprise.
The software, which learns by interacting with and being advised by its users, will handle a wide range of interdependent decision-making tasks that in the past have resisted automation. A CALO will have the ability to engage in and direct routine tasks and assist when the unexpected occurs. To focus research on real-world problems and ensure that the software meets requirements such as privacy, security and trust, CALO researchers are themselves using the technology during its development.
This project brought together more than 300 researchers from 25 of the best academic and commercial research institutions. Siri was therefore initially developed by SRI International through this DARPA project. However, SRI separated from Siri, Inc. in 2007, and the now independent company launched a personal assistant application in February 2010.
However, this application will not have time to live its life since the company Siri, Inc. bought that year by Apple to become the personal assistant of all iPhone users. The company’s web page disappears the day after Apple’s announcement to the press. Although Siri is not adaptable for professional use, it will remain the starting point for the large-scale deployment of this technology, driven by sales of this famous smartphone.
So what is the difference between a solution and a “chatbot”? The fact that a person, external to the coding team that designs the solution, can configure the chatbot to make it say what it intends. As Richard Wallace wrote in his A.L.I. C.E. “Don’t Read Me” webpage .
The botmaster is you, the master of your chat robot.
It is difficult to find which the first real commercial solution is, but the Pandorabots solution is certainly one of the very first. It is the result of Richard Wallace’s collaboration with Franz, Inc. Pandorabots  created in 2002 and the company is still active in 2020 with more than 350,000 bots to date (indicated on the website home page).
History will remember that the hosting was free and that the Shop menu (which appeared in November 2005) allowed buying only Pandorabots T-shirts and not Bots! .
The year 2015 sees the release (almost on the sly) of IBM’s Watson Dialog, introducing on a large scale and carried by an international player, the learning capabilities of AI to serve language processing and thus improve the mechanisms for recognizing user requests.
The graph showing the distribution of solutions according to the year of creation of the company highlights the explosion of solutions during the years 2013 to 2018 with the creation of 2/3 of them. The years 2015 to 2017 are particularly noteworthy as these 3 years saw half of the solutions created worldwide. 20% of the companies publishing a conversational solution were created in 2016, a real pivotal year for creation. There was one before and one after this date.
Even if companies continue to be created, the pace has slowed down dramatically with only 9 solutions created in 2019 and the year 2020 is likely to be on the same trend.
The solutions are sorted by location of the company’s headquarters. As you might expect, the United States of America is the country that hosts the most solutions with 219 chatbot solutions identified. Surprisingly and although it is a fairly small market, France comes second with 77 solutions. Then we find India with 40 solutions. Then the United Kingdom with 28 and Canada with 20. Germany, Spain, Israel and Singapore have between 10 to 20 solutions each. The rest is below 10 solutions per country according to my list. I count only 1 solution per publisher even if this one publishes several different solutions.
48 countries in total have at least 1 solution. All continents are represented.
The analysis of editor typologies allows classifying them according to the following main categories:
- The pure players
- Natural Language Processing Solutions Providers
- ERP and/or CRM solutions publishers
- Call Center, Contact Center and/or Livechat Solutions Publishers
- Cloud Services Solution Providers
- Digital Service Companies (DSCs) and other IT companies
- The others
Pure players are among those publishers which base almost all of their revenues on the sale of a Bot solution and associated services. Most of these solutions are multi-tenant and hosted in the Cloud.
Publishers of Natural language processing (NLP) solutions, often established in their markets, have seen opportunities in the explosion of chatbots demands and have created solutions based on their own algorithms.
An Enterprise Resource Planning (ERP) or Customer Relationship Management (CRM) solution will benefit from hosting a conversational solution, especially if this solution is complemented by connectors and privileged links to the modules of the software package. These editors generally do not have conversational skills and usually proceed by Mergers & Acquisitions.
For solutions used to manage the exchanges between users, customers, suppliers with a company, automation via a conversational solution becomes a mandatory functionality. In Livechat solutions, one of the agents becomes virtual and is able to take care of part of the requests and thus respond autonomously.
Enterprise software publishers which have migrated to the Cloud or which are Cloud pure-players such as Amazon WS, Baidu Cloud, Google Cloud Platform, IBM Cloud, Microsoft Azure, SalesForce.com, etc. for the best known, provide a suite of services and must include a conversational solution in their “toolkit”.
Digital and technological services companies (Consulting and IT companies), experts in information technology, must master these technologies since their customers request them. Many integrate existing solutions and some took the gamble to create their own.
Finally, individuals who generally make the source code available (often in Open Source mode) create solutions. There are also companies that have developed their own solution and make it available to the greatest number. There are also researchers in Artificial Intelligence and language processing who share their discoveries, or even their thesis results.
Unsurprisingly, the number of employees at Bots solution providers have less than 50 employees (73%).
1/3 of the companies created in 2017 have less than 10 employees, which remain very small companies. 1/3 of the companies created in 2016 have between 11 and 50 employees and often provide a mature and recognized solution on its market. Companies that publish several solutions or that already had a software base (for example a contact center) generally have the capacity to exceed 50 employees.
If we take the raw figures for company sizes, we see that, proportionally, France is the country in the Top 10 with the most “small” companies, with almost half of the companies not exceeding 10 employees. Italy and the United Kingdom are close behind. It should be noted that for France, there will be more company creations in 2017 than in 2016 a contrario of certain countries, which could explain this one year delay compared to countries that have more companies with more than 10 employees.
Each solution is different from the other. It offers services, interfaces and facilities that are specific to it. Nevertheless, when studying a large number of these solutions, it is possible to identify several categories of solutions. Either the editor expresses it clearly, or all the functionalities correspond to an identifiable category.
Here are these categories:
- General purpose; 30% of the market solutions. I start with the most representative but also the most general category. The solutions in this category allow you to configure all types of wizards without special features. They recognize the question (the user’s intent) and give the most appropriate answer from an answer that has been configured by a subject matter expert. The solution is often presented as being able to manage several categories. When the solution is too general, it is often too general and does not stand out from its competitors.
- Retail and e-commerce (Marketing); 26%. This is certainly the most represented specialized category. It covers the commercial field in support of the sales site. It is used to retain the customer, to help him choose a product, to answer questions about product availability. The chat window opens when the user remains blocked on the page. The solution provides tools such as carousel, product sheet, and delay can be added before answering in order to appear more human, friendlier. It has a wide range of statistics about the user such as where he comes from, if it is his first visit, his browser’s name and version. Many solutions have a kind of mailbox for unresolved requests. The Bot will seek by all means to obtain the email address of the user, make him subscribe to the newsletter. The assistant becomes a sales advisor and the solution does everything to facilitate the Bot’s configuration in this sense. The solution has a wide range of connectors to e-commerce software, contact management software, etc. allowing to know the customer better and to respond in the most targeted way possible. Often, it also has an extensive list of connectors to social networks.
- Customer support, sales, helpdesk; 17%. These solutions are designed to respond specifically to user issues. They can handle a large volume of questions and have links to documentation. They facilitate the creation of complex dialogues whose objective is to help the user find an answer. These solutions based on the principle of knowledge, guides to unlock the user. They also participate in the automation of processes by allowing the user to upload documents, to retrieve information from the information system.
- SDK (Software Development Kit); 5%. This solution is a toolbox allowing developers to build their conversational solution. It provides a whole set of functionalities (libraries) allowing managing the understanding of the request, the execution of functions, and the management of the dialog context… It is necessary to have programming skills to realize a Bot. Most of the time, these solutions are available in Open Source. Sometimes the publisher provides a Cloud execution environment, which is then paid for.
- Banking, Financial Services and Insurance (BFSI); 4%. These solutions target the banking and insurance market by providing functionalities specific to these areas: banking management, solvency analysis, quotes, recommendations/advice, routing of a prospect to an agent, claims management, etc.
- Personal Assistant; 4%. Probably the most “public at large” of the solutions since they allow you to create assistants which come to help you in your everyday life.
- Healthcare; 2%. Few of healthcare solutions are configurable. Most of them are already configured and do not allow you to change their settings.
- Travel and Hospitality (Tourism); 2%. These solutions specialize in tourism activities and allow the configuration of Bots that can help users plan their vacations, indicate which tourist places to visit, available hotels, etc.
- Recruitment; 2%. These solutions enable the creation of assistants for the process of recruiting new employees. They can answer questions about the company, take competencies tests, gather CVs, serve as a point of contact for the hiring, etc.
- Device Assistant; 1%. These solutions are designed to be embedded in an electronic device such as an oven, fridge, television, etc. They are generally voice based and embed the AI necessary for request recognition.
- Automotive; 1%. These solutions are designed to be embedded in a car and help the driver on his journeys. This category is a specialization of the Device Assistant category even if the car is a rather “big” device.
- Conversational Process Automation; 1%. This solution combines the capabilities of a Bot with the process automation functionalities generally found in BPM (Business Process Management) software or RPA (Robotic Process Automation).
- Conversational Search; 1%. Its goal is to allow you to create wizards that can help you find the right document, the right paragraph, in a content manager.
- Project Management; 1%. This solution allows you to create project assistants. They support the project manager in his daily management activities. Bots created with these solutions become a kind of PMO (Project Management Office) assistant.
- Other; 5%. This last category is not one. It is used to group together all categories that are below 1% of presence. These solutions make it possible to respond to administrative requests, they graphically display data from databases where the user formulates his or her request in natural language, help with the reception of new employees (onboarding), are specialized in real estate, manage support tickets, are specialized in agriculture, are specialized in content management and its restitution.
As in any active and growing market, we are witnessing company movements. When there is an acquisition, the solution is either integrated into a larger offering or remains as a product supported by a larger company and can retain their name. Some solutions die for reasons that often remain unknown.
In the case of a purchase of existing skills and functionalities, the solutions disappear completely and are integrated into a product that no longer allows the creation of a chatbot (e.g. Snips bought by Sonos in 2019 ).
Sometimes a software suite integrates a chatbot solution to add those specific features into the platform that hosts it. One example is Speaktoit’s api.ai, acquired by Google in 2016 , which became Google DialogFlow, thus joining the Google Cloud Platform functionalities. On the other hand, the French company Recast.ai acquired by SAP in 2018 so that the German publisher could add a wizard solution to its ERP development platform .
The death of a company is not always pleasant, especially for those who live it, but they are part of the law of the market and adaptation. Although it is difficult to determine the exact date of death of a solution (the company is generally no longer able or willing to communicate), the disappearance of the website is often indicative of the end of the company’s life.
Of the 31 identified deceased solutions, 50% are American. 2/3 of the missing solutions are less than 5 years old.
Note Bene: Mergers and Acquisitions solutions with disappearance are not included in these statistics. Open Source solutions that are no longer maintained are also not counted.
It is not possible to finish these statistics without a touch of lightness. It is interesting to note that half of the solutions begin with one of the following letters: C A S B M I.
6% of the solutions start with “Bot” and 3% with “Chat”. One solution has a single letter name.
Here are my predictions for the future of conversational assistance solutions:
- Unless there is a technological breakthrough, the number of new solutions created will not exceed dozens per year.
- The new general purpose solutions will not be able to survive among the hundreds of existing solutions. These solutions will have to make a choice and become specialists in a specific niche.
- A solution that is too generalist can only fight by reducing its prices and will therefore have to make volume in order to survive, a volume that can only be achieved if it already has an established base, live references.
- Some solutions have made design choices that will penalize them in the future. They will have to make a profound change or will have to fail.
- The solutions will have to go to level 4 complexity and thus allow dialoguers to reuse several different business domains in the same interface. To find out what this implies, I invite you to read my article on dialog complexity.
- The language processing technologies currently used by the solutions are mature and allow good recognition rates (intent detection and entity extraction) with a reasonable number of examples. These solutions are mature and robust and today meet the challenges of quality and robustness of service. They will have to work on analytics to help the people who maintain the solutions in their improvement work.
- Due to the advance of Artificial Intelligence technologies, we will see the appearance of modules or new solutions that will be able to converse without the need to create the dialogues “by hand”.
- In the years to come we should see corporate movements with Mergers & Acquisitions and corporate deaths. The market is certainly growing, but not all solutions will survive. Acquisitions should allow publishers, not having yet an assistant solution, to integrate conversational solutions into their own solution.
- The solutions will have to manage more and more the context of their interlocutor in order to provide an increasingly personalized service.
Solutions for creating on-demand conversational assistants have gradually moved from the shadows to the light in just a few years. The market is maturing as most solutions have the same basic functionality and try to stand out with related features. The United States of America is the country with the most publishers, followed by France and India. Most of these companies have less than 50 people, especially when they only provide this type of solutions. The e-commerce sector is the most covered by solutions that can replace a sales consultant and help users to choose. 2/3 of the solutions that disappeared from the market were less than 5 years old.
The tricky question: “What’s the best solution?” There is no one best above all others, but several adapted to each use case. In order to be able to choose among all these solutions, a list of criteria must be drawn up to quickly filter out a dozen solutions (e.g. localization, language, etc.). Then, a second list of criteria must be applied to arrive at 2–3 solutions. Then you will need to test them so that you can check their overall performance and judge the quality of the configuration interface.
1] What is Google Duplex? The smartest chatbot ever, explained- Article DigitalTrends https://www.digitaltrends.com/home/what-is-google-duplex/
2] A.L.I.C.E — Wikipedia — https://en.wikipedia.org/wiki/Artificial_Linguistic_Internet_Computer_Entity
3] SIRI project CALO — Site- http://www.ai.sri.com/project/CALO
 Don’t read me — A. L. I. C. E. and AIML Documentation — https://www.mediensprache.net/archiv/pubs/2760.html
5] Pandorabots in October 2002 — Internet Archive-https://web.archive.org/web/20021001074218/http://www.pandorabots.com:80/pandora
6] Shop menu of Pandorabots in November 2005 — https://web.archive.org/web/20051127022134/http://www.pandorabots.com:80/botmaster/en/store.html
7] Sonos buys Snips, a privacy-focused voice assistant — https://www.theverge.com/2019/11/21/20975607/sonos-buys-snips-ai-voice-assistant-privacy
 Google acquires api.ai — TechCrunch- www.techcrunch.com/2016/09/19/google-acquires-api-ai-a-company-helping-developers-build-bots-that-arent-awful-to-talk-to/
 SAP Increases Commitment to Powering Innovation in France — https://news.sap.com/2018/01/sap-powers-innovation-france-acquires-recastai/
This post was originally published by Patrick Meyer at Medium [AI]