UPDATE: The Watson price table below is now updated as of April 24th 2017. Differences from the prior table are documented in my Watson Pricing and Language Updates post, where I have documented the differences in pricing as well as changes in what languages Watson supports since my last post on Watson language support from Dec 2015. Only the table below is updated. The table from Aug 2016 is still available here.
How much does Watson cost? What is the price model? Enterprise software price models are never easy and when entering the API-era it does not get simpler, but it often gets more open. This is all applicable to Watson. Unfortunately, the Watson cost is split, much like what languages that Watson supports, it is listed in different places for each service / API. For my own pleasure (not entirely true) I created a cheat sheet. It is included below both as a link to the actual Google Spreadsheet as well as an embedded table.
Worth noting is that these are the public pricing and I am fairly confident that they stick pretty hard to this structure, but there are discounts involved for partners and enterprises, these are not entirely public, but since being involved with some products / projects I know they are pretty fair.
Also worth noting is that most of the services has a premium offering to move from the public shared cloud and instead go for a “single tenant instance” / private cloud setup that will ensure better encryption and compute-level isolation. I have an upcoming post on security coming up as well.
UPDATE: The supported language table below is now updated as of April 24th 2017. Differences from the prior table are documented in my Watson Pricing and Language Updates post, where I have documented the differences in language support as well as changes in Watson Pricing since my last post on Watson pricing from Aug 2016. Only the table is updated (and some typos corrected).
Unstructured data and languages is a tough nut to crack. To support all languages is not easy, but what is a bit frustrating with IBM Watson is that each service support different languages, naturally logical since it is different technologies and as an example it is easier to classify a blogpost as Swedish or Turkish compared to extract relationships or concepts as well as translate a text or sense emotions. Also, there is no Watson language support page that you can go to.
Challenges aside IBM Watson actually already support several languages for most services available today. This is impressive from several angles since many of the services actually require that Watson to be cognitive and learn the language from the beginning, just as we as humans do.
Watson language support per service
What I discovered is that it is really hard to find what language each service support, it is often buried deep in the documentation and only in a few cases listed on the main page for the service.
So, I decided to dig through the documentation and create a document that lists all services and what languages they support.
The list below is the result of me digging through the documentation of each and every Watson Service currently available.
Brazilian Portuguese, English, French, Italian, Spanish, German, Traditional Chinese, Simplified Chinese, and Dutch. Arabic is supported through the use of the Conversation API but not through the tooling interface.
Classify image method: english, spanish, arabic and japanese.
Custom and all others: english
English, Spanish, Portuguese-BR (updated 2017-09-27) and German (Discovery News only english language news).
I hope that IBM and the IBM Watson team soon will publish an official list that they maintain and update. The above list is accurate as of the date this post is published (UPDATE: This table is updated as of 24th of April 2017). I will try my best to update when new services becomes available or an existing one supports additional languages. If I find an official IBM Watson document that lists language support per service I will naturally update as well.
With this post I will try to bring some light on what Watson is. What are the components and products etc. I will also dive into the partner channel and price-model. IBM have really tried to make Watson easy to understand, almost too easy, which is very good, being IBM and all. Watson will likely assist us in ways we have not seen before.
As IBM have described it in the header-image of this post, that I have borrowed:
Watson is a cognitive technology that processes informaiton more like a human than a computer – by understanding natural language, generating hypotheses based on evidence, and learning as it goes. And learn it does. Watson “goes smarter” in three ways: by being taught by its users, by learning from prior interactions, and by being presented with new information. This mean organizations can more fully inderstand and use the data that surrounds them, and use that data to make better decisions
Watson has the ability to reason and work with hypotheses. It means that Watson will not always respond with the, according to the raw data, best suited answer, but will instead reason and weight other dependencies into the picture, just like we humans do.
As humans we base our decisions on a wast amount of data, a huge amount of data naturally, but we also take outside things into considerations, that are not pure data, but related info that we kind of know anyway and then we put all the alternatives on the table and reason about the best solution. That is how Watson does it as well.
The impressive thing with Watson is that when most of us actually forget things, Watson has access to all the data, always and dont forget, on top of that he reason and work with hypothesis. A great example here is how Watson works with oncology. Watson has access to huge amount of data and then the entire journal of the patient.
When Watson is asked about the best treatment for the patient he produces a couple of suggestions, including the evidence, and ranks them. The Doctor naturally makes the final decision, but with this, it is no risk that the Doctor misses out on some information and go on routine as “he has always done”, the Doctor now has several options that probably will contain information and suggestions that he / she (as the human being he / she is) probably did not consider nd even if he did, he gets all the supporting evidence to make the right decision for the patient.
Was that not possible prior with the classic big data / analytics tools we hear all about. Well, I would argue NO. This since most of the data in healthcare research and medical records are simply just text and images. Traditional, yet smart, analytics tools cannot read that kind of data, Watson can. it is not only about being able to read the information, it is also about actually understanding it, learning from it and draw conclusions from it, that makes the big difference and also what makes Watson so impressive. Agian, Watson think and acts a lot like a human, or the work IBM is pushing, being cognitive.
A neat thing with most solutions that are built on Watson is that we can communicate in natural language. We can simply write or talk in a natural way. Watson will even learn how we talk in different domains. This is very good since a doctor has one type of vocabulary and there is another in school or in the military. Again, as with us humans, Watson needs to be taught every language and every domain. This might sound cumbersome, but think about it like this. A traditional software is as stupid or intelligent as it is programmed, it simply only does what it is programmed to do. It is comparable to a person being told “Read what is on page 32, second row”. Watson on the other hand will after he is taught be able to answer “Watson, can you explain to me how subject X works?”, which is the summary of the data on page 32 but also with additional info from other things he has learnt.
Above is what Watson is mainly known for and we all want to know how we get our hands on that nifty software, so how do we?
The picture communicated from IBM usually looks similar to the below chart.
The reason for the confusion is pretty simple I would guess. Watson is developed in such speed that documentation and communication is not always keeping up with the updates that are being implemented, I would say that is a very good thing, rather some confusion about the amount of APIs then a technology that is updated at a slow pace. A few examples that might confuse are the following:
Message Resonance. It is a service that is available in the IBM Watson Developer Cloud, have a demo and some documentation. It is not listed on the services page, but hidden.
All APIs are avialble to play with for free in the Watson Developer Cloud. A Bluemix account is all that is needed. Every API has documentation and most have tutorials and example code. Additional code examples can be found on the IBM Github site.
High and low!
Herewithin lies a small challenge. I have followed the progress of Watson over some time now and the “Jeopardy” vs “28 APIs” is the challenge. On a highlevel it seems like a magic machine (which it really is), but scratching the surface reveals an API based technology. To start, that is fantastic since the tech-world we live in today is an API-based world. The tricky thing is to connect the one-box-solution aka Jeopardy or Oncology with these 28 APIs. When we come up with a brilliant use case and want to start exploring how to build it, it takes a lot of time to realize how the APIs connect and which ones that work best if combined with other APIs etc (Dialog and NLC as an example). When that phase is over it all becomes clearer, but to draw the lines between the goal and the services can be a challenge.
IBM Watson Knowledge and Expertice
Knowledge is naturally a challenge for IBM. A new product has been released and it takes time before it generates revenue, therefore, IBM sales will still focus on opportunities that will generate revenue now. With that said, there is no doubt that IBM is putting in huge amount of resources to make Watson succeed, but from a knowledge perspective it is still 51 Astor Place, Manhattan, NYC that is the epicentre or knowledge, with is smaller sibling in the new Watson office located in San Fransisco since recently.
The rest of IBM seems to be in a learning phase currently, but I except that to be improved significantly shortly.
In terms of support in EU, I think the UK office has a small team in place. I am also starting to see quite a few IBM Watson-positions available in most countries incl Europe.
What about partners, Watson Partners?
Since I have been around the block a few laps in terms of the IBM partner-channel, I could probably write a book about that, but lets keep focus on Watson. It seems that the channel-strategy for Watson is somewhat different and I am not even sure the classic IBM partner organisations around the world have started to look at Watson just yet.
It might even be that the partner-channel is different with Watson. It is all new models and there is also the IBM Watson Ecosystem.
IBM Watson Ecosystem
There will be a seperate post on the Ecosystem soon. This since the company I have co-founded Monies, just have been accepeted as an IBM Watson Ecosystem partner. All papers are signed, but we have not really started yet, so more info will follow.
As a summary the IBM Watson Ecosystem consist of about 400 companies that have been accepted and fulfilled the pre-requisits that IBM has in place. It has a kind of incubator / accellerator 3months program included as well, this to get your product out the door quickly, this with support from IBM in many different ways, not only from a tech-perspecitve.
It also opens up the opportunity to pitch for a piece of the $100M that IBM has put aside to invest in companies within the Watson Ecosystem.
The Watson Ecossytem is fully managed from Watson Group in NYC. The agreements are partially local, which I guess is for legislation reasons. In our case no one from our local country have been involved, 100% New York.
How much does IBM Watson cost? Pricemodel?
As usual with these large corporations it is a jungle to find out, but let me give it a try. Currently there are three price-models.
PayGo. A straight-forward pay-as-you-go subscription model based on usage. It has two tiers, one free and one paid. The free one is usually limited to amount of API calls or similar. A modern model. Most prices are public and can be found on each service, like Natural Language Classifier as an example.
IBM Watson Ecosystem. This is an entirely new model and I like it a lot. It is a revenue-share model where you share a part of the revenue with IBM based on how much of the product that relies on Watson (or if the full product does). The best thing is that you do not pay until the product generates revenue. This creates a good partnership with shared interest in the success.
Enterprises. I have not found any formal info on this and I actually guess it is pretty much up-in-the-air still. My personal guess is that IBM currently creates custom packages until they have found a good model for Watson within the enterprise segment.
But what about Watson Analytics and Watson Explorer etc?
In short it is the APIs that are the Watson product. There are also some support services like the Watson Experience Manager. Watson Group have also created a few applications based on the APIs, like Watson for Oncology, Chef Watson (together with Bon Appetite and I recommend a dinner based on suggestion by Chef Watson, great fun) and Watson for Wealth Management, but they are still based on the 28 APIs….or the 19.
Watson Analytics and Watson Explorer are simply already existing products that has been sprinkled with some magic Watson-dust. Watson Analytics has gotten itself a natural language interface, but underneeth it is still an SPSS predicitive analytics product. Watson Explorer is the old IBM Infosphere product etc. These products are also sold as classic software within (I guess) the Analytics Software brand.
That concludes this humungusly long post. I will try to keep them shorter in the future. I have simply been digging in the throwes of Watson info for a while and thought it was time to let some steam out and share my thoughts. Almost 2000 words.