Inforum 2019 - Can Coleman AI make self-service data science a reality?
- Behind the Inforum 2019 GA announcement of the Coleman AI platform lies an intriguing data science debate: will customers embrace self-service data science? Here's my look into how Infor weighed this out internally, and what they decided.
Quick question: what do you think was the overriding theme of Inforum 2019? Cloud? No - though cloud is clearly Infor's future. User adoption of Infor's multi-tenant CloudSuite remains their central focus/challenge.
Verticals? No, though Infor is convinced that its industry approach to ERP clouds will be its market edge.
AI? Not quite - though the Coleman Platform GA (General Availability) announcement was arguably the biggest news announcement of a show that was not breaking-news-heavy.
Process intelligence was an interesting twist, one that ties into Infor's expressed commitment to make ERP projects easier. I covered that in Inforum 2019 keynote lead-up - executives air challenges and customers share wins.
No, the winner goes to something more practical: customer productivity. Infor adds marketing sizzle by calling their approach "productivity delivered." This is where Infor's AI and automation pursuits come into focus. As in this slide from our Infor press and analyst day:
To improve productivity, you have to define it
I could squander an entire blog post about whether ERP - or technology in general - has truly made companies more productive. To sharpen this discussion, we need to define what productivity actually means. During the day two Inforum keynote, Rod Johnson declared Infor's intentions:
We want to be the ones helping you to achieve productivity outcomes for your business.
No customer would object to that. But how? Johnson spelled it out:
- Productivity of people - including software with fewer clicks, and better access to data
- Process level productivity - a "relentless focus on automating processes and eliminating manual tasks from processes"
- Macro issues - addressing big picture issues around labor, inventory, and capital allocation
Those are worthy ambitions. But in reality, each industry has different productivity obstacles. Infor delved into that in manufacturing and health care, but there's only so much you can do in one keynote.
Connecting AI to productivity
So how does Coleman AI factor into Infor's productivity message? Start with Coleman's three components:
- Digital assistant (announced at Inforum 2018)
- Coleman platform (GA announced Inforum 2019)
- Infor RPA (expected release calendar year 2020)
I'm not a big fan of product slides. But I'd much rather show you this Coleman platform slide than try to define it myself:
Infor believes that this "ensemble" approach to AI is differentiated from what other enterprise software vendors are doing. If you're skeptical about "Our AI is better than your AI," well, welcome to the club. But the real story to watch here is Infor's attempt to "operationalize" AI.
In other words: can we make it easy for business users to serve their own AI/predictive needs with their own data - without elaborate projects and IT support? As Adrian Bridgwater wrote:
Infor CEO Kevin Samuelson says that the practical operationalized use of AI and machine learning in the enterprise remains low because most tools are deeply technical and developer-centric. He suggests that too many of these AI tools have been designed for experimental projects and are therefore difficult to implement complete projects with.
In an attempt to resolve this situation, Samuelson says that his firm’s Coleman AI platform provides industry-specific starter packs (templates) to accelerate the development of repeatable big data, machine learning-based AI projects. These templates are highly personalized and tailored to specific customer data and usage patterns. Further, they are designed for use by ‘citizen developers’, who don’t need extensive data modeling skills.
So will this work? In my own discussions with Infor, they acknowledged there is internal debate on whether customers will embrace what you might call self-service data science. But we've certainly seen widespread adoption of self-service BI, so I get why Infor is betting on "operational AI." To get a handle on Infor's motivations here, there's no one better to talk to than Ziad Nejmeldeen, Chief Scientist, Infor.
Dynamic Science Labs and their Infor mission
I first met Nejmeldeen and his team during an on-site visit to Infor's Dynamic Science Labs at MIT in Cambridge (Bring your own data (science) - a day with Infor's Dynamic Science Labs at MIT). Quite a bit has changed since 2015. Several of the pilots I wrote about in that piece have become Infor products:
- The Pricing Science product (see my use case here) is now part of CloudSuite Distribution, and accessible to any CSD customers.
- Inventory Intelligence for Health Care is now productized, with three customers live,
Making data science accessible is core to Nejmeldeen's mission. As he said to me about Pricing Science:
One thing we're really proud of was we have so streamlined the implementation configuration, that it can be done with non-data scientists out of ICS (Infor Consulting Services), and it's a four week process. So that was something we weren't sure about, whether it was even possible to do or not.
Nejmeldeen's team has three dedicated scientists with Infor Nexus, working on a range of supply chain network projects. One they are hoping to expose to the public:
We're looking at whether the community information we have from Nexus has predictive power on macro-economic indicators that we can publish to the world and share with everyone.
I joked with Nejmeldeen about using Nexus data to predict the next recession. Not the happiest goal, but Nexus data can be a leading indicator:
Just based on the orders, there's already been a major shift on the fashion manufacturing side, from China to Vietnam. Vietnam is taking off, China is coming down - just based on what we're seeing with our customers. That's interesting because we get somewhere around the four to six months lead time over what you would see on actual receipts.
Business users and the citizen data scientist debate
And what about this Coleman platform debate? Nejmeldeen:
What's going to be interesting is seeing which idea becomes true here. One idea is that customers are ready for a platform, the Coleman platform if you will. But any platform that will allow a citizen data scientist, somebody that's not trained as a data scientist. Just like in the past you didn't have people that were trained on BI, but then everyone starts using all of these different BI tools, and now they're basically power users on that.
Will citizen data scientists emerge?
The idea here is: can you have citizen data scientists on the customer side that are comfortable taking a platform like Coleman, putting their own data into it and running ML, coming up with an answer, having Coleman help them interpret that answer, and then utilizing that in the business. That's one possibility.
So what would be the contrarian view?
The other school of thought is that customers aren't ready for that... They weren't ready when Amazon introduced TensorFlow, which is a machine learning library. Then Amazon introduced Sagemaker as basically a front end to TensorFlow, individual customers weren't ready for that either.
Of course, data science teams were all over it:
You have data science groups that were ready; I have people that are very comfortable with both TensorFlow and SageMaker.
Can Coleman's toolkit win the contrarian view over?
Coleman is supposed to be a much more friendly way of utilizing the library that is in TensorFlow, and it gives you a means of taking data and cleansing it, formatting it before you even apply the ML. It gives you an ability to then understand why the ML chose the attributes that it did. Even then, you could argue that customers aren't ready for Coleman either; it's still too much of a reach.
The market will decide. But Nejmeldeen is clear on his purpose:
I think if at the end of this, every customer we deliver Coleman to, if they end up having to come back to us in order to repeat the process over and over again - I'd like that to not be the case. So I'm excited at the prospect of: are we going to be successful here at actually getting customers empowered or not? Because I really feel like we're at a point now where this will revolutionize companies the same way BI did.
One thing Nejmeldeen is happy about: even if customers end up needing more AI support, "Coleman as a service" should allow Infor to get AI services up and running quickly for whatever use case is needed, from predictive maintenance to next-best sales offers.
Infor's Rick Rider: AI can't be out-of-the-box, it has to be personalized
Before presstime, I had the chance to ask Rick Rider, Infor's Senior Director, Product Management to weigh in. Rider made a key point: Coleman AI can't be entirely self-service, or it will lose its competitive edge for customers:
It does require a little bit of science, in that if two customers have the same footprint, we're not going to develop ML for every customer to use in the exact same. Because even if two customers have the same footprint, and maybe even the same data schema, the data is still significantly different. So you have to train it. It's got to be personalized for it to give a customer a competitive advantage by using it.
Rider brought up that "operationalizing AI" lingo. It's basically an antidote to customers feeling overwhelmed by AI possibilities, or getting caught in long-winded pilots.
If I can show you something that you can do in a matter of weeks, and literally have a drag and drop interface to where then I can really start to take advantage and input something into my system or give me an answer. So we started there, we started building and building.
I told Rider is that this reminds of our recent diginomica low-code debate. I believe "low-code" gives business users the chance to build useful workflows and productivity apps, but it doesn't replace the need for sophisticated mobile app design and coding for customer apps. Does the same principle hold true here? Rider:
It's absolutely spot on. If you take a look at our roadmap and some of the sessions we've been doing here, you see there is a line of, "Hey, half of our roadmap is focused on how can we get towards auto-ML and the citizen user interface." But that's not to say that the hardcore data scientists won't also use that because it makes it faster. Then we've got another path for the advanced users because they realized that the infrastructure that we have with Infor OS makes things real for them very, very, very quickly.
My take - bring on the customer proof points
Infor put an unspectacular day one keynote behind them with a tour de force keynote loaded with customer interviews on day two. When it comes to Coleman projects, I suspect next year will be the chance for Infor to get more of those on the keynote stage. Infor also presented on another Dynamic Science Lab product in the keynote, Infor Talent Science, including new "intelligent" team-building features called Team Insight.
I was glad that Infor tied in the impact of diversity and inclusion on effective teams. That's another sign that the commitments Infor has made under Charles Phillips will continue under CEO Kevin Samuelson, who made a point of coming on stage to discuss these issues.
On the AI/ML adoption front, EAM customer CERN spoke with us yesterday about their Coleman project - I'll write about that next. Whether Infor will succeed in a self-service data science approach remains to be seen. Obviously they are ensuring these tools will be used by data scientists as well, so even if business analysts don't take to the tools, the investment won't go to waste.
I suspect the bigger obstacles will remain getting the right depth and variety of data necessary for predictive and prescriptive models to be relevant - and the tedious chores of cleaning/merging disparate data sets. Infor thinks the Coleman platform has answers there too, via their Infor Data Lake and automated data cleansing. That's a topic for another day.
The customers I spoke with about this fell into two camps. The first camp is already on a path towards transformation and improved analytics, in some cases via Infor Birst. These customers were all interested-to-enthusiastic about getting Coleman going. Other customers, not yet on such transformation journeys, were looking for enhancements connected to their existing software - which, in many cases, is not CloudSuite (yet).
This second group of customers also wanted to learn more, but they were thinking more about first steps with Birst and perhaps data lakes - and they weren't sure how Infor was going to get that functionality in their hands quickly. That ties into Infor's improved project delivery initiative, which I got into yesterday. That's going to be important if Infor wants the "productivity delivered" narrative to stick.
But for now, customer demand for Coleman doesn't look like an issue for Rider's team:
We've gotten some referenceable customers that have come out very quickly because of the progressive things that we're doing. Our prospect list - it's almost hard to manage right now.
Classify that under problems you want to have.
Updated, 10am PT Thursday the 26th, with additional resource links. Also note reader comment below.