Essential Guide

Browse Sections
This content is part of the Essential Guide: How to build an AI services business in the channel
Evaluate Weigh the pros and cons of technologies, products and projects you are considering.

Machine learning consultants explore use cases

Machine learning consultants and developers are among the IT channel partners discovering AI-driven business opportunities across a range of vertical markets.

In the rapidly changing world of technology, it's hard to make predictions. Here's a prediction that's likely to come true: Machine learning is poised to transform businesses, industries and, yes, the entire economy.

Enterprises across a range of vertical markets are poised to take advantage of machine learning algorithms' ability to learn from data, spot patterns and produce actionable information from voluminous data sets. Against that backdrop, IT service providers may find themselves becoming machine learning consultants. Some channel partners have already created new lines of business in the machine learning field, delivering IT skills, technology and services that incorporate machine learning into their customers' business operations.

Similar to technologies such as cloud computing, ERP, customer relationship management and supply chain software, artificial intelligence's machine learning tools, part of the broader field of data science, are primed to help businesses understand their customers better, accomplish tasks more efficiently, improve revenue and reduce business costs. Machine learning use cases span a number of fields, including telecommunications, healthcare and professional services.

As companies discover that the use of machine learning can help them make more informed business decisions, a report from McKinsey Global Institute (MGI), the business and economics research arm of McKinsey & Company, revealed that adoption of the technology is still in its infancy.

The report, "Artificial Intelligence: The Next Digital Frontier?" found only 20% of respondents said they currently use any AI-related technology at a level or scale that could be considered a significant part of their business. The report is based on a survey of 3,000 AI-aware C-level executives in 10 countries and 14 sectors.

Nevertheless, investments in the technology are an indication of things to come.

"Globally, we estimate tech giants spent $20 billion to $30 billion on AI in 2016, with 90% of this spent on R&D and deployment, and 10% on AI acquisitions," the MGI report stated. "[Venture capital and private equity] financing, grants, and seed investments also grew rapidly, albeit from a small base, to a combined total of $6 billion to $9 billion. Machine learning, as an enabling technology, received the largest share of both internal and external investment," the report noted.

Graphic showing the six steps of a machine learning project
IT service providers focus on the business problem and use case as the step of a machine learning project.

Accenture explores machine learning use cases

One role for emerging machine learning consultants is to help customers nail down appropriate use cases. Systems integrator and IT consulting firm Accenture provides an example. The company launched a variety of tools and services in 2018 to test AI systems, including machine learning algorithms. The company said its approach will help clients build, monitor and measure reliable AI systems at customer data center sites, as well as in the cloud.

Jean-Luc Chatelain, CTO, AccentureJean-Luc Chatelain

Jean-Luc Chatelain, CTO of Accenture's applied intelligence practice, said machine learning use cases include everything from analyzing and predicting the consumption of Coca-Cola to evaluating the effectiveness of drug treatments among a group of patients, as well as detecting patterns of fraud in the banking system.

The technology is also undergoing a form of democratization, Chatelain said, as use cases are not only applicable to enterprise customers, but also to small and medium-sized businesses. Still, there are challenges, and one big concern is the ability of companies and organizations to collect relevant data.

"You can't get good insights out of bad data," he said. "If you want to derive benefits from machine learning, you have to have an IT infrastructure that supports a whole spectrum of techniques involved in analytics. It's not really just about machine learning, but it's about how you capture your data. From an IT point of view, it's a matter of discipline and adding capabilities, both human and technological, to get at the data and its preparation. In addition to that, you have to then have very good algorithms that can extract the right signals."

The growing variety of machine learning use cases can benefit partners such as Accenture, as well as their clients. Accenture, for example, is using machine learning to help improve contract work processes, reduce costs and, ultimately, ensure delivery of better results for clients.

You can't get good insights out of bad data.
Jean-Luc ChatelainCTO, Accenture

"Our client contracts are complex documents that ensure we equally protect our clients, as well as ourselves, clearly defining respective roles, responsibilities and accountability," Chatelain said. "We use machine learning to 'read' these contracts, proactively assessing potential risks and defining remediation. Or, in the case of an adverse event such as the flooding of a delivery center, it helps us quickly understand liabilities that allow us to actively deploy alternative solutions, ensuring our client's delivery commitments are not affected by adverse effects."

Chatelain said the use of machine learning dramatically reduces the time it takes to read these contracts, allowing Accenture to be more agile and ultimately ensure sustained client satisfaction. In short, machine learning helps boost Accenture's business analytics and data management offering.

"For instance, we are seeing significant use of machine learning in the data veracity and quality space, and we're helping to unlock business value for clients by offering data quality as a service," Chatelain added.

Chatelain also gave the example of a client in the international telecommunications industry, whose revenue was being affected by an ineffective and expensive billing and customer service solution. While Chatelain didn't want to identify the customer, he did say the company benefited when it shifted from a time-and-materials format for evaluating labor and product efficiency to an AI-enabled, as-a-service platform.

"The company experienced a variety of positive shifts, such as enhanced data quality, reduced revenue leakage, an almost 90% reduction in cycle time and an approximately 30% reduction in cost, leading to a multimillion-dollar business benefit," Chatelain said.

Microsoft prepares partners for machine learning

At Microsoft, the company is preparing its channel partners for the uptick in machine learning opportunities. Melissa Mulholland, Microsoft's director of cloud profitability, said in addition to technical AI enablement, the company helps partners learn to frame customer problems in terms of machine learning and AI, and to develop a roadmap to where AI becomes more and more a part of the customer's daily operations.

She said the question of whose machine learning algorithms go into a partner solution often depends on the skill and sophistication of the partner's team and the application requirements. The partnership includes guidance, out-of-the-box APIs, tools and services and training.

"Partners may choose to leverage one or more of our prebuilt AI APIs, and there are even pretrained models ready to load," Mulholland said. "We also provide the tools and services to support the full AI development lifecycle and help partners develop the skills to build custom machine learning algorithms. The partner then collaborates with the customer to select the data to be used to train the model."

Mulholland also said that while partners are excited about AI and machine learning, they are to the field. She said partners are learning to build teams with appropriate skill sets, envision the possibilities for AI and machine learning in their domains, frame the customer problem in terms of an AI and machine learning and understand the AI development lifecycle and best practices in terms of processes.

She added that partners are also educating themselves on using AI tools like Azure Machine Learning, Data Science Virtual Machines and other tools that can make building AI systems a lot easier.

Mulholland expects partners that build an AI practice and integrate machine learning tools at their customer sites will see the benefits further down the road.

"Partners have found opportunity providing horizontal solutions or by using their domain expertise on industry-specific solutions," she explained. Partners, she added, are finding revenue opportunities in each of these high-level AI processes:

  • envisioning, helping customers see a roadmap where intelligent solutions take on an ever-expanding role in the customers' operations;
  • implementation, integrating prebuilt AI APIs or developing custom AI models or solutions; and
  • deployment, helping customers get their AI into production and supporting the customer's internal teams.

Using machine learning in risk prevention

One Microsoft partner is KenSci, a Seattle-based company that uses machine learning in its Azure-based risk prevention platform for healthcare. Sunny Neogi, chief growth officer at KenSci, said machine learning's ability to continuously learn from data and assist caregivers with predictions is the most valuable use case in healthcare.

Sunny Neogi, chief growth officer, KenSciSunny Neogi

"Applying machine learning to patient data is a game-changer," Neogi said. "Healthcare providers can make far better use of their time if they can predict whether a person is at risk for a chronic condition before they fall sick and direct preventive care at them. They'll spend less time fighting fires and more time preventing fires."

Neogi added there is a data access challenge in healthcare, in part because of restrictions outlined in patient data privacy laws and also due to the lack of interoperability between electronic health record (EHR) systems. In light of those factors, successful machine learning projects in healthcare must effectively incorporate machine learning tools into EHRs and other hospital systems, as well as make sure the initiative is backed by the healthcare organization's clinical and executive leadership.

"Like most successful digital transformation projects, it's absolutely critical to develop a business case and having a clear sense of the return on investment and success metrics at the outset," Neogi said. "The clinical team has to believe in the machine learning output and the impact it might drive. They have to look at the evidence, and the math has to justify taking a course of action before they take action based on those predictions."

Identifying machine learning use cases

Channel partners turned machine learning consultants must look for ways to build a framework for how they'll incorporate machine learning tools on customer projects.

Chatelain said Accenture's approach is to identify and understand the use case scenario and the business outcome the client is seeking to achieve.

"The use case will drive the architecture and whether or not a big data approach is needed and will also determine the data science approach that may or may not use machine learning -- depending on the type and volume of data," Chatelain said.

IT service providers should also keep in mind that the complexity of projects will vary. Complexity, Chatelain said, is a function of the cardinality of the number of data sources that need to be involved; the nature of these sources, structured, poly-structured or unstructured; and the type of machine learning used. For example, deep learning tends to demand a more complex infrastructure, especially for model training.

As IT service providers expand their machine learning engagements, they should remember that most IT departments are ill-equipped in term of capabilities and trained resources in these techniques.

"This is an opportunity for providers within the ecosystem to help clients achieve their goals," said Chatelain, who advised customers to conduct due diligence when choosing providers.

Dig Deeper on Emerging Technology Markets

Join the conversation


Send me notifications when other members comment.

Please create a username to comment.

What steps are you taking to pursue machine learning AI systems this year?
Hi Good and interesting! Thanks.
Stanislas Kaggle Master Atlanta.
At work we are using DATAROBOT and it is very good. We did a great benchmark in January. Actually, we decided to keep DATAROBOT but we keep also two others H2O.AI and PREVISION.IO for outstanding accuracy.
AML is cool but so hard to use with time series like Amazon ML
Stop talking about AutoML, it is too expensive
Stanislasbv thank you for exposing solutions, we also use datarobot but I did not know it seems to be free? I just launched fraud detection based on logs on the cloud version, seems powerful, already done with h2o versus datarobot - same results. Very hard to got same results on Amazon ML too long to build.
Guys, we did bigbench this year on fraud detection :
- Datarobot
- H2O
- BigML
- GoogleML
- AmazonML
- AzurML

Dataset : 13 480 000 lines , 232 columns

Feature engineering : I did it myself.

Content of file : Text, numbers, dates

Context : Fraud Detection

Results (on 32 vcores , 128 gb ram)  :

- Datarobot : 34.2 hours calculation , results 0.33941

- H2O : 28.3 hours calculation, results 0.32751

- BigML : 40 hours calculation, results 0.36

- : 27.4 hours calculation, results 0.32875

- GoogleML : 33 hours calculation, results 0.35111

- AmazonML : 34 hours calculation, results 0.34981

- AzurML : 33.5 hours calculation, results 0.35871

Top 3 ranks performance :

1. First : H2O ( 0.32751 )
2. Second : ( 0.32875 )
3. Third : Datarobot ( 0.33941 ) 

Top 3 calculation  :
1. First : ( 27.4 hours )
2. Second : H2O ( 28.3 hours )
3. Third : Azur ML ( 33.5 hours )

Feel free to ask me more information.

Top 3 ranks performance :

1. First : H2O ( 0.32751 )
2. Second : ( 0.32875 )
3. Third : Datarobot ( 0.33941 ) 

Top 3 calculation  :
1. First : ( 27.4 hours )
2. Second : H2O ( 28.3 hours )
3. Third : Google ML ( 33 hours )

Sorry Google ML not Azur ML on third.

Did you test the open source version of H2O or Driverless AI?
Too expensive?
More expensive than kaggler? Not sure
Actually we are using one automl in cloud, 40K$ per year. Salary of 1 expert : 150k$.
Michael sounds interesting , we launched in 2017 at Citi a benchmark but we did not use bigML due to poor accuracy we already knew that
Loans scoring on Citi's assets (we used a small dataset of 20M rows and 265 columns)

Good idea to share results

We had Datarobot and on top, Google and Amazon totally out on accuracy
H2O has particular good results.

No idea about, only knowing they beat Datarobot at Goldman benchmark on credit scoring but we are not able to test them on cloud due to our policy on this use case, do you know if there is any on premise version? Is it french ? Perhaps we will test it on log analysis or other use case need to know more about this company.

We have some other use cases with OCR and pattern without any policy, we are using both Amazon and Google and it works well :-)
Very nice landscape and benchmark MichaelJHR thanks, we have H2O Driverless we know it very well and we love it. We tested also many many solutions in the market. Don't say BigML , bad.

We know datarobot but we did not know and...what an accuracy and what a surprise.

Results on some scoring for marketing : : 0.384
H20 : 0.39
Datarobot : 0.409

We launched many others , no leaks , just front of or equal H2O , Datarobot just behind. We tried also one or two models we built quickly manually with some xgb ...0.43 no way.
Colinmiller , it is available online we launched easily many tests on their website.

We love H20 so we keep it. Will see in a few months regarding discussions we have actually with this challenger
Nice. Like that kind of benchmarks very hard to find on comments. I saw some good comments with H2O and and their nice accuracy on kdnuggets and actual top 2 autoML solutions in terms of accuracy.

Hopefully some tests, we use Driverless here in my company (bank), Datarobot too expensive for the same job as others. We don't know seems to be new (2016 - crunchbase), curious so we will try but not now.
I found same results with both Driverless H2O Datarobot
We shift to , the times series module on our sales forecasting use case really powerful.
We use H2O and Microsoft Azure. Datarobot is great but too expensive. Never heard about
Is it a benchmark workspace? What impresses me here is the benchmark on AutoML products. Really interesting.