Open Data Group (ODG) has rebranded as ModelOp, reflecting the company’s sole focus on Model Operations and the rapidly growing need in large enterprises for this critical new capability, which is essential for realizing the value from their investments in AI.
According to Gartner, “The democratization of ML techniques in the last few years has seen the proliferation of model development practices but unfortunately a majority of these models are neither operationalized nor deployed at scale. This capability is becoming critical for the survival of data science teams and this urgency will push MLOps toward the Plateau of Productivity in two to five years.”*
ModelOp was founded as ODG in 2016 by Pete Foley, CEO and Stu Bailey, CTO, with funding from the founders, advisors and Silicon Valley Data Capital. Bailey and Foley previously worked together at Infoblox, where they pioneered a new category of IT infrastructure, leading to a successful IPO in 2011 and an acquisition of the company by Vista Equity for $1.6B in 2016. Prior to co-founding Infoblox, Bailey as the technical lead at the National Center for Data Mining, worked closely with the pioneers of data science. Foley previously served as CEO at Port Authority (acquired by Websense) and Ring Cube (acquired by Citrix), and as chairman of Graphite Systems (acquired by EMC).
“We got ahead of the market when we started this venture, convinced that the unique characteristics of data science and machine learning (DS/ML) models would require new organizational and technical approaches in order to realize their value, especially within large enterprises” said Pete Foley, CEO of ModelOp. Foley continued, “Our expanding customer base, and the level of activity we are seeing in the industry overall strongly validates our vision and makes us extremely excited for this next phase of growth.” ModelOp’s customers include five of the top 10 largest financial institutions, as well as Fortune 500 manufacturers, insurers, and credit bureaus.
Models are a New Kind of Software
DS/ML models are the engines that organizations use to turn their data into value. In order to capture that value, DS/ML models must be integrated with enterprise applications that use the models’ predictions to automate and improve decisions like credit scoring, fraud detection, bond trading, customer retention, manufacturing operations, supply chain optimization, ecommerce sales, ad conversions and virtually anything that can be driven by data.
DS/ML models are unlike conventional software in several key respects:
● Conventional software is deterministic and, once deployed, operates as written without change. DS/ML models are probabilistic, and are “trained” using large quantities of data. Over time, DS/ML models decay, producing degraded results unless they’re re-trained or re-written.
● There are many different methods and tools for producing DS/ML models, but no standards for how they’re deployed and managed. This makes it very difficult for large organizations to implement effective pipelines for moving models into production at scale while maintaining corporate standards for performance, availability and compliance. It’s not uncommon for models to lie fallow for 6-12 months from the time they’re complete until they’re deployed in enterprise applications, which can cost millions in lost sales and inefficiencies.
● The full lifecycle for DS/ML models – commissioning, development, deployment, monitoring, retraining, governance, and retirement – touches all aspects of the enterprise including lines-of-business, IT and compliance organizations. In many companies there’s no clear “owner” for defining and managing the strategy and processes necessary to implement and manage the model lifecycle. This translates directly into the organizational friction and corresponding delays that keep models from reaching production.
● Models are increasingly subject to regulatory scrutiny, driving new demands for robust governance processes and tools for maintaining and demonstrating compliance.
Model Operations, “DevOps for AI”
According to The Forrester Wave™: Multimodal Predictive Analytics And Machine Learning Solutions, Q3 2018, “Data scientists regularly complain that their models are only sometimes or never deployed. A big part of the problem is organizational chaos in understanding how to apply and design models into applications. But another big part of the problem is technology. Models aren’t like software code, because they need model management. And models must make it into applications.”
Model Operations addresses these issues by automating the release, activation, monitoring, performance tracking, management, reuse, maintenance and governance of AI and ML models. Just as the rise of SaaS has driven wide adoption of DevOps processes and tools across the enterprise, the rapid adoption of AI is driving the need for enterprise-class model operations.
ModelOp’s unique approach starts with its solutions portfolio and expertise in data science, ML engineering, IT operations and business transformation, to drive customer success throughout their AI journey, backed by its enterprise software which automates model deployment, monitoring and governance independent of the data science workbench, modeling language and underlying technology platforms.
“Rapid advances in AI and ML are putting powerful tools into the hands of a growing group of professional and “citizen” data scientists, resulting in an explosion of models that can add significant value across the enterprise,” said Stu Bailey, co-founder and CTO of ModelOp. Bailey continued, “In this new, model-driven world, no single development platform will prevail, as organizations will continually look for the best tool for each model use case. Our mission at ModelOp is to provide our customers with an independent model operations platform that maximizes model velocity, effectiveness and accountability and avoids dev platform lock-in.”
Exos, a next-gen financial technology platform and B2B institutional finance firm, is using ModelOp’s solutions to assist them in enabling clients greater insight, exceptional client experience and transparency across all areas of capital markets. “Model Operations is a critical part of our technology platform to ensure we can scale the use of our data science/machine learning models across our business,” said Joe Squeri, CTO/COO of Exos. “ModelOp has been a trusted partner for us from the start, helping to accelerate our journey with their expertise and software.”
*Gartner, Hype Cycle for Data Science and Machine Learning, 2019, Shubhangi Vashisth, 6 August 2019.
Discussion about this post