The world’s leading publication for data science, AI, and ML professionals.

Enterprise Data Science Workbench with Marketplace – Ecosystem to Democratize AI

"A centralized workbench and marketplace, the opportunity that exists for enterprises to create a multi-tenant organisation wide ecosystem"

Introduction

Gartner says by 2025, 80% of data and analytics governance initiatives, focused on business outcomes, rather than data standards, will be considered essential business capabilities. The ask is, "businesses should be involved in the end to end lifecycle of AI systems and in the decision making process". Nearly every enterprise has processes for their data science team, which focus on teaching computers and making predictions, often faster and more accurate than humans. Amidst taking these predictions to commercial applications is when they realised that the runway is not equipped to take off.

The best way to bring AI close to the SME’s is to Democratize Ai. This directionally puts the AI tooling industry in a better positioning with all the young players championing the market space. As the mindset is shifting towards "Value with AI and showcasing AI enabled product platforms with greater maturity and scale", enterprises are parking a large size budget towards AI operational toolsets.

AI's VoID is "Value Creation with AI" [Image by Author]
AI’s VoID is "Value Creation with AI" [Image by Author]

Also study by informatica revealed that data science teams spend nearly three fourths of their time building and maintaining the AI services. Last few years, this space has been exploited with technical innovations by industry giants like Google, Apple, IBM, AWS, Uber, Facebook and other enterprise products bringing in unified platforms. The claim now is, data science teams should be able to spend more time building structures and decision models. This era of innovations has brought in specialized platforms and tools with respect to data platforms, model training and development , productionization and serving of AI models.

As per Gartner by 2024, 70% of enterprises will use cloud and cloud-based AI infrastructure to Operationalize Ai, thereby significantly alleviating concerns about integration and upscaling. This peak rise in availability of toolsets brings in challenges around enterprise visibility of toolsets, optimised use of resources, reinvention of model artefacts, inconsistency in business processes and thus still keeps the businesses away from data science teams. Here are the top 3 challenges faced:

Enterprise Challenges to Operationalise AI [Image by Author]
Enterprise Challenges to Operationalise AI [Image by Author]

The Opportunity

The opportunity that exists for enterprises is "Creating a multi-tenant organisation wide ecosystem to operationalize AI". Data science is a team sport and all teams in the enterprise which includes business, analytics, science, engineering and operations needs to be directionally aligned to showcase business value. Business metrics should evaluate the models for fitness and should be enabled with continuous monitoring. Discoverability of existing information pushes the boundaries beyond siloed teams. Existing enterprise-wide systems needs to be taken in conjunction before any AI tool investment, for long term maintainability and cost saving.

"Democratization" is one of the key objectives when leaders talk about the value of AI and AI for everyone. Democratization brings in minimal barriers to embed intelligence in product lines, standardization and reusability, increased adoption of AI by all teams in the enterprise, thus the ability to manage end to end AI workflow.

With this, enterprises purposely forgo "Silos",as it’s the enemy for AI powered growth.

Opportunity to operationalise AI [Image by Author]
Opportunity to operationalise AI [Image by Author]

Foundational Tentacles – Data, People, Process, Product and Platform (D4P)

Enterprise learning suggests that, to create an organisation wide ecosystem, there is "No one size which fits all". The ecosystem needs to be stitched , glued and extended to build a standardized and compliance platform for running production ready AI Use Cases. For enterprises, foundational tentacles for this ecosystem includes data, people, process, product and platform [D4P]. To illustrate let’s take the workflow of a "Fraud detection during new customer registration", which involves

  • [Data] detecting duplicates in seconds comparing with millions of records
  • [People] operator needs to be trained on intelligent features like fraud detection
  • [Process] the business workflow needs to be changed to include live fraud detection as compared to post registration detection
  • [Product]The new customer registration screens needs to be enhanced to embed the AI intelligence
  • [Platform] the AI services which needs to be hosted need to co-exist in the existing product infrastructure adhering to the system performance SLA’s

Thus the foundation of this ecosystem needs to be built with "D4P" in synergy. Design and product thinking remains as a foundational layer to help build the ecosystem that supports requirements from business, tech and operations.

Core Principles

The age of AI now is being ushered by the value served by algorithms. For enterprises, analytics and algorithms are woven into the fabric of firm’s businesses and thus demand the businesses to function differently with the primary focus being governance and life cycle management of AI. Thus, here are the core principles which define the features of this ecosystem.

Core Principles - Data Science Workbench and Marketplace [Image by Author]
Core Principles – Data Science Workbench and Marketplace [Image by Author]

The definition of "Value of AI" is augmentation of human decision making and interactions, as compared to focus on automation. This thought process provides businesses the complete autonomy to own the analytics and algorithms. Businesses define augmentation needs along with data, science and engineering teams. Data and model definitions are governed by measurable KPI’s, identified by the businesses. The engineering and IT teams are onboarded with the change in product and infra requirements, thus enabling seamless integration of AI to the enterprise product suite.

Building Blocks

"Self-service" is an essential characteristic that should be equally applied in this ecosystem. To complement this characteristic, industry leaders and technology vendors have all released drag and drop, no code AI tools to integrate predictions into applications. With this, the cautionary tale which emerges is victim to bias, discoverability and explainability issues. The value against time for self service is illustrated here:

Value of Enterprise AI Marketplace [Image by Author]
Value of Enterprise AI Marketplace [Image by Author]

The strategy here is to build a workbench with a marketplace, thus enabling SME’s, Engineering and Science to work as AI experts. The building blocks for Data science workbench will take into consideration the discoverability, self-service, standardization and overall governance. With access, consume, governance and ownership built-in, the overall block structure is illustrated below:

Building blocks for Data Science Workbench and Marketplace [Image by Author]
Building blocks for Data Science Workbench and Marketplace [Image by Author]

Approach and Strategy towards building a workbench

We discussed earlier that the AI tools and libraries marketplace is flooded with technology vendors introducing new products every day. Typically when there is a surplus of options available, teams get excited to bring in tools of their choices to suit their needs. I would say this approach is a great strategy to bring the best breed of toolsets into the organization. Now with more toolsets, additional challenges around visibility and maintainability develop over a period of time. "A centralized workbench and marketplace address a lot of these challenges and is managed by a central governing body with CDO, CIO, CTO and AI officer."

The fundamental element of the workbench is to build a communication layer to exchange information across these different toolsets and thus teams still have their autonomy to work on the tools/libraries which suits their needs. Concept of marketplace, provides the opportunity to onboard any idea and thus providing the ability to manage the end to end lifecycle which include data, modelling, metrics tracking and model management. Ability to create workflows provides the flexibility to design an end to end process for a data/model taking into consideration the access policies, data tokenization, business approvals, metrics and thresholds for model management and ownership. Open cockpit design inspired dashboards provide high visibility and fine tuning controls, which could be operated by all personas involved. Here are the core capabilities ,

Core Capabilities - Data Science Workbench with Marketplace [Image by Author]
Core Capabilities – Data Science Workbench with Marketplace [Image by Author]

Conclusion

Organizations will need a large number of bespoke solutions to address a diverse set of AI use-cases. In most enterprises, it’s quick to get the initial start-up investment to build initial few models, but over time the expectation is to build a "Bootstrap" kit and a central unified platform for the future. Workbench with marketplace products will be key enablers for "data-centric AI development, which is focused more on business value with comprehensive data quality as compared to data quantity". With workbench, flexible model development and testing environments will be accessible with limited access mechanisms, which enables innovations in AI augmentation.

Butterfield noted that long-term success depends on agility of the collaboration, interdisciplinary, cross sectoral, integrity of the data, accuracy of risk assessments and multi-model way. "Workbench with marketplace product platforms", helps build the culture of businesses owning the AI developments collaborating with science, engineering and operations. Machine learning PaaS solutions will be the focus and remains as a key enabler for businesses adopting AI.


Related Articles