The Challenges of Scaling for AI Transformation

Horizon by QuantumBlack Labs, AI by McKinsey

Projects using artificial intelligence and machine learning (AI/ML) are a key driver of business value. Companies have shifted from exploring what the technology can do to exploiting it at scale to gain market share. At QuantumBlack, AI by McKinsey, we recognize that the complexity involved in building, deploying, and managing AI/ML at scale should not be underestimated.

We help organizations achieve accelerated, sustainable, and inclusive growth with AI. QuantumBlack Labs is the R&D and software development hub within QuantumBlack. We use our colleagues’ experience to develop suites of tools and assets that ensure AI/ML models scale into production and achieve sustained impact.

This article summarizes a recent series of posts about some of the challenges of scaling AI transformations, and the solutions we have built to address them:

• Scaling from prototype to production
• Establishing reuse through assetization
• Combining assets into use cases
• Communicating model insights
• Measuring AI project performance

Challenge 1: Scaling from prototype to production

At McKinsey, we know that turning AI ideas into real, working solutions can be tough.

It’s important to move quickly from a test version to the final product, and if teams don’t follow best practices, they find their projects incur “technical debt,” which means:

• Things break or don’t work well often
• It’s hard to fix and make the code better
• It costs a lot to keep things running
• It takes a long time to finish projects
• There’s less time to create new cool stuff

Solution: Kedro

• An open-source Python framework.
• Promotes the creation of reproducible, maintainable, and modular code.
• Encourages best practices from the start.
• Over 17M downloads and 10K stars on GitHub.
• Used by developers from over 250 companies.

[Read our recent article about Kedro]

Challenge 2: Establishing reuse through assetization

To scale AI effectively, it’s necessary to reuse code, insights, and resources as much as possible. This helps increase time-to-value and boosts productivity. Many companies struggle with this because:

• Their solutions are not designed for reuse.
• There’s no central repository for reusable AI assets.
• Teams are reluctant to reuse code if they didn’t create it themselves.

Solution: Brix

Brix is a tool for assetizing reusable use cases. It has five major offerings:

• Reusable code discovery.
• Telemetry.
• Siloed code consolidation.
• Easy project integration.
• Clear versioning.

Benefits of Brix:

• Faster deployment.
• Lower costs.
• Reduced risks.
• Easier evolution.

[Read our recent article about Brix]

Challenge 3: Combining assets into use cases

Engineering for reuse is a strategic move towards sustainable and successful AI/ML initiatives.

• Repurposing models across customer segments or business units is challenging.
• Requires distinct skills, processes, and tools.
• Teams struggle to organize and manage their code, leading to disappointing productivity and low performance.

Creating reusable assets for AI and machine learning is a smart strategy for long-term success. However, sharing models between different customer groups or parts of a company is tough:

• It needs special skills, processes, and tools.
• Teams often have trouble keeping their code organized and easy to manage.
• This results in slow progress and poor performance.

Solution: Alloy

• A framework for reusable software component development.
• Helps AI teams turn their knowledge into working code.
• Creates a ‘factory’ for AI use cases.
• Allows creation of modular AI components.
• Organizes components into enterprise-grade products.
• Teams can concentrate on solving industry problems instead of making custom tools.

[Read our recent article about Alloy]

Challenge 4: Communicating model insights

To scale an AI/ML project successfully, everyone involved must be able to understand it well to make good decisions. It helps to be able to communicate model insights using a high-quality data dashboard, but some teams face problems because:

• There are not enough skilled front-end engineers or designers to create the dashboards.
• It takes a lot of time and resources.
• No standardized tool kit for dashboards.

Solution: Vizro

• An open-source framework for low-code, high-quality data visualizations.
• Enables data scientists to create dashboards easily, even without design skills.
• Helps share complex information clearly.
• Features a generative AI component for faster insight sharing.
• Over 2.5K GitHub stars and 15K monthly downloads.

[Read our recent article about Vizro]

Challenge 5: Measuring AI project performance

Companies that get the most from digital innovations closely track the value and performance of each analytics project as a business metric. This can be difficult because:

• Careful tracking of AI projects’ value and performance is essential.
• The complexity of AI/ML projects makes this tracking difficult.
• Different tools and no central inventory of use cases add to the difficulty.
• Stakeholders struggle to monitor use cases, spending, and ROI.

Solution: Turo

• A performance management tool for AI projects offered via an accessible web application.
• Provides fast and simple insights-reporting.
• Enables deep dives into individual use cases.
• Tracks KPIs, compliance, and budget issues.
• Enhances communication and builds business confidence.

[Read our recent article about Turo]

QuantumBlack Horizon is a family of enterprise AI products, including Kedro, Brix, Alloy, Vizro and Turo. It addresses the pain points associated with scaling and AI transformation. QuantumBlack Horizon is a first-of-its-kind product suite that helps McKinsey clients discover, assemble, tailor, and orchestrate AI projects.

To learn more about what QuantumBlack Horizon products can do for you, please email Yetunde Dada