Why the Globe and Mail Bet on AWS

Like most legacy media organisations, Canada’s Globe and Mail spent years wrestling with the transition from print to digital revenue. The paper ultimately settled on a clear strategic direction: grow digital subscriptions and deepen reader engagement, with Amazon Web Services (AWS) as the core technological foundation.

The AWS migration began several years before this reporting (circa 2019) and was driven largely by the paper’s desire to exploit AI and machine learning capabilities at scale — capabilities that were difficult or cost-prohibitive to run on-premises.

AWS Services in the Editorial Workflow

The Globe and Mail adopted a suite of managed AWS AI services to automate and enrich editorial processes:

• Amazon SageMaker — building, training, and deploying custom machine learning models.
• Amazon Comprehend — natural language processing for topic extraction and content classification.
• Amazon Rekognition — image analysis for photo tagging and content moderation.
• Amazon Textract — extracting structured data from documents.

These services feed into automated workflows that tag and categorise content, match articles to reader interests, and reduce manual editorial overhead.

Sophi: The Globe’s Custom Predictive Analytics Platform

The most strategically significant piece of the Globe’s AWS deployment is Sophi, a recommendation engine and predictive analytics platform built in-house and hosted on AWS. Sophi uses machine learning to forecast how much reader interest a story is likely to attract after publication.

That predictive capability directly informs two high-stakes editorial decisions:

1. Paywall placement — which articles should sit behind the subscription wall and which should remain free to maximise reach and conversion.
2. Promotional effort — how much push, newsletter placement, or social promotion a given story deserves.

The Globe has also begun commercialising Sophi, licensing it to other news organisations — following a model similar to the one that inspired it (see Arc Publishing below).

Arc Publishing: Cloud-Native CMS as a Service

The vast majority of the Globe and Mail’s technology footprint now runs on AWS, including its content management system. The paper was the first launch partner for Arc Publishing, the CMS originally developed by the Washington Post for its own newsroom. The Post subsequently began licensing Arc to other news organisations; the platform is hosted through AWS.

Globe and Mail CTO Greg Doufas acknowledged the partnership carried risk at the outset — adopting a CMS that was still being productised by a direct editorial peer. In practice, Arc proved to be a major improvement over the paper’s previous custom-built CMS, delivering the scalability and reliability a modern digital newsroom requires.

Moving from Dashboards to Raw Data

Doufas joined the Globe and Mail through its data science and analytics team before becoming CTO. He now leads a team of roughly 120 people — approximately half developers, UX designers, and data scientists.

When he arrived, the Globe relied on off-the-shelf web analytics products from vendors such as Google and Adobe. The core problem was access to raw data:

“You’d put the tags on [content], and it would tell you as much as you wanted to know about traffic and patterns and traditional web metrics. But everything was just a report. I couldn’t get to the raw data. I couldn’t imagine doing my job here and evolving the business without manipulating raw data. What were my data scientists going to do? Use dashboards and reports?”

— Greg Doufas, CTO, Globe and Mail

The AWS migration gave the Globe’s data science team direct access to event-level data they could query, model, and experiment with — a prerequisite for building platforms like Sophi.

Audience Understanding as Competitive Advantage

Doufas frames the Globe’s analytical capability in explicitly competitive terms:

“Our intimate and scientific understanding of our audience and the content we create is our competitive advantage. No matter how gorgeous the design is … we will leave it on the cutting floor if it doesn’t perform.”

— Greg Doufas, CTO, Globe and Mail

Custom audience research conducted by the paper found that speed, performance, trust, and credibility were the top drivers of reader loyalty — and that page load speed is tightly coupled with perceptions of trustworthiness. Migrating to a scalable cloud infrastructure directly improved the reader experience on both dimensions.

Multi-Cloud Pragmatism

Despite AWS being the dominant provider, the Globe maintains a deliberately multi-cloud posture. Doufas put it plainly:

“Any CTO worth their salt will want to hedge their bets. We will always have a bit of a toe in the water with other [cloud providers].”

— Greg Doufas, CTO, Globe and Mail

He also noted that concentrating spend with a single provider can unlock volume-based pricing incentives — a practical consideration for organisations running AI workloads at scale.

Key Lessons for Media and Publishing DevOps Teams

• Raw data access matters more than dashboards. Managed analytics tools are useful for reporting, but building ML-driven products requires direct access to event-level data streams.
• Managed AI services lower the barrier to ML adoption. Services like SageMaker, Comprehend, and Rekognition let a mid-sized editorial engineering team operate at a level of ML sophistication that would otherwise require a much larger infrastructure team.
• Shared-platform CMS can pay off. Adopting a purpose-built publishing CMS (Arc) developed by a peer organisation reduced technical debt and accelerated feature delivery.
• Performance is a product decision. Page load speed and reliability are not purely infrastructure concerns — they directly influence audience loyalty and, in a subscription model, revenue.
• Commercialising internal tools creates new revenue streams. Both the Washington Post (Arc) and the Globe and Mail (Sophi) turned internal platforms into licensable products, offsetting development costs.