Q&A: Globo Director of Product & Engineering Igor Maciel Macaubas Talks Content Delivery, Discovery, and Build vs. Buy
In this wide-ranging conversation with Globo Director of Product & Engineering for Digital Products Igor Maciel Macaubas, we discussed Grupo Globo digital products that deliver content at scale monthly to 90 million unique Brazilian viewers. The main streaming product Globoplay is an ad-supported service which was started ten years ago in 2015. In reaching their audience Globo faces a number of challenges that would test the limits of most streaming services. The company has over 14,000 employees, and Macaubas’ digital technology division accounts for a significant share of this workforce.
Macaubas oversees the development and maintenance of Globo’s entire portfolio of digital products, including the following:
- Globoplay, a subscription VOD streaming service
- Premiere, a premium soccer channel
- G1, Globo’s hard news portal
- ge, a sports portal
- gshow, an entertainment platform
- and Globo.com, Brazil’s largest homepage
Globoplay, Globo’s subscription VOD service
Macaubas has been with Globo for 15 years and says “he’s worked in digital and streaming since day one.”
Our interview has been edited for length and clarity.
To The Cloud (And Beyond)
Nadine Krefetz: Can you talk about your transition to the cloud?
Igor Macaubas: We closed a deal with Google in April 2021, mid-pandemic. The audience tripled, and it was incredibly challenging for us to keep things running. We prioritized migrating edge services, our customer identity management and access management system, our CMS, and the delivery part of our APIs—everything that was connected to delivering Globoplay and all of our other digital products to the internet. After that, we started doing other migrations. The VOD transcoding part of our live packaging came after.
All of our edge CDN is still on-prem. To reach the amount of scale we have is really hard if you’re going to a cloud deployment or a third-party deployment for the edge CDN. There are different challenges with the shape of the internet in Brazil that some cloud vendors can’t grasp: where the backbone is and what are the most propitious places to put points of presence and to do interconnections. [Keeping our edge CDN on-prem] was a critical decision at the end of the day because, otherwise, we wouldn’t have been able to reach the scale that we have today. Some other stuff is still on-prem, like master control for some linear channels and SDI infrastructure. We prioritized migrating everything related to digital and finished that around the end of 2022 or the beginning of 2023.
Krefetz: What was your key takeaway from this transition?
Macaubas: The first thing is you have to be really, really careful with FinOps. We have a very granular view of what every single system costs in terms of cloud. Otherwise, you’re dealing with a huge bill that you have no idea how it came to be that amount. This is one of the main takeaways in terms of governance going to the cloud.
A second lesson is that you cannot just lift and shift to the cloud; you have to redo the architecture to leverage autoscaling and other native functionalities. It’s more efficient, cheaper, and also a better fit to scale if you go with a native cloud solution like a SaaS instead of infrastructure-as-a-service.
Krefetz: You’ve been running your own CDN for a few years.
Macaubas: Yes. Globo built, owns, and operates its own CDN in Brazil. This decision was driven by two factors: the high cost of commercial CDNs for video delivery at scale and the need to control quality across a country with vast geography and uneven internet infrastructure.
Back in 2019, we began deploying our CDN by leveraging Globo’s existing terrestrial infrastructure—a legacy of our TV network footprint. With over 100 geographically distributed facilities—some owned, others operated by affiliates—we had a strong foundation with mission-critical capabilities, like edge data centers. Today, Globo operates one of the largest CDNs in Brazil, with over 270 PoPs and more than 20 terabits per second of capacity. This gives us multiple advantages: We deliver 4K content at no additional cost to subscribers (unlike most competitors who reserve it for higher-tier plans), we maintain control over QoE, and we’ve recently started commercializing excess CDN capacity, further improving our unit economics.
Krefetz: And you’re on the verge of bringing in ATSC 3.0.
Macaubas: This is still undergoing regulatory approval. Every day, we have 30–40 million people tuned to Globo. Imagine the complexity of delivering that over digital streaming internet, unicast, CDN. If you look at the subscription numbers for Netflix in Brazil, they have 20 million subscribers, which is a good number. But we have two or three times that every night on television, by the reach of radio frequency. [The new] DTV+ [standard] will bring streaming and broadcast [together].
Viewing Surges
Krefetz: How do you handle surges in streaming viewers?
Macaubas: Over the past few years, Globo has migrated most of its digital infrastructure to the cloud. As part of this transition, we adopted a “cloudification” strategy, transforming our backend systems and APIs into cloud-native applications capable of autoscaling and leveraging modern cloud functionalities. This architecture allows us to absorb typical traffic fluctuations.
However, some peak events go far beyond what autoscaling can handle—especially Big Brother Brasil (BBB), one of the largest digital engagement events in Brazil. It’s a reality show where contestants live under 24/7 surveillance, isolated from the outside world, with viewers voting weekly to eliminate one participant. The final winner is chosen by the audience in a massive online vote. Engagement happens across multiple touchpoints: live streaming on Globoplay, voting via gshow and Globoplay, and widespread activity across social media platforms.
Big Brother Brasil
The real challenge comes during primetime TV. Because of the massive reach of Globo’s free-to-air TV, when the BBB episode ends—usually around 11 p.m.—we see a massive “migration effect” as viewers instantly shift from broadcast TV to digital platforms. This switch happens within seconds, creating a traffic spike of up to 200 times the regular load across our APIs, voting systems, and streaming infrastructure.
Cloud autoscaling alone cannot absorb this kind of synchronous load. To handle these surges, we’ve invested heavily in our backend architecture—including aggressive load-testing, predictive audience modeling, and performance tuning. Today, our systems can process hundreds of thousands of requests per second, ensuring viewers can launch the Globoplay app, start playback, vote, and interact in real time, even during these peak moments.
It’s a mission-critical operation that requires constant readiness.
Globoplay Premiere premium bundle
Krefetz: How do you minimize latency for tentpole events like the World Cup?
Macaubas: We had to optimize our entire live video pipeline—starting from the stadium capture to our encoding and delivery processes. Traditional broadcast workflows often include components like statmux (statistical multiplexing) encoders, which help optimize bandwidth for satellite delivery but add latency.
We also had to improve our live packaging stack and evolve our video player to support ultra-low-latency playback. As a result, we now deliver live sports content, particularly in our Premiere premium bundle, with end-to-end latency between 5 to 7 seconds. It’s a significant achievement, although it’s currently limited to select sports events.
Globoplay Premiere premium bundle
Merging Broadcast and Digital Workflows
Krefetz: Have you merged some of the broadcast and digital workflows at Globo?
Macaubas: Not yet, but we’re advancing well and integrating as much as possible. One good example was the effort to reduce latency on our live streaming. Another example is evident from our SSAI/DAI work on live linear channels. To make it “broadcast grade,” we had to tightly integrate broadcast and digital workflows.
There are still fundamental differences between traditional broadcast and streaming technologies—from signal transport to encoding, playback, and ad delivery. However, these differences are gradually diminishing with the advent of DTV+ [ATSC 3.0], Brazil’s next-generation free-to-air television standard, which is currently undergoing regulatory approval.
Globo has played an active role in the technical development of DTV+, working in collaboration with industry and regulatory bodies to help shape a more modern and flexible broadcasting framework. One of the key innovations of DTV+ is the adoption of IP-based transport and hybrid delivery, which brings broadcast infrastructure closer to the architecture used in OTT services. This includes support for adaptive bitrate streaming, targeted advertising, and integrated broadband experiences—capabilities that were previously exclusive to streaming platforms.
Once DTV+ is officially rolled out, it will enable a convergence of workflows, opening new possibilities for synchronized broadcast and digital experiences and simplifying the integration of broadcast-grade quality into appbased environments like Globoplay.
The average age of a TV set is between 8 and 10 years old in Brazil. These TVs are very low power in terms of processing and memory. We designed an update of Globoplay in February: a lighter app with less eye candy, with more focus on delivering exactly what the user wants, which is to find content and hit play. Oftentimes, these TV sets don’t even support Netflix anymore, but Globoplay is still working. We saw a huge spike in consumption, both in terms of hours, frequency, and engagement for older TV sets—like 200% growth.
Krefetz: What is the biggest technical challenge as you continue to run two workflows for digital and broadcast?
Macaubas: One challenge was implementing dynamic insertion on the live linear TV channels. We did that a couple of years ago. We had to integrate the digital streaming workflow with the broadcast workflow. On the broadcast side, you have a bunch of tools to do ad insertion, slicing, and splicing. You have ad servers that are running connected to the playout. You have scripts for the breaks. When you are going to streaming and you’re putting SCTE markers on everything, it’s a huge challenge.
We had to do a lot of groundwork, a lot of foundation work in order to integrate these two workflows. If you look at TV Globo, for instance, we have a very dynamic programming grid. We have live events going on, and live events are not of fixed duration. You have to move things forward or back depending on what goes on. Oftentimes, you’re going to cut some prerecorded TV shows, oftentimes, you’re also going to cut some commercials, and all that new script has to go through the ad server, and then it has to go through the SCTE markers as well.
Another big challenge was that streaming has inherent latency because HTTP-based protocols require some level of buffer before starting to deliver video. The typical latency in OTT is often between 20 and 30 seconds.
We had to do a lot of foundation work with the broadcast team in order to modify some workflows on the broadcast side and be able to reduce this latency over there. We were able to reduce the OTT latency between 5 to 7 seconds, which is pretty good.
Another challenge is the regional rights. Globo is one of the main rightsholders for soccer in Brazil, and there are a lot of different rules in terms of blackouts. In 2017, we started doing the linear feed of Globo on the internet. This required us to collaborate heavily not only with the broadcast team, but also with our regional affiliates in order to have the feeds from the affiliates concentrated in our data center and ingested into our CDN packager. We had to develop a mechanism on the front-end app to detect the user location in order to deliver the correct feed and also be able to implement a couple of blackout rules to the linear pay TV channels that were also going on live on Globoplay.
For mobile phones, we’re using GPS information. When the user is watching on TV or on a desktop computer, it’s going to be an IP address. That was a huge challenge that we had to undertake early on in Globoplay.
Krefetz: Do you have digital-only content?
Macaubas: We do. In 2022, we launched FAST channels available through Globoplay and Samsung TV Plus. That was a whole different challenge for us to put together because FAST channels are very different from traditional linear pay TV channels or traditional broadcast channels. We had to update a completely different play-out infrastructure [using] vendors to help us.
Build vs. Buy
Krefetz: When it comes to your tech stack, is Globo a builder or a buyer?
Macaubas: We’re mostly builders. Our product development strategy is centered around retaining control over what’s strategically important to us—especially the user experience and the evolution of our technology stack. At the same time, we actively leverage best-in-class solutions available in the market to accelerate time to market and deliver the best possible experiences to our consumers.
Our fully microservices-based architecture gives us the flexibility to switch, combine, or evolve between custom-built and vendorsupplied components. When we do buy, it’s typically for commoditized parts of the stack, such as DRM systems, video packagers, or analytics SDKs. We also make extensive use of open source technologies.
Krefetz: If you’ve moved from in-house development on specific services to using vendors, what was the biggest reason?
Macaubas: Here’s a recent example. Our search engine for Globoplay was developed in-house [open source Elasticsearch]. We recently discovered that the results and conversion were poor in terms of quality and CTR [click-through rate]. After A/B testing and an MAB [multiarmed bandit], we’ve decided to switch to a vendor. [This] is already delivering meaningful improvements to the consumer experience.
Accelerated AI Evolution
Krefetz: Did you see things at NAB this year that surprised you?
Macaubas: When I was at NAB 2024, I started seeing a lot of stuff related to AI in all points of the workflow. What kind of surprised me at NAB is how fast these workflows and tools have evolved since 2024—dubbing, automated scene description. I’ve seen a solution so cool that you can just ingest a movie or something like that, and you can go through a search box and you can query stuff. Where is this scene where the guy jumps off a cliff on a motorcycle? Then it will go right there to the exact scene where that happens. Automatic generated clipping has been done for the past 5 years, but [this software would] generate a teaser of this content and versions in multiple formats.
Krefetz: Are you using generative AI in any parts of your workflow that you can speak about?
Macaubas: We began exploring this capability early. We’re using gen AI in our software development process, enabling the usage of such tools for both our development and product teams.
We’re also leveraging gen AI for postproduction tasks such as dubbing, subtitling, closed-captioning, and color grading. Additionally, it’s being used to enrich content metadata. Our next step is to integrate gen AI into consumer-facing features. And while specific projects are in development, we’re not disclosing details just yet.
Audience Data and Measurement
Krefetz: What kind of information can you collect on viewers?
Macaubas: Globo has a whole ecosystem of digital products that are used by the same users on different platforms and at different moments of their day. To unify our audience, we’ve created a customer identity platform, called Conta Globo, which provides authentication and
Globo’s G1 news portal
authorization. This platform also offers single sign-on.
We track everything that our users do on our properties, from reading an article on our hard news portal, G1, to watching a TV show or a live linear channel on Globoplay. All those records go into our data lake, which supports both targeted advertising and content personalization, with the goal of offering ads and content that are meaningful for that user.
Globo's G1 news portal
The volume of data we collect is about 10 billion data events per day and is fully compliant with the data processing and protection laws [Brazilian viewers supply an email address, telephone number, and government tax certificate number when they subscribe or purchase content].
Krefetz: Is this different for ad-supported and subscription-based platforms?
Macaubas: No, we collect the same information for all of our users.
Krefetz: Are you using server-side or client-side ad insertion?
Macaubas: We’re using both technologies on Globoplay: SSAI for live linear streaming ad replacement and CSAI for VOD. This allows us to deliver ads with minimal buffering and maintain consistent playback.
Krefetz: What type of measurements do you use?
Macaubas: We have three signals that we constantly monitor and try to find ways to improve:
- Weekly frequency—how many days within the week the consumer is using Globoplay
- Consumed hours—the engagement of each consumer with the platform
- Content diversity—how deep into our catalog that consumer is
These are our main North Star metrics. Consumers that [measure well on these] have proven to be less prone to churn. We also monitor and act on inactivity—if a subscriber is not using the product, it’s very possible that they’ll churn in a bit. The CAC [customer acquisition cost] of reactivating an existing subscriber, and avoiding the churn, is better than the CAC of acquiring a new subscriber.
Krefetz: How many people do you reach?
Macaubas: Globo’s digital products reach around 90 million unique users monthly, rep-
resenting approximately 70% of Brazil’s internet audience. Globoplay alone sees between 25 to 30 million monthly active users. It is the second largest streaming platform in Brazil. Our content consumption (on digital) is evenly split between linear and on demand.
Krefetz: Is there a typical viewing time you can share?
Macaubas: In 2024, Globoplay streamed over 4 billion hours of content. We deliver about 10,000 hours of streamed content every 60 seconds. That’s the equivalent of over 1 million years of non-stop viewing for a single person. More than half of this consumption happens on big screens—connected TVs and dongles.
Globoplay Viewing and Discovery
Krefetz: What is your typical stream resolution?
Macaubas: Full HD 1080p is our standard for most of our content library. We also stream in 4K HDR for linear live sports and premium VOD content. 8K is also available on select content, but the adoption of this is still limited.
Krefetz: Where do viewers consume Globo content?
Macaubas: Most of the consumed hours are streamed on big screens (connected TVs and set-top boxes). This environment is responsible for over half of Globoplay’s streamed hours. The distribution across CTV brands reflects the general market share of each manufacturer in Brazil. Android on mobile phones is also very popular, but the consumption pattern is different: more video views and less time spent.
Krefetz: What are your thoughts on metadata?
Macaubas: We work with three different kinds of content: our owned-and-produced content, licensed content, and live feeds. Live feeds are different because it’s the same pipe with different content going in, and you have to be able to tie all that up. We haven’t resolved that yet, and I don’t think that anyone has.
For linear programming, we have been working with companies like Gracenote and others to keep a good programming grid available. We are using EIDR [Entertainment Identifier Registry]. We are also using some internal crossmapping of IDs to correlate what’s happening on linear. We’ve been working with a couple of partners to help us better deliver ads based on the content that’s going on.
You have to have quality in terms of what [value the] metadata brings and also a semantic understanding. That’s more important to algorithms like search and recommendations than having a huge amount of metadata that’s of no use.
Krefetz: What kinds of tests would you generate to make sure that your metadata is good?
Macaubas: It depends. I’ll give you an example. Let’s say you’re following a soap opera on linear, and you went out on Friday and didn’t watch then. A good search system should be able to return you to the correct chapter. This is not exactly metadata that you are going to see, but it’s something that your search system should be able to handle. We are still working toward that. If you want to have a good recommendation system, you have to balance the customer taste and your content library. You have to have recommendations, collaborative filtering, embedding models, and reranking. Also, you have to be able to ingest events from the behavior of the user and have that as a key input to your recommendation algorithms that will make the user identify with whatever you are returning.
There is a very high risk of you creating a bubble for that consumer.
You also have to create ways to burst the customer out of that bubble. But at the same time, you have to be able to burst the bubble of that consumer in order to increase the retention and the engagement of that customer.
Competing for Viewer Attention
Krefetz: Who are your competitors?
Macaubas: It’s easy to assume we’re competing against other streaming platforms. Today, Globoplay is the number-two streaming platform in Brazil, according to Comscore. But the real competition here is for attention, as the consumer now has a lot of entertainment options, ranging from social networks, video games, UGC platforms, and so on. We compete for time, attention, and relevance in an increasingly crowded digital ecosystem.
Krefetz: Are your viewers only in Brazil?
Macaubas: Mostly in Brazil. Globoplay has an international offering primarily aimed at Brazilian expats, as most of the content is in Portuguese, with availability in countries like Portugal and some Portuguese-speaking regions in Southern Africa. But the majority of our user base and consumption is in Brazil.
That said, it’s worth noting that Globoplay’s content has increasingly crossed borders through international distribution agreements that allow our productions to reach global audiences. Our stories are traveling the world.
Related Articles
Akhil Ramachandran addresses some key challenges he faced when he added low latency streaming support for MediaStore, leveraging the state-of-the-art Chunked Transfer Encoding, and the strategies adopted to solve them.
25 Mar 2025
This article offers a primer on how a traditional broadcaster could start to move services to a cloud provider.
07 Jun 2023
Where exactly is the edge in streaming content delivery? According to leading figures from the Streaming Video Technology Alliance, Amazon Web Services, and Fortinet, defining what edge computing means for streaming varies by use cases that involve factors such as streamlining user experiences, taking security measures, and evaluating data costs.
08 Nov 2022