AI is moving fast, and the companies staying ahead are the ones building research into their product strategy.
At Infobip, AI isn’t just a buzzword – it’s how we make messaging smarter, personalization sharper, and users safer. And we’re not the only ones. The top tech players are showing that real innovation comes from staying curious: sharing what they learn, giving teams room to explore, building strong foundations, and teaming up with the academic world.
Here’s what we’ve learned from them and how we apply it.
Tie research efforts to business goals
Criteo’s AI Lab was designed to drive real impact across its ad tech platform. The lab works on product-integrated AI (e.g., scalable recommendation engines) while staying connected to state-of-the-art research through publication and collaboration.
Amazon’s AGI Lab follows a similar strategy by developing intelligent agents whose capabilities are directly aligned with Amazon’s long-term business goals like automating web-based tasks and enhancing customer experiences through AI-driven services.
Takeaway: Align R&D teams with product outcomes to make sure research delivers a measurable value.
Foster a culture of open research and knowledge sharing
Top tech companies know that publishing is good business. It attracts talent, builds credibility, and ensures your research is tested and improved by peer review.
Microsoft Research has operated like an academic institution since 1991. Researchers are encouraged (you could even say expected) to publish openly, often co-authoring with university labs and contributing to conferences like NeurIPS, IUI, or ICLR, among others
Google Research supports both product-aligned and foundational AI research, with teams regularly publishing hundreds of peer-reviewed papers annually.
Meta AI runs open research units like FAIR, where scientists are free to publish and collaborate externally.
Takeaway: Invest in processes that support publication. Publishing boosts brand visibility, validates technical depth, and strengthens recruiting.
Make space to explore
Breakthroughs often begin with unlikely ideas.
At X, the Moonshot Factory, Google encourages teams to tackle “10x problems”. Such ambitious projects like Project Loon and Waymo started as wild experiments and became real businesses.
DeepMind’s AlphaGo wasn’t tied to a product roadmap. It was a deep research challenge that led to advances in reinforcement learning, and now powers Google’s data center cooling optimization.
Takeaway: Make room for structured exploration. Create space to experiment, then give teams pathways to scale what works.
Build scalable infrastructure
Infrastructure turns ideas into impact.
The introduction of LinkedIn’s Pro-ML platform enabled ML model development, monitoring, and deployment across various features (e.g., content feeds, talent matching, and messaging systems, among others).
Similarly, Netflix built Metaflow, a human-centric framework to simplify end-to-end machine learning workflows which enables scalable experimentation and production deployment for their recommendation and personalization systems.
Takeaway: MLOps isn’t optional. Build infrastructure that supports reproducibility, monitoring, and reliable deployment.
Build with ethics in mind from day one
Responsible AI is no longer optional, rather, it is expected.
Anthropic built its company around “Constitutional AI”, which is a set of rules that guide model behavior toward human-aligned responses, transparency, and safety.
Similarly, Google’s AI Principles help internal teams evaluate every model for fairness, privacy, and social impact before deployment.
Takeaway: Integrate ethics into R&D tooling and workflows early – don’t retrofit.
Use conferences as strategic input
NVIDIA Research treats AI conferences like NeurIPS, ICLR, and ICML as two-way streets. They not only share breakthroughs, but also scout for collaborators, trends for new research, and hiring prospects.
Another example comes from Hugging Face’s Science Team which actively engages with the academic and open-source communities through conference participation and paper submissions. They use events like ACL and EMNLP to shape both research direction and community-driven innovation.
Takeaway: Don’t just present – listen. Use conferences as feedback and inspiration loops for your research roadmap.
Work closely with academic researchers
The MIT–IBM Watson AI Lab demonstrates how long-term academic partnerships can create a dual impact, i.e., advancing foundational science while solving industrial challenges.
Another example comes from Spotify Research which reports how they actively work to bridge the gap between academic-style research and product development through embedded collaborations and cross-functional research initiatives.
Takeaway: Don’t just recruit researchers – build ecosystems with them. Academia is your best ally for long-term innovation.
Infobip’s AI R&D formula
At Infobip, AI R&D is tightly integrated into product development using the UNFIX framework, with a focus on messaging, personalization, and user protection.
We encourage open research through publications (e.g., on Voice AI and HCI) and academic collaborations (e.g., with FIPU and TUG) as well as support exploration via hackathons and research spikes to bridge science and real-world application. We’re building not just models, but an ecosystem of learning, exploration, and impact.
Into AI research? Come see what Infobip has to offer.