Why product integrations are key for your AI strategy by Shensi Ding (Merge) - INDUSTRY 2025
(Sponsored post) At Industry 2025, Shensi Ding, CEO and co-founder of Merge, delivered a timely message for product leaders who are building AI tools and products. Yes, everyone is talking about AI. But few are talking about the underlying infrastructure that makes AI genuinely useful: integrations.
Shensi focused on the unglamorous but essential work of ensuring that your AI features can access the right customer data and what happens when they can't.
The cost of ignoring integrations
AI is only as good as the data you feed it. That was the central argument Shensi made, and she backed it up with examples from her time at Expanse, a cybersecurity startup. There, the lack of integrations was a key challenge. Customers expected tools that worked seamlessly with ServiceNow, Jira, and Asana. Without those integrations, its product adoption stalled.
When Expanse eventually built those connections, it came at a steep cost. Engineering resources were diverted away from core product work. This pain point led to the founding of Merge, a platform that now supports over 220 integrations and processes over 175 million API requests per day.
AI needs more than a model
Many product teams are still fixated on improving the model or tweaking the LLM. But as Shensi pointed out, even the most advanced model is useless without access to customer data.
Why this matters now
“We're at the start of what analysts are calling the Agentic AI era,” Shensi explains. The stats are hard to ignore:
- Only 1% of enterprise software had agentic AI capabilities in 2024.
- That number is projected to hit 33% by 2028.
AI is already showing up everywhere, but the difference between a feature that drives real value comes down to high-quality data to power AI, Shensi explains. This is why integrations are so important; they are the pipelines that connect all of these different systems into something that your AI can actually use. Without integrations, AI can only give users broad surface-level answers and scrape what it finds on the internet.
Integration is key
With integrations in place, AI can deliver highly personalised experiences. One common use case is enterprise search. A user might ask: “Can you aggregate user feedback on onboarding friction across Intercom and Slack over the past 30 days?”
This kind of responsiveness leads to products that feel more intuitive. Customers use them more often, stick around longer, and rate them higher on NPS surveys.
But it’s not easy
Integrations are hard. Building them takes time, costs money, and often requires formal partnerships. Even once they’re built, you still need to clean the data, maintain syncs, and ensure permissions are respected.
Ding outlined several core challenges:
Choosing what to build first
Another key challenge is prioritisation. Customers will ask for a long list of integrations. But Ding warned against letting that wishlist dictate your roadmap. Your roadmap should remain focused on the core product.
Instead, teams should look for ways to outsource the burden. Merge, for instance, offers a unified API to access over 220 platforms, from ATSs and CRMs to ticketing and file storage. That means your AI can start delivering value without waiting for your team to build bespoke integrations.
What Merge adds
For AI-specific use cases, Merge has added a number of capabilities:
- Normalised data: Makes your vector database more accurate.
- Access to raw and custom data: Useful for complex edge cases.
- Built-in ACL support: Permissions synced in real-time.
- High throughput support: Designed for large datasets.
The takeaway for product teams
If you’re adding AI features to your product, your job doesn’t end at choosing the right model. You need a robust integration strategy.
As Shensi made clear, integrations are a product decision that determines what your AI can see, understand, and do. Ignore them, and you risk building a product that only scratches the surface of what AI can offer.
