Building The 'Context Graph' For Effective AI Deployments

Tom Dunlop is cofounder and CEO of Summize, an AI-powered CLM solution, and a former General Counsel for high-growth technology companies.

​The past decade taught organizations that tools spread faster than systems can evolve. As companies rushed to digitalize workflows, teams adopted software at an unprecedented rate, and often without considering how these tools could work together and share data. Sales bought one platform, finance another and customer success a third. According to IBM, “Only 36% of enterprise tech executives reported that their investments in cloud, data, AI and product engineering are managed as integrated portfolios defined by business objectives and common architecture.”​

When it comes to AI, we see a similar problem. Business-wide teams are deploying copilots, agents and automations to increase productivity. And while they are seeing early gains, much like the SaaS boom before it, adoption is often happening at the team or use-case level rather than as part of a broad, coordinated strategy. That is a problem, because a disconnected AI approach makes it difficult to answer critical questions around governance, oversight and token consumption. It also limits the value that teams can get from AI, as siloed tools and agents cannot work effectively across the wider business.

As a result, AI can often automate individual tasks, but it struggles to run effectively across end-to-end processes.

Why AI Requires A Unified View Of The Business

All this missing information exists somewhere in the business, but rarely in a form that can be easily connected and understood, which means AI outputs often lack the context needed to make confident decisions. Most work does not happen within a single system or department, but instead, moves across teams. Without a way to connect those relationships, critical context gets lost, or misinterpreted, at every handoff.

The Context Graph: Connecting How Work Actually Gets Done

A practical starting point looks like this: