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.

getty​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.

Most AI tools are remarkably good at solving individual problems, but without the ability to share information, their effectiveness will be limited. For example, an AI agent helping a sales team may not understand the challenges a customer faces during implementation, which would affect how subsequent sales efforts should proceed. A support agent may not know why a customer received a contractual exception, which could complicate troubleshooting. A legal agent may lack visibility into commitments made during the sales process, which could result in faulty contracts.

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 more organizations deploy tools and agents, the more the complexity compounds. AI often fails because it lacks the full business context.

If the data is fragmented or buried across systems, no amount of AI sophistication can overcome that. This is why so many organizations are turning to context graphs. At its simplest, a context graph connects three critical layers of an organization: data (what happened), knowledge (what it means) and process (how work moves). The graph identifies connections between decisions, workflows, people and outcomes.

For most organizations, building a context graph starts with identifying where context is routinely lost, such as customer handoffs, contract approvals, incident management, employee onboarding and sales-to-implementation transitions. These are the moments where information, decisions and reasoning become fragmented as work moves between teams and systems.

1. Start with one workflow where teams repeatedly ask the same questions, recreate information or lose visibility into why a decision was made.

2. Figure out which information should travel through the process. This might include customer requirements, business risks, approvals, exceptions, commitments or key decisions.

3. Move critical context out of calls, emails, chat threads and disconnected documents. The goal is not just to store information, but to make it findable, connected and usable.

Practically speaking, when a company works to create the knowledge graph, it’s important to remember that it’s less about new infrastructure and more about making existing relationships explicit. Teams can start by connecting their systems that already hold the records they want to access, like their CRM, support ticketing, Slack or other project management tools.

Then, they should review together and agree on what the business needs to recognize consistently across all this data at a macro level. That could include which customer, contract, promise, risk or decision each record refers to, so the company can consistently identify these items across every system. These are the anchors, and every connection should carry its source, so anyone, whether that is a person or an AI agent, can understand where the information came from and whether it is trustworthy.

Lastly, it is critical to remember that context has a shelf life. Delivering yesterday’s answers confidently today is worse than no answer at all.

Most businesses already have the information they need, but lack a means to connect it across systems, teams and workflows. Without that foundation, every new AI tool, agent or automation risks becoming another isolated source of complexity rather than a driver of transformation.

The organizations that create lasting value from AI will do the best job of capturing, connecting and preserving organizational knowledge. Because as AI becomes increasingly capable, the competitive advantage will come from how well organizations understand the information, decisions and relationships that drive their business, as well as how effectively they make that knowledge available to both people and AI.

So, before investing in the next AI tool or agent, leaders should ask a more fundamental question: Does it have access to the context it needs to make the right decision? Getting the foundation right first is what scales AI’s value across the business.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?