Jun 16, 2026 · 1h 13m · big-technology
Is Unstructured Data The Key To Successful AI Deployments?
Jitesh Gai · 18m spoken Partha Srinivasa · 17m spoken Alex Kantrowitz · 16m spoken Mike Campbell · 12m spoken
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Hosted by Alex Kantrowitz at Hyland's Community Live, this broadcast explores how enterprises can move beyond stalled AI pilots by combining unstructured proprietary data, domain ontologies, and governed agentic architectures. Through insights from Hyland executives and Erie Insurance leadership, the discussion outlines blueprints for driving tangible business ROI across healthcare, banking, and insurance.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Alex holds 24.4% of the talking time here. How this is scored →
Alex as informed peer 5.0 Guest teaching 4.0 Guest disagreement 1.1 Alex pushing back 1.8
speaking balance: gold is Alex, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 31:37 Pivoting the discussion topic
Mike immediately redirects Alex's question on patient scheduling to focus specifically on referrals where he believes the enterprise context value is clearer.
Hardest push from Alex ▶ 38:59 Challenging model safety and kill switch limitsAlex challenges Mike's security framework by citing instances where frontier models bypassed containment fields and question whether software kill switches can truly reign them in.
Biggest teaching moment ▶ 53:29 Explaining the synergy of structured and unstructured insurance dataPartha educates Alex on regulatory determinism in underwriting and illustrates why unstructured conversational data supercharges rather than replaces structured risk models.
Alex holds their own ▶ 1:04 Demonstrating command of enterprise AI failure ratesAlex sets a strong, analytical tone by citing specific enterprise AI stall examples at Starbucks and Uber alongside industry research showing that 82% of token usage fails to make production.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Alex as informed peer | Guest teaching | Guest disagreement | Alex pushing back | Why |
|---|---|---|---|---|---|---|
| Overcoming AI Pilot FOMO with Business Intentionality | 6 | 3 | 1 | 2 | Alex opens with industry data points regarding high pilot failure rates, Starbucks, and Uber. Jitesh agrees and explains the psychology of corporate AI pilot FOMO. | |
| Proprietary Enterprise Data as the True Competitive Moat | 6 | 4 | 1 | 2 | Alex quotes Larry Ellison on data moats versus commoditized models. Jitesh articulates why graph-linked industry ontologies surpass simple brute-force vector search. | |
| AI Governance Lessons from the Big Data Era | 5 | 4 | 1 | 1 | Alex synthesizes the architecture requirement, and Jitesh draws an analogy to the Hadoop/Spark big data era when unmanaged data lakes degraded into swamps. | |
| Structuring Unstructured Enterprise Data with LLMs | 4 | 5 | 1 | 1 | Alex asks how unstructured data is made actionable. Jitesh explains that 90% of enterprise information is unstructured and LLMs provide the structure needed to extract value. | |
| Automating Document-Centric Business Processes | 3 | 4 | 1 | 2 | Alex questions why content like documents would matter for AI agents. Jitesh clarifies what enterprise content entails in banking and core operational workflows. | |
| Transforming Regulated Industries: Healthcare Case Study | 5 | 4 | 1 | 1 | Alex cites the live keynote demo about emergency triage and referral routing. Jitesh explains how automating document workflows accelerates regulated healthcare operations. | |
| Capturing Patient Context to Free Doctors from Paperwork | 5 | 3 | 1 | 1 | Alex draws on his personal family background with physicians buried under paperwork to highlight clinical data capture. Jitesh affirms the goal of unencumbering knowledge workers. | |
| Building the Content-Powered Agentic Enterprise on ECM | 5 | 3 | 1 | 1 | Alex summarizes that moving past pilot euphoria requires architectural rigor. Jitesh emphasizes how enterprise content management governance extends naturally to agentic systems. | |
| Introducing Hyland's Product Strategy for Operational AI | 4 | 3 | 1 | 1 | Alex introduces Mike Campbell to discuss product implementation in clinical and billing workflows. Mike outlines how Hyland focuses on operational and clinically adjacent tasks. | |
| Accelerating Referrals with the Enterprise Context Engine and Ontologies | 4 | 5 | 2 | 2 | Alex asks about scheduling, but Mike pivots to specialist referrals to demonstrate the Enterprise Context Engine and ontologies in cross-system data retrieval. | |
| Establishing AI Trust: Agent Passports, Control Towers, and Kill Switches | 6 | 4 | 2 | 5 | Alex challenges Mike on AI trust and containment vulnerabilities, citing Anthropic's red-teaming examples. Mike describes agent passports, control towers, and kill switches. | |
| Banking Automation: Multi-Agent Mesh in Indirect Lending | 4 | 4 | 1 | 1 | Alex asks about banking applications. Mike explains indirect lending workflows and how an agent mesh coordinates fraud checks, completeness, and underwriting creditworthiness. | |
| Elevating Knowledge Workers and Automating Public Benefits | 5 | 4 | 1 | 3 | Alex asks about workforce impact and whether frontier base models might render specialized software obsolete. Mike argues custom vertical workflow software remains indispensable. | |
| Erie Insurance's Scale and the Reality of Insurance Analytics | 6 | 4 | 1 | 1 | Alex welcomes Partha Srinivasa and references Bezos's famous comments on insurance automation. Partha details Erie's scale and reframes insurers as long-standing data and analytics companies. | |
| Supercharging Deterministic Underwriting with Unstructured Intelligence | 6 | 5 | 2 | 4 | Alex plays devil's advocate, arguing structured actuarial tables already work effectively. Partha explains that unstructured conversational signals supercharge deterministic underwriting. | |
| Real-Time Agentic Co-Pilots in Claims and Subrogation | 5 | 4 | 1 | 1 | Alex digs into claims processing, and Partha explains how real-time co-pilots assist adjusters in collecting contemporaneous evidence needed for subrogation and fraud detection. | |
| Delivering Empathetic Customer Service with Human-in-the-Loop AI | 6 | 4 | 1 | 4 | Alex asks whether automated inconsistency tracking will be weaponized to deny claims. Partha reframes the system around customer empathy, immediate assistance, and human-in-the-loop validation. | |
| Orchestrating Workflows and Managing Time-Limit Legal Demands | 5 | 4 | 1 | 1 | Alex relates personal frustrations with missing claim documents. Partha explains how AI detects time-limit legal demands in incoming scans to prevent bad-faith liability. | |
| Erie's Unified Content Architecture with Hyland and Agentic Roadmap | 4 | 4 | 1 | 1 | Alex inquires about Erie's technology partnership with Hyland. Partha explains consolidating multi-vendor content into a single repository to feed agentic workflows. | |
| Controlling Token Costs with an AI Business Value Office | 5 | 4 | 1 | 1 | Alex asks for the core takeaway on controlling pilot waste. Partha details Erie's AI Center of Excellence and AI Business Office that monitors token expenditures against clear business ROI. |
Statements from this episode (22)
Ghai: AI pilots launched to cure FOMO are bound to fail
“The psychological element is, you know, sometimes it's easier to do a pilot and cure your FOMO, your fear of missing out. Make yourself feel better that you're doing something. Those are bound to fail because nobody's earnestly trying to engage with this trans…”
Ghai: Proprietary enterprise data is AI's only truly novel asset
“I think what is truly novel and unique Is an enterprise's data. It's their data. It is a living record of their enterprise, its decisions, its actions, its operations, and through it, you can get, you can realize enterprise context for that bespoke organizatio…”
Ghai: Brute-force vectorizing of unstructured enterprise data causes AI hallucinations
“Let's vectorize petabytes of data and hope we can figure it out. That's when hallucinations happen. That's when other issues happen. So there needs to be some structure around it.”
Ghai: Enterprise AI agents require role-based access controls
“Not everybody in the organization has access to this content and data, and not every agent in the organization should have access to it as well. So, the concept of agentic roles mapping to access rights so that they can make the decisions and take the actions …”
Ghai: 90% of Global 2000 enterprise data is unstructured
“It's not entirely well appreciated that 90% of an enterprise, the global 2090% of their enterprise data is actually unstructured.”
Ghai: All document-centric tasks in mission-critical processes can be automated
“There's a lot of document-centric tasks involved in, in these very valuable and mission-critical business processes, and all of these can be automated.”
Ghai: Automating document workflows speeds up hospital emergency intake and referrals
“If we can automate all of these document centric tasks at machine speed, patients get onboarded and assigned a bed in the emergency department much faster. Nurses are freed up to deliver care with more of their time. Referrals can happen at machine speed so pa…”
Campbell: Specialist medical referrals often take weeks due to administrative friction
“In many cases when you're getting referred to a specialist, That can take weeks, weeks and weeks, and if you've got an ailment that needs to be treated, that's just wasted time.”
Campbell: Hyland requires agent passports before deploying AI agents
“Every agent in the Highland system has what we call an agent passport. This is basically the declaration of what this agent can do, what its role is, what information it can access, what privileges it has. So all of that is defined and nothing gets deployed in…”
Campbell: Enterprise AI architectures need instant kill switches for agents
“The other advantage of the agent control tower is that it provides what we call kill switches. So If you don't want an agent doing anything anymore, you can immediately shut it down, and it will completely stop.”
Campbell: Multi-agent meshes can orchestrate complex enterprise lending workflows
“And that's this idea of an agent mesh, right? This isn't just one agent doing one thing. It's a collection of agents working together, checking for completeness, assessing fraud, checking for credit worthiness and assessing what are the available loan products…”
Campbell: Government benefits eligibility and verification can be automated
“We do a ton around government funded benefits eligibility.
The, you've lost your job, unfortunately.
Now you're applying for unemployment insurance.
Somebody is assigned your case.
Somebody is checking all of your information.
Somebody is determining maybe you…”
Campbell: AI agents coding complex enterprise systems from scratch is decades away
“Yeah, I think we're a long way from Any individual company just saying, agent, go, go, go code me my healthcare revenue cycle system from scratch, right? The, could we get there eventually? I think maybe we could, but I think that that's decades away, well bey…”
Srinivasa: Insurance companies have used AI and predictive algorithms for decades
“Insurance companies have been using AI for many years, for decades. When we look into our actual department, the way how we price it, the way how we underwrite a risk, it's based on thousands of different data points what we take and we apply a lot of algorith…”
Srinivasa: Insurers will never replace structured data with unstructured data
“I don't think that we will ever replace our unstructured, our structured data the way how we are doing it. We're going to supercharge the structured data with unstructured data because we are highly regulated”
Srinivasa: Erie uses background agentic AI to prompt claims adjusters
“In near real time, this is an AI agent, call it an agentic AI, which is running behind the scene, listening into every conversation about your claim. And it is acting upon it and prompting you to do certain things.”
Srinivasa: AI will automatically pull police records for insurance claims workflows
“Because what is going to happen is the content intelligence is going to create a set of workflow. What we call as orchestrate a workflow to do certain things to collect information. In many cases, insurance company will automatically go pull that information f…”
Srinivasa: AI tools will parse legal demands to prevent bad-faith lawsuits
“The power of the tools now which are available is going to automatically look into it and say there's a time limit demand. If you settle it now, you're done with 5000 dollars because that's what the customer feels is the right number. If not, the customer is g…”
Srinivasa: 80% to 90% of Erie Insurance's data is unstructured
“80 to 90% of our information are on unstructured data, and then structured data is what we use primarily for a lot of things.”
Srinivasa: Erie Insurance will always keep humans in the AI loop
“Insurance is a trust business. We will always have a human in the loop, ok? Because trust is built based on human intelligence and et cetera, and we want somebody to validate it.”
Srinivasa: Erie Insurance created an AI office to monitor token usage
“What we are setting up in our organization is there is an AI center of excellence and AI business office, similar to how cloud used to have a cloud business office whose primary responsibility to watch how people are picking up compute on a self-service mode. …”