Signal › Feed

OpenAI Just Launched a Google Business Profile for ChatGPT. What It Means for Every Brand That Wants to Be Found in AI Search.

Deloitte launched Edisyl on November 12 with a simple thesis: enterprise AI doesn't fail because models aren't smart enough. It fails because models don't know enough about the organization they're serving. Here's what that means for every enterprise deployment.


Deloitte's November 12 announcement of Edisyl Inc. — a new enterprise AI company launched under its D71 Ventures portfolio — frames its product around a problem that most enterprise AI deployments never name correctly. The bottleneck in enterprise AI is not the model. It is context.

Edisyl's CEO, Dave Balter — who co-founded Flipside Crypto and moved to Edisyl after Deloitte acquired Flipside's AI business in July 2026 — describes the enterprise AI problem this way: AI doesn't fail because the models aren't smart enough. It fails because the models don't know enough about the specific organization, domain, or workflow they're being asked to serve. The technical name for this failure is the context gap — the distance between what a general-purpose model knows about the world and what it needs to know about a specific enterprise to generate accurate output.

The context gap problem is widespread and expensive. Kyndryl's 2026 AI readiness report found that 78% of enterprises say their AI deployments are delivering less value than expected. Genpact's research found that 85% of enterprises with AI agents in production have not tied those agents' output to measurable P&L outcomes. And Gartner reported that 40% of enterprise AI agent projects are canceled before reaching production. Edisyl's thesis is that the context gap is the single most common cause running through all three failure patterns.

Why Enterprise AI Fails Before the First Prompt

General-purpose AI models — Claude, GPT-4, Gemini, Llama — are trained on broad internet data. They are extraordinarily capable at language tasks. What they do not have, by design, is knowledge of a specific organization's internal structures, terminology, policies, and workflows.

When a healthcare system deploys a general-purpose model for clinical documentation, the model knows what healthcare documentation looks like broadly. It does not know that this health system uses a proprietary billing code taxonomy, that its prior authorization policy changed in September, that the clinical terms its physicians use in notes differ from standard medical terminology, or that its compliance team requires specific language in patient communications. When the model is deployed for clinical tasks without that context, it generates plausible but inaccurate output — wrong codes, outdated policies, non-compliant language. That is the context gap in action.

The same pattern plays out across every vertical. A financial services firm's AI assistant that doesn't know the firm's specific credit policies, risk thresholds, or regulatory interpretations generates recommendations that look reasonable but fail compliance review. An insurance carrier's claims AI that doesn't know the carrier's specific coverage definitions and adjudication rules produces settlements that require costly manual correction. A retail chain's inventory AI that doesn't know the chain's specific supplier relationships, shelf layout constraints, and regional demand patterns generates replenishment recommendations that look algorithmically correct but are operationally wrong.

None of these failures are caused by model capability limitations. They are caused by context gaps — by deploying a general-purpose model on a task that requires specialized organizational knowledge the model was never trained to have.

What Edisyl Builds: The Knowledge Pack Architecture

Edisyl's answer to the context gap is the Knowledge Pack — a portable, structured layer of domain and proprietary knowledge that installs alongside an enterprise's existing AI tools rather than replacing them.

The Knowledge Pack has two components with different ownership and maintenance models.

The Domain Pack is a curated, continuously updated body of knowledge for a specific field or industry — healthcare regulations, financial compliance, insurance underwriting, retail operations management, legal practice areas. It is maintained by Deloitte and by domain experts who actively work in that field. The continuous updating is important: compliance regulations change, clinical guidelines are revised, financial rules are amended. A static domain knowledge layer degrades as the domain evolves; a continuously maintained one stays current.

The Proprietary Pack holds a company's internal terms, definitions, policies, workflows, and data. Critically, the Proprietary Pack stays under the company's control — it is not absorbed into Edisyl's or Deloitte's systems. The company installs it in its own cloud or on-premises environment and maintains control of its own organizational knowledge. This addresses one of the most significant enterprise AI concerns: data sovereignty.

Together, the two packs are designed to be portable across AI tools. Rather than rebuilding enterprise context for each AI tool, organizations install the Knowledge Pack once and use it alongside the AI tools already in their stack — Claude, Codex, Gemini, Microsoft Copilot. The portability claim is central to Edisyl's value proposition: context is rebuilt once, maintained as a managed asset, and carried across tools rather than reconstructed from scratch for each point deployment.

Knowledge Pack ComponentWhat it containsWho maintains itWhere it lives
Domain PackIndustry regulations, standards, terminology, best practicesDeloitte + domain expertsEdisyl-managed, deployed to customer environment
Proprietary PackInternal policies, workflows, terminology, dataThe enterprise itselfCustomer-controlled (own cloud or on-premises)

Why Flipside Crypto's AI Team Was the Right Acquisition

The choice of Flipside Crypto's AI business as the foundation for Edisyl is not an obvious one — blockchain data analytics and enterprise knowledge layers look like different businesses. But the overlap is deeper than the surface.

Flipside built its blockchain analytics products on a core capability: transforming complex, domain-specific raw data into structured, queryable knowledge layers that non-specialists could use. Blockchain data is notoriously difficult to interpret — raw transaction data is cryptographically structured in ways that are opaque to anyone without deep protocol knowledge. Flipside's business was making that data legible and useful for specific audiences: traders, protocol teams, venture investors, analysts. That is exactly the same problem as making organizational knowledge legible and useful for enterprise AI tools.

Flipside sold its blockchain data business to SonarX in May 2026 to focus entirely on its AI work, and Deloitte acquired that AI business in July. The full 24-person team moved to Edisyl, including Dave Balter as CEO. The acquisition gave Deloitte a team with five years of experience building knowledge layers from complex, domain-specific data — not a team that needed to invent the approach from scratch.

The D71 Ventures Model: How Deloitte Is Building AI Companies, Not Just Consulting on Them

Edisyl's placement under D71 Ventures — Deloitte's AI-focused venture arm — is as significant as the product itself. It signals a strategic shift in how Deloitte is positioning its AI capabilities.

Consulting firms have historically responded to new technology platforms by developing advisory practices and helping clients deploy third-party technology. D71 Ventures represents a different model: building and operating the technology companies that solve the problems Deloitte's consulting teams identify at scale. Edisyl is the first company where this model is fully visible: Deloitte identified the context gap problem through its enterprise AI consulting work, acquired a team that had relevant technical capability, and built a company to solve it.

This has distribution implications that matter for the enterprise AI market. IBM and Microsoft's Imagine Enterprise demonstrated that the enterprise AI market rewards companies that can combine catalog depth with existing client relationships. Edisyl has an analogous distribution advantage: Deloitte's consulting relationships with Fortune 500 companies across every major vertical create a natural initial distribution channel for Knowledge Packs. When Deloitte's consultants are advising a financial services firm on AI deployment and the firm has a context gap problem, recommending Edisyl's Knowledge Pack is a natural part of that engagement.

The design partner model confirms this: Edisyl is building its initial Knowledge Packs with Deloitte programs and a growing set of Fortune 500 organizations. The first customers are inside the existing consulting relationship network.

The Four Ways Enterprise AI Fails Without Context

Understanding the context gap in the abstract is less useful than understanding the four specific failure modes it produces in production enterprise AI deployments.

Failure mode 1: Accurate-sounding but incorrect domain output. The model generates output that is grammatically and structurally correct, follows the patterns of the domain, and passes non-specialist review — but contains factual errors specific to the organization's context. A compliance AI that cites the wrong regulatory version. An insurance claims AI that applies the wrong coverage definition. A clinical documentation AI that uses a deprecated code. These failures are expensive because they pass initial quality filters and reach downstream processes or customers before the error is caught.

Failure mode 2: Hallucinated internal policies. When an AI tool is asked about an organization's internal policies without access to those policies, it generates plausible-sounding responses based on how policies typically work in that industry — not how they actually work at that organization. This is not random hallucination; it is systematic context gap. The AI fills the absence of real context with statistically likely context. The result is authoritative-sounding misinformation about the organization's own rules.

Failure mode 3: Terminology misalignment. Organizations develop specialized internal terminology — product names, process names, role names, customer segments, geographic regions — that differs from industry-standard terminology. When an AI tool uses industry-standard terms to describe internal processes, the output is technically accurate but operationally confusing. When it misinterprets internal terms as industry-standard terms, it produces wrong output. Both failure modes are caused by terminology context gaps.

Failure mode 4: Workflow ignorance. AI tools that don't know an organization's specific workflow sequences — which approvals are required at which steps, which exceptions exist to standard processes, which systems receive which outputs — generate recommendations that are correct in isolation but wrong in operational context. A procurement AI that doesn't know the organization's approval hierarchy recommends steps out of sequence. A customer service AI that doesn't know the escalation policy applies the wrong resolution path.

Portable Knowledge vs. Platform Lock-In: The Strategic Architecture Choice

The portability claim at the center of Edisyl's design is a direct response to one of the most consequential strategic traps in enterprise AI deployment: rebuilding context for each AI tool, creating multiple parallel knowledge layers that diverge over time and are expensive to maintain separately.

The alternative most enterprises have pursued — building RAG pipelines on top of internal document stores — tends to produce point solutions. A RAG pipeline built for one AI tool's API does not transfer to another tool's API without reconstruction. As enterprises add AI tools to their stack (a reality in most large organizations, where Microsoft Copilot, Salesforce Einstein, and multiple purpose-built AI agents coexist), the cost of building and maintaining separate context layers for each tool compounds quickly.

The AI agent stack analysis found that the most expensive layer in enterprise AI deployments is not the model cost but the integration and context maintenance layer — the work of making models useful for specific organizational tasks. Edisyl's thesis is that this cost should be absorbed by a managed knowledge layer rather than rebuilt independently by each enterprise and for each AI tool.

The build-versus-buy calculation depends on three factors: how rapidly the domain evolves (a rapidly evolving regulatory domain benefits more from managed maintenance), how specialized the domain expertise needs to be (deep specialization is expensive to maintain internally), and how many AI tools need the same context (portable knowledge has higher leverage in multi-tool environments).

What Edisyl Means for the Enterprise AI Market in 2027

Edisyl is at an early stage — it is building its first Knowledge Packs with design partners and has not released pricing, customer count, or outcome data. But the problem it is addressing is real, the team is experienced in building knowledge layers from complex domain-specific data, and the distribution advantage through Deloitte's consulting relationships is genuine.

The broader market signal is about where the enterprise AI value chain is maturing. For the past three years, the enterprise AI conversation has been dominated by model capability and infrastructure — which model is most capable, which cloud has the best AI infrastructure, which agentic framework is most production-ready. Those questions are still important, but they are increasingly table stakes. The differentiation in enterprise AI outcomes is shifting to the layers that sit between general-purpose models and specific organizational tasks: context, governance, workflow integration, and measurement.

Edisyl is a bet that the context layer is the most valuable of those — that the company which provides portable, maintained, domain-specific and organizational knowledge infrastructure will capture a significant portion of enterprise AI deployment value as the market matures. The analogous bet in the data infrastructure market was that the query and analytics layer would capture more value than the raw storage layer. Databricks, Snowflake, and dbt have validated that thesis. The enterprise AI context layer may follow the same trajectory.

Five Steps for Enterprise Technology Leaders Responding to the Context Gap

The context gap is not a problem Edisyl alone can solve for every enterprise. Many organizations will build internal context layers, work with other vendors, or develop hybrid approaches. The following five-step framework is designed to structure how enterprise technology leaders approach the context gap regardless of which solution they choose.

1. Audit your current AI deployments for context gap failures. Run a structured audit of your production AI deployments, looking specifically for the four failure modes: accurate-sounding but incorrect domain output, hallucinated internal policies, terminology misalignment, and workflow ignorance. The audit should include sampling AI outputs against known-correct answers, reviewing escalation logs for AI-generated errors, and interviewing the users who work with AI outputs daily about where they don't trust the output.

2. Map your context gap by deployment type. Not all context gaps are equally expensive. A context gap in a customer-facing AI that generates wrong compliance language is a regulatory risk. A context gap in an internal productivity tool that uses wrong internal terminology is a nuisance. Rank your deployments by the cost of the context gap failures they're producing, and prioritize context investment accordingly.

3. Decide on the build-versus-buy threshold for each domain. For each domain where you have a context gap, assess the build-versus-buy calculus: how much internal engineering capacity exists to build and maintain a RAG pipeline, how rapidly the domain evolves, and how many AI tools need the same context. Domains that evolve rapidly (regulatory, clinical guidelines) and tools that span multiple AI platforms favor a managed approach. Static internal domains that don't change frequently may be better candidates for internal RAG investment.

4. Assess data sovereignty requirements before any vendor conversation. The most common enterprise AI context gap vendor concern is data sovereignty: does providing context data to a vendor create compliance, regulatory, or competitive exposure? Edisyl's Proprietary Pack model — where the enterprise keeps its own data in its own environment — addresses this directly. Any context layer vendor conversation should start with a detailed data sovereignty assessment: which context data is permissible to share, which must remain on-premises, and which requires specific data handling commitments.

5. Set context quality metrics before deployment, not after. Enterprise AI teams that measure context quality after problems emerge are always reactive. Set context quality metrics at deployment time: accuracy rates on domain-specific tasks, hallucination rates on internal policy questions, terminology accuracy against the organization's internal glossary. These baselines make it possible to detect context gap deterioration as models update, as domain knowledge evolves, or as the organization changes — and to take corrective action before failures reach customers or regulators.

Takeaway: Edisyl's launch is a signal about the enterprise AI market's maturation. The model capability race is real and ongoing. But the deployments that are delivering value in 2026 and that will deliver more in 2027 are the ones that solve the context gap — that give general-purpose models the specific organizational knowledge they need to generate accurate, compliant, and useful output for real enterprise tasks. Whether the solution is Edisyl's Knowledge Pack, an internal RAG pipeline, or some other architecture, the enterprise AI leaders who explicitly address the context gap will outperform those who assume that model capability alone is sufficient. It is not. The context layer is not optional infrastructure. It is the difference between a demonstration and a deployment.

Frequently Asked Questions

What is Deloitte's Edisyl and what problem does it solve?

Edisyl is an enterprise AI company launched by Deloitte under its D71 Ventures portfolio on November 12, 2026. It was formed around Deloitte's acquisition of Flipside Crypto's AI business, which closed in July 2026. Edisyl's thesis is that enterprise AI deployments fail not because the underlying models are insufficiently capable, but because general-purpose AI models lack the specific organizational context — domain expertise, internal policies, proprietary terminology, workflow-specific data — required to generate accurate and useful output for enterprise tasks. Its product is the Knowledge Pack: a portable, structured layer of domain and proprietary knowledge that installs alongside existing AI tools rather than replacing them. The goal is to close what Edisyl calls the context gap — the distance between what a general-purpose model knows and what an enterprise deployment actually needs it to know.

What is a Knowledge Pack and how does it work?

A Knowledge Pack is Edisyl's core product: a portable, structured knowledge layer that sits alongside an enterprise's existing AI tools. It has two components. The Domain Pack is a curated, continuously updated body of knowledge for a specific field or industry — healthcare regulations, financial compliance, insurance underwriting, retail operations — annotated by practicing domain experts and updated as the domain evolves. The Proprietary Pack holds a company's internal terms, definitions, policies, workflows, and data. Critically, the Proprietary Pack stays under the company's control — it is not absorbed into Edisyl's or Deloitte's systems. The Knowledge Pack is designed to be portable: it installs in the company's own cloud or on-premises environment and works alongside AI tools including Claude, Codex, Gemini, and Copilot. The idea is that domain and organizational context is rebuilt once, maintained as a managed asset, and carried across AI tools rather than reconstructed from scratch for each deployment.

How did Deloitte acquire Flipside Crypto's AI business to form Edisyl?

Flipside Crypto was a blockchain data analytics company that pivoted to AI. In May 2026, Flipside sold its blockchain data business to SonarX to focus entirely on its AI work. Deloitte acquired Flipside's AI business in July 2026, and Flipside's full 24-person operating team — including co-founder and CEO Dave Balter — moved to the new entity. Edisyl was publicly announced on November 12, 2026, as a Boston-based enterprise AI company under D71 Ventures, Deloitte's AI-focused venture portfolio. The acquisition gave Deloitte an operating team with deep expertise in building knowledge layers from complex, domain-specific data — the same skill set Flipside had developed in building analytics products for the blockchain ecosystem. Dave Balter brings the perspective that the enterprise AI problem is fundamentally a knowledge and context problem, not a model capability problem.

Which AI tools does Edisyl's Knowledge Pack work with?

Edisyl's Knowledge Pack is designed to be portable across the AI tools enterprises already use, rather than requiring them to adopt a new AI platform. The tools cited in coverage of the launch include Claude, Codex, Gemini, and Microsoft Copilot. This portability claim is central to Edisyl's value proposition: rather than rebuilding enterprise context for each AI tool, organizations install the Knowledge Pack once in their own cloud or on-premises environment, and it works alongside the AI tools already in their stack. This positions Edisyl as infrastructure that reduces the coordination cost of enterprise AI deployments rather than adding another AI tool to manage. However, specific integration documentation and technical architecture details are not yet publicly available, as Edisyl is working with design partners including Deloitte programs and Fortune 500 companies while building out its initial Knowledge Pack catalog.

What is the enterprise AI context gap and why does it cause deployments to fail?

The enterprise AI context gap is the difference between what a general-purpose AI model knows about the world and what it needs to know about a specific organization to generate accurate and useful output for enterprise tasks. General-purpose models like GPT-4, Claude, and Gemini are trained on broad internet data. They know what healthcare regulations look like in general. They do not know that a specific health system uses a proprietary billing code taxonomy, that its prior authorization policy changed in September, that the clinical terms its physicians use in notes differ from standard terminology, or that its compliance team requires specific language in patient communications. When those models are deployed for enterprise healthcare tasks without that context, they generate plausible but inaccurate output — wrong codes, outdated policies, non-compliant language. That is the context gap. It is the most common reason enterprise AI deployments fail to reach production or fail to generate value after reaching it.

How does Edisyl compare to building a knowledge layer internally?

Building an internal knowledge layer is the alternative to Edisyl that most enterprise technology teams have attempted. The internal approach typically involves a RAG pipeline (retrieval-augmented generation) built on top of internal document stores, vector databases, and enterprise search infrastructure. It is expensive in engineering time, expensive in maintenance as domain knowledge evolves, and difficult to extend across AI tools without rebuilding the retrieval layer for each one. Edisyl's Knowledge Pack is a managed alternative: Deloitte and domain experts maintain the Domain Pack, the company maintains control of its Proprietary Pack, and the Knowledge Pack is designed to be portable rather than point-solution. The trade-off is that you are depending on Edisyl for domain pack maintenance and accuracy. The build-versus-buy calculation depends on how rapidly the domain evolves, how specialized the domain expertise needs to be, and how much of the enterprise's AI use cases share the same domain context versus requiring bespoke knowledge layers.