Deloitte Bets on the Enterprise Knowledge Layer: Dissecting Edisyl and Its Knowledge Pack
Enterprise AI fails not because models are not smart enough, but because they lack accurate context - that is the entire business logic of Deloitte's new company Edisyl.
On November 12, Deloitte announced the launch of Edisyl, an enterprise AI company. The company emerged from a "split acquisition" completed by Deloitte in July 2026: it bought the AI business of on-chain data firm Flipside Crypto, while Flipside's crypto data business was sold separately to Sonaverse. Flipside's entire 24-person operations team joined Edisyl, with former CEO Dave Balter serving as co-founder.
Among the Big Four consulting firms, Deloitte is not the first to go all-in on AI, but this move deserves a closer look: instead of folding the acquired team into the consulting practice for delivery work, Deloitte placed it in its independent venture portfolio D71 Ventures, building it into a product company that sells the "knowledge layer."
1. The Deal: Buying the Team and the Know-How, Not the Code
| Deal element | Detail |
|---|---|
| Acquisition target | AI business of Flipside Crypto (closed July 2026) |
| Crypto data business | Excluded from this deal, sold separately to Sonaverse Inc. |
| Team | Flipside's entire 24-person operations team joins Edisyl |
| Leader | Dave Balter, Flipside's former CEO, serves as Edisyl co-founder |
| Ownership | Part of Deloitte's D71 Ventures portfolio (focused on AI-native businesses) |
| Headquarters | Boston |
| Commercialization stage | Developing initial Knowledge Packs with design partners, including Deloitte internal programs and a growing set of Fortune 500 organizations across industries |
Balter's track record explains the pricing logic: he has founded eight companies, including BzzAgent (acquired by dunnhumby) and Smarterer (acquired by Pluralsight), led M&A at both acquirers, built dunnhumby's venture arm, and served on its global executive team - a serial founder who knows how to sell companies to large enterprises and how large enterprises buy them. The core team also includes COO Kyle Fugere, CTO Jim Myers, CFO Andrew "Mo" Septimus, and Chief Data Scientist Eric Stone.
2. The Core Product: Knowledge Pack, One Package in Two Halves
Edisyl's prescription for enterprise AI is called the Knowledge Pack, officially defined as "the unit of value" - two halves built together, installed together, and owned by the customer:
| Component | What it contains | Who maintains it |
|---|---|---|
| Domain Pack | Everything outside the enterprise walls: partner data, government and regulatory sources, market data, industry standards and codes | Continuously captured and built by fleets of agents, steered by subject matter experts (SMEs), kept current |
| Proprietary Pack | The enterprise's own terms, definitions, policies, and data | Mapped into the same structure, with data remaining under the enterprise's control |
| Semantic Layer | The foundation underneath: reconciles the vocabulary in enterprise systems with the vocabulary people actually use | Solves the problem that "what the system says" and "what people say" are not the same thing |
Three key deployment choices:
- Installed on the customer's turf: deployed in the customer's own cloud or on-premises, not routed through Edisyl servers;
- Embedded in the customer's existing AI tools: the website explicitly lists Claude, Codex, Gemini, Copilot - it does not sell chatbots, only serves as the knowledge supply layer behind them;
- Portable across systems: the same Knowledge Pack moves with the enterprise when switching models, without being rebuilt in every tool.
3. How Edisyl "Uses Enterprise Data": Treating Context as an Asset
This is the core question: how does enterprise data help enterprises use AI well? Edisyl's answer is not "feed data into the model" but a three-step process:
| Step | Stage | Explanation |
|---|---|---|
| 1 | Locate the gap | Enterprise data lakes contain "facts," but the definitions, regulations, and judgment standards in experts' heads are not in any database - this missing layer is the context gap |
| 2 | Two packs together | Domain Pack fills in public industry knowledge (regulations, standards, markets); Proprietary Pack maps the enterprise's private terminology and data into the same structure, so the AI gets "how your company defines this problem" |
| 3 | Semantic layer alignment | The system field is called patient_status_code, but the business people call it "active patient" - and a company's commercial team simultaneously has three definitions of "active patient." The semantic layer ensures the agent picks the right one |
Balter puts the bet plainly:
"Enterprise AI doesn't fail because models aren't smart enough. It fails because it lacks accurate context. Today, intelligence is widely available. What AI tools can't access is the judgment of the person who has worked a specific problem for 20 years. That's what we set out to package, and it's why the annotation layer matters just as much as, if not more than, the interface."
The website's "What we believe" section, four points, is almost written directly for enterprise CIOs:
| # | Belief | Translation |
|---|---|---|
| 01 | Models will keep getting better, and can pass specialist board exams - but that is not the constraint | No matter how smart the model, it still cannot guess which of your three "active patient" definitions your commercial team uses |
| 02 | Most of what an expert knows is not in your data | It sits in regulations, standards, taxonomies, and codes |
| 03 | Context is an asset, not a prompt | Assets need owners, versions, and maintenance costs - managed like any system of record |
| 04 | The knowledge layer should outlive the model | Context that has to be rebuilt every time you switch vendors was never a "layer" in the first place |
Point four deserves its own callout: this is deliberately written against the model vendors. Switching from Claude to Gemini is fine, the knowledge layer does not move - "anti-lock-in" is both the product selling point and the only reason it dares to compete head-on with Microsoft Graph and Google Workspace native integrations.
4. What This Business Actually Is: The Coordinate System
In one sentence: Edisyl is "enterprise knowledge middleware" - it does not train models, does not build the application layer, it sits between the model and the enterprise.
| Approach | What it stores / retrieves | How it differs from Edisyl |
|---|---|---|
| Data warehouses (Snowflake, Databricks) | Structured raw data | Edisyl does not store data, it stores "the structured version of expert judgment" |
| Traditional RAG (retrieval-augmented generation) | Document fragments | RAG retrieves raw text; Edisyl delivers expert-annotated, versioned, owned semantic knowledge |
| Model vendor enterprise editions (Copilot + Graph, etc.) | Platform-native context | Edisyl's selling point is cross-platform portability and customer ownership - precisely the lock-in pain point of platform solutions |
| Consulting firm delivery (Deloitte's old business) | Person-days | One-time PPTs do not become reusable product assets - which is exactly why D71 spun it out separately |
An easily overlooked background: Flipside's core business was on-chain data analysis, and its core capability was exactly translating highly complex professional data into human-understandable queries and dashboards. This "translation pipeline" maps directly to the enterprise AI scenario: translating domain knowledge into structures that AI can consume. The 24-person team plus Balter's serial entrepreneurship experience - what Deloitte bought was this pipeline, not the codebase.
5. Upsides and Risks
Upsides:
- Built-in world-class distribution channel: Deloitte serves nearly 90% of the Fortune 500, spans more than 150 markets, and has about 500,000 people - every Knowledge Pack built for an industry has Deloitte's own projects as the largest test bed and customer acquisition entry;
- "Annotation layer" positioning: in 2026, as model capabilities converge, the industry consensus is shifting from "competing on models" to "competing on context," and the annotation and governance of enterprise proprietary knowledge is a real demand gap;
- Crypto background revaluation: a Web3 infrastructure data company being absorbed by a Big Four as an AI asset is a signal of "data talent RWA-ization" beyond RWA - Web3 infrastructure talent is now being priced by the traditional enterprise software market.
Risks:
- Still in the design partner stage, no public commercial customer cases, no pricing, no delivery cycle or maintenance cost disclosures;
- How the "agent fleet maintenance + expert oversight" knowledge freshness and correctness is handled - and who is responsible when it goes wrong - is not addressed in the announcement;
- Direct head-on collision with native memory / context features from Microsoft, Google, and OpenAI enterprise editions;
- "Consulting firm doing product" is a repeatedly validated high-failure-rate track; whether the D71 model can give Edisyl the rhythm of an independent product company remains to be seen.
6. Conclusion
Edisyl's judgment can be summarized in one sentence: intelligence is cheap, judgment is expensive. When models can pass specialist board exams, the bottleneck of enterprise AI shifts from "how smart" to "do you know how your company defines the problem." Deloitte transformed a crypto data company's AI team into a knowledge layer company, betting that the moat of next-generation enterprise software is not in the model, but in carefully managed context - context is an asset, not a prompt.
Sources and Verification
| Content | Nature | Source |
|---|---|---|
| Deal structure, team, executives, progress | Verified | Deloitte PR Newswire official announcement (2026-11-12, full text fetched directly this round) |
| Knowledge Pack two-pack structure, semantic layer, deployment, four beliefs, Balter quote | Verified | Edisyl website edisyl.com (fetched directly this round) |
| Chinese flash report | Verified | TechFlow (2026-11-12), consistent with English announcement |
| Comparative analysis, upsides and risks | Author analysis | Inference based on the above facts, not investment advice |
| Undisclosed items | Explicitly noted | Pricing, customer list, delivery cycle, liability terms not publicly disclosed, noted in text |
This article is compiled based on official announcements and public information, for industry analysis only, and does not constitute investment advice or a recommendation for any company.