BIDeck Notes

Snowflake Summit 26 · BI233 · speaker-notes deck

Deck notes you can rehearse from, not just read.

A phone-first HTML version of the BI233 deck with speaker notes, slide-by-slide talk tracks, pacing cues, and the phrases worth keeping under pressure.

The value is governed connectivity, not centralization. PE tax data becomes useful when the model preserves how funds, entities, investors, jurisdictions, legal terms, allocations, source evidence, and risk signals relate.

Phrases to keep in mind

These are the lines that make the talk crisp. If you blank, pick one of these and rebuild from there.

Thesis

“The data was always connected. The work was connected. The risk was connected. The problem was that the systems were not.”

Use on Slide 7 or 13. This is the cleanest version of the whole point.
Anti-generic

“We weren’t trying to build one giant database. We were trying to model how the tax business actually works.”

Use before architecture/CDM so Snowflake sounds like operating model, not plumbing.
AI guardrail

“AI is only useful here if it can reason over governed context and trace the answer back to source evidence.”

Keeps the AI section disciplined and non-hypey.
Close

“The chemistry isn’t in the data by itself. It’s in the relationships — the bonds between the data.”

Use as the landing line. Do not rush it.

18-minute pacing rail

Your audio run-through was about 26:55 at ~179 wpm. This version is built to force sharper cuts and leave room for handoff/Q&A.

0:00–2:00 · Orient

PE tax is layered by design. A dollar can pass through many entities and jurisdictions before it reaches an investor.

2:00–4:00 · Mismatch

The business relationships are connected, but the tools are siloed: CRM, accounting, entity management, legal documents, fund-admin files.

4:00–6:00 · Outcomes

Transparency, confidence, and risk-driven decisioning. The platform goal is better-targeted tax review.

6:00–9:00 · Operating change

Known issues get caught early by DQ gates. Unknown risk gets prioritized by scoring and patterns.

9:00–12:00 · Architecture

Raw to cleanse to CDM to consume. Governance, lineage, and meaning travel with the data.

12:00–15:00 · Functional areas

Heart of the talk: entities to allocations, legal terms to tax outcomes, source events to risk signals, curated data to client systems.

15:00–17:00 · Analytics + AI

Anomaly detection and dashboards are practical today; document extraction/NLQ become stronger as the governed model matures.

17:00–18:00 · Close

Connected models outlast workflows. Governance travels with the data. The chemistry is in the relationships.

Slide-by-slide deck notes

Each card has the deck content, what to say, the transition into the next slide, phrases to keep, and what to cut if timing gets tight.

2

Title · Data Relationship Chemistry

Claire Callahan + Michael Tran · EY
0:00–0:30

Deck says

  • How EY is Building a Financial Services Intelligence Platform on Snowflake
  • The Art of Data Relationship Chemistry

Say

“The title is relationship chemistry because the value is not just storing financial-services data. The value comes from understanding the relationships inside that data well enough to support review, risk, and future AI workflows.”

Transition out

“That’s the theme I want to keep coming back to: the value is not just the data, it’s the relationships inside the data. So let’s start with the operating challenge before we get into the technology.”

Job: title → agenda. Make the talk feel like a story, not a slide tour.

Keep in mind

Set the theme early: chemistry = relationships, not AI sparkle.

Cut if tight

No long intro. Get to the challenge quickly.

4

Agenda

Challenge → outcomes → Snowflake foundation → AI in practice → three lessons
0:30–1:00

Deck says

  • Fragmented data landscape
  • Approach and outcomes
  • Data foundation on Snowflake
  • AI from concept to capability
  • Three things that mattered

Say

“I’m going to keep this grounded in the operating problem: PE tax data is already connected in the business, but historically it has not been connected in the systems.”

Transition out

“To make the platform choices make sense, it helps to start with why this data is different in the first place.”

Job: agenda → PE tax complexity.

Keep in mind

Promise a case story, not a product tour.

Cut if tight

Do not read the agenda. One sentence only.

6

Why PE tax data is different

Layered funds, entities, investors, jurisdictions, and tax events
1:00–2:30

Deck says

  • GP manages the fund
  • LP investors put capital in
  • 200+ entities: blockers, feeders, SPVs
  • Portfolio investments receive capital out
  • Every layer creates tax work; every investor is different

Say

“Private equity tax data is hard because the business itself is layered. Investments sit inside entities, entities roll up through funds, funds have investors, and each investor may have different tax attributes, jurisdictions, agreements, and expectations. A single dollar of income can flow through multiple entities before it reaches an investor, and its tax treatment can change based on that path.”

Transition out

“So that’s the business reality: layered structures, different investors, different jurisdictions, and tax treatment that depends on the path through the structure. The problem is that the systems supporting that work usually don’t preserve those relationships cleanly.”

Job: complexity in the business → fragmentation in the tools.

Keep in mind

Use this to orient non-financial-services people. Do not teach PE 101 for five minutes.

Cut if tight

Skip detailed GP/fund/entity mechanics after the audience has the mental model.

7

The data is connected. The tools are siloed.

Fund structures, ownership changes, tiered tax risk, spreadsheets
2:30–4:00

Deck says

  • Entities connected through fund structures
  • Ownership shifts with each capital call
  • Tax risk propagates through tiers
  • Reviewers hunt through file systems and spreadsheets
  • No cross-jurisdiction view

Say

“The data was always connected. Capital calls change ownership. Ownership affects allocations. Entity structure affects tax treatment. Legal terms can change investor outcomes. But the work was spread across accounting systems, entity systems, CRM, file shares, fund-admin spreadsheets, and local-country processes.”

Transition out

“Once you see that mismatch, the design question changes. It’s not just, ‘How do we move data into one place?’ It’s, ‘What outcomes do we need the platform to produce for tax teams and clients?’”

Job: fragmented systems → outcome-led design.

Keep in mind

This is the first big thesis slide: business connected, systems fragmented.

Cut if tight

Use one concrete example: ownership changes with a capital call, then move on.

8

Target outcomes

Transparency · confidence · risk-driven decisioning
4:00–5:30

Deck says

  • Transparency: one connected view with lineage
  • Confidence: 167 quality rules catch relationship breaks
  • Risk-driven decision: scoring highlights material risks
  • Goal: faster, better-targeted tax review

Say

“Before we designed the architecture, we aligned around outcomes. We needed transparency across the business, confidence in the data through quality checks and lineage, and better decision-making by focusing reviewer attention where risk is highest.”

Transition out

“Those outcomes became the principles for the build. If we wanted transparency, confidence, and better decisions, the platform had to carry lineage, quality, risk signals, and business context all the way through.”

Job: outcomes → design principles.

Keep in mind

Architecture choices should sound driven by business outcomes.

Cut if tight

Do not explain all before/now copy. Say the three outcomes and transition.

9

Built to principles, not just to spec

Compliance, risk management, elevated human focus, knowledge capture, data-driven process, improvement
5:30–6:45

Deck says

  • RAW → CLEANSE → CDM → CONSUME with lineage and DQ
  • Materiality rules, anomaly scoring, reviewer dashboards
  • CDM encodes fund/entity/investor/jurisdiction relationships
  • Traceability and modular AI services

Say

“These principles kept us from building technology for technology’s sake. Compliance required lineage and DQ gates. Risk management required materiality rules and dashboards. Knowledge capture required a common data model that preserves the relationships between funds, entities, investors, and jurisdictions.”

Transition out

“That can sound abstract, so let’s make it practical. Imagine a busy quarter where fund-admin data is coming in and reviewers are trying to decide what needs attention.”

Job: principles → day-to-day reviewer workflow.

Keep in mind

Pick three principles, not six, unless you have time.

Cut if tight

Do not say “I don’t know if I want to go through all of these.” Choose your path confidently.

11

A busy quarter. Two signals.

Without connected model vs with connected model
6:45–9:00

Deck says

  • Without: wait for fund-admin data, reconcile 200+ partner percentages, cross-check capital accounts, catch errors during K-1 review
  • With: DQ catches ownership totals 100.7%; scoring elevates Fund II for review

Say

“This is where the operating model changes. Without a connected model, each reviewer is rebuilding context in spreadsheets. With the platform, known issues are caught as data moves through the gates — for example, ownership totals that do not equal 100%. Then scoring helps reviewers find the less obvious risk: the fund or entity that deserves heightened attention during a busy quarter.”

Transition out

“That reviewer experience depends on the architecture underneath it. To catch issues earlier and focus attention intelligently, we need lineage, quality checks, and a common model that travel with the data.”

Job: operating example → Snowflake foundation without making it plumbing.

Keep in mind

Known failures = DQ. Unknown/material risk = scoring + human judgment.

Cut if tight

One example each: 100.7% ownership and busy-quarter pattern.

12

Four governed layers on Snowflake

RAW · CLEANSE · CDM · CONSUME with DQ gates
9:00–10:45

Deck says

  • Raw intake
  • Cleanse
  • Common Data Model
  • Consume
  • DQ gates between layers

Say

“Each layer solves a tax-data problem, not just a data-engineering problem. Raw preserves source evidence. Cleanse standardizes and validates. The CDM defines what each data point means and how it relates to the business. The consume layer gives reviewers, dashboards, client systems, and AI services governed access to the same trusted context.”

Transition out

“But the most important point is not that the data moves through layers. The important point is what survives those layers: the relationships. This is the heart of the talk.”

Job: architecture → centerpiece thesis. Slow down here.

Keep in mind

Governance travels with the data. Lineage is not a side document.

Cut if tight

Do not over-explain medallion architecture. Tie it back to trust, lineage, and business meaning.

13

Connecting functional areas, not just files

The heart of the talk
10:45–13:45

Deck says

  • Trial Balance & GL
  • Investor Master
  • Capital Activity
  • Legal & Contractual
  • Portfolio Investments
  • Tax Filing History
  • One connected model: entities ↔ allocations, legal terms ↔ tax outcomes, source events ↔ risk signals, curated data ↔ client systems

Say

“This slide is the center of the story. Centralized means the data is in one place. Connected means the model understands how the business works. Entity structure validates allocations. Legal terms explain tax outcomes. Portfolio events become risk signals. Tax filing history gives us comparability across years and jurisdictions. The same governed model can support preparation, review, audit sharing, client systems, and natural-language querying without rebuilding the business logic each time.”

Transition out

“Once those relationships are modeled, the analytics become much more practical. We’re no longer just asking, ‘What does this field say?’ We can ask, ‘Does this treatment make sense given the investor, the entity, the jurisdiction, the agreement, and the history?’”

Job: connected model → analytics and AI use cases.

Keep in mind

Slow down. Say: “The value is governed connectivity, not centralization.” This is the line people should remember.

Cut if tight

Skip naming every functional area. Use the four relationship pairs and move on.

14

Analytics dashboard in practice

Anomaly detection · document extraction · NLQ · ML reconciliation
13:45–15:30

Deck says

  • In production: anomaly detection for investor withholding
  • Emerging: document extraction from agreements and side letters
  • Roadmap: Cortex Analyst over CDM; ML-powered reconciliation

Say

“The practical use cases follow from the model. For investor withholding, anomaly detection can flag when one investor is being treated differently from similar investors — not to declare it wrong, but to focus human review. For documents, LLM extraction becomes useful when extracted terms land in governed fields connected to entities, investors, and tax outcomes. For natural-language querying, the answer has to be traceable back to governed data and lineage.”

Transition out

“And the common thread across those examples is focus. The goal is not to automate judgment away. The goal is to focus human judgment where it matters most.”

Job: analytics examples → risk-driven review.

Keep in mind

AI is a payoff of the foundation, not the protagonist.

Cut if tight

Use withholding as the concrete example; mention document extraction and NLQ as future-facing capabilities.

15

Risk-driven review in practice

Multi-factor scoring surfaces entities that need attention first
15:30–16:30

Deck says

  • Rules catch the known
  • Scoring prioritizes where to look for the unknown
  • Dashboard focuses review on higher-risk filings/entities

Say

“The goal is not to remove judgment. It is to aim judgment better. If a team has a hundred filings to review, the platform helps identify which five may deserve the sharpest attention based on materiality, anomaly signals, historical outcomes, and business context.”

Transition out

“So stepping back from the dashboard, there are three patterns from this work that are worth carrying forward.”

Job: specific dashboard workflow → broader lessons.

Keep in mind

Human attention moves to where it matters. That is the business value.

Cut if tight

Do not explain every KPI. Say what the dashboard changes about reviewer behavior.

16

Three patterns to build on

Connected models · governance travels · AI foundation
16:30–17:30

Deck says

  • Connected models outlast any single workflow
  • Governance travels with the data
  • AI gets useful when the foundation is right

Say

“The three lessons are simple. Connected models outlast individual workflows because the relationships are reusable. Governance travels with the data because lineage and quality are built into the platform. And AI becomes operational only when it is grounded in connected, validated, governed data.”

Transition out

“Connected models outlast individual workflows. Governance travels with the data. And AI becomes useful only when it is grounded in that governed context. That brings us back to the chemistry idea from the title.”

Job: lessons → closing chemistry line.

Keep in mind

This is the recap. Keep it clean and confident.

Cut if tight

One sentence per pattern.

17

Closing chemistry line

Land the session title
17:30–18:00

Deck says

  • The chemistry isn't in the data.
  • It's in the depth of relationship connections between the data.

Say

“The title of this session is about data relationship chemistry. For us, the chemistry is not just the data. The data points are the atoms. The value comes from the bonds: how entities connect to allocations, how legal terms affect tax outcomes, how source events become risk signals, and how all of that can be traced through governance and lineage.”

Landing cue

“The data points are the atoms. The value is in the bonds.”

No new content after this. Pause, then hand off or close.

Keep in mind

Pause before the final sentence. Let it land.

Cut if tight

Do not add new details. Close.

xAI clean narration audio

Listen through once to hear the cleaner pacing, or play individual slides before rehearsal. Total combined run: about 6:55.

Download full MP3 Download slide ZIP Open script

Full clean narration

All slide narrations stitched together.

Slide 2 · Title

Data Relationship Chemistry.

Slide 4 · Agenda

Ground the session in the operating problem.

Slide 6 · Why PE tax data is different

Layered structures, investor attributes, and path-dependent tax treatment.

Slide 7 · Connected data, siloed tools

The business was connected; reviewers rebuilt context manually.

Slide 8 · Target outcomes

Transparency, confidence, and risk-driven decisions.

Slide 9 · Principles

Compliance, risk management, and knowledge capture drive architecture.

Slide 11 · A busy quarter

DQ catches known failures; scoring points to material unknown risk.

Slide 12 · Governed layers

Raw, cleanse, CDM, consume — governance travels with data.

Slide 13 · Functional areas

The heart of the talk: governed connectivity, not centralization.

Slide 14 · Analytics + AI

Withholding anomaly detection, document extraction, and traceable NLQ.

Slide 15 · Risk-driven review

Aim judgment better; focus attention where it matters.

Slide 16 · Three patterns

Connected models, governance, and AI-ready foundation.

Slide 17 · Closing

The chemistry is in the relationships — the bonds between data.

Speaker-coach reminders

Use this right before another run-through. The goal is fewer words, more signposts, cleaner thesis.

Pacing
18:00
Checklist

Slow down here

PE complexity, connected model, AI traceability, and the final chemistry line.

Cut first

Detailed walkthroughs of every principle, every architecture layer, and every dashboard KPI.

Watch habit words

Your run-through had lots of “you know,” “so,” “um,” “kind of,” and “right.” One pause beats three hedges.

Prep capture

Saved locally in this browser. Use it for revised lines, Q&A, and post-run timing notes.

Open prep room