Partner AI Training · Part 2 · May 2026
01 / 00
Advent International
Partner AI Training · Part 2 · May 2026

Advent proprietary intelligencevia various AI tools.

The capabilities we have built, the ones we are building, and the path forward together.

Front Office Technology & Innovation · Advent Labs
Our tools

Three platforms. One purpose.

AdventLabs is tool-agnostic. We work across all three to bring the best capabilities to every deal team.

KEY INSIGHT
Our team's job goes far beyond building AdventGPT. We build data connectors, develop workflows, educate users, and surface what's possible. The skills and best practices we create work across platforms, and they are what make each tool actually useful for PE work.
Where AdventGPT fits

Filling gaps the out-of-the-box tools don't address.

Tools like Claude Code, Codex, and the Claude/ChatGPT chat apps are powerful harnesses around a state-of-the-art model, but they aren't purpose-built for large-scale document indexing or company-specific workflows. Pointed at a deal folder, they have no prior knowledge of the firm and no structured understanding of our deal drives. Every search is like a new hire with zero training, feeling their way through the room. AdventGPT shows up with that structured understanding already in place, which is why it produces richer answers than a cold chat harness can. The four below are concrete examples; the pattern generalizes.

EXAMPLE · VDR MAPPER
Built for thousands of files at once.
A data room can contain thousands of files. The chat harnesses give the model a fixed context window per call, and a 5,000-file VDR overwhelms it. AdventGPT chunks the work across 1,000 cheaper-model calls in parallel and returns a structured map of the entire VDR in minutes. A cheaper model run 1,000× beats the best model run once, and costs about 100× less.
EXAMPLE · FISHER DASHBOARD
Scheduled, persistent, production-grade.
Wakes itself every morning, ingests portfolio company emails, extracts KPIs, updates a database, and keeps a secure dashboard live 24/7 in production. A chat tool can run the analysis once when asked. AdventGPT runs it on a schedule, without anyone re-prompting.
EXAMPLE · MEDIA DIGEST
Internal plumbing, end-to-end.
A custom pipeline that ingests the daily FSG media report, de-duplicates across sources, ranks by relevance, formats it for the executive audience, and emails it firmwide. Not something a chat harness does on its own; it is a stitched-together flow with schedulers, email, and internal distribution wired in.
EXAMPLE · 50+ PORTFOLIO AGENTS
Prior knowledge, before the first prompt.
Each agent has already consumed the full deal folder, thousands of files, before any question is asked. A cold search through Claude or ChatGPT might surface 3 sources. AdventGPT, with structured prior understanding of the company, draws answers from 40+ documents. "How has the CFO performed over time?" returns a sourced answer instead of a guess.

The boundary keeps moving. Claude and ChatGPT will absorb more features over time, and AdventGPT will stay focused on the gaps that matter for PE work: deep document indexing, persistent scheduled workflows, and structured prior knowledge of the firm's deals.

How AdventLabs operates

We are wiring our institutional knowledge into every AI interaction by building use-case-driven, model-agnostic skills and micro agents.

  • State-of-the-art models need purpose-built data connectors to engage with organizational data
    e.g., Outlook email search beyond the 10-result limit
  • Skills are organizational reasoning encoded as plain-English instructions for how to think about a problem
    e.g., LOI drafting: locale, signatories, comparable deals
    Empowers a generic model to do Advent-specific, sophisticated work
  • Built on Model Context Protocol (MCP); interoperable no matter the model
  • Built for both commercial tools (Claude, ChatGPT) and in-house AdventGPT
  • Empower teams to build their own apps. Don't outlaw vibecoding, enable it
  • Great for solving tightly-scoped, low-integration, low-security problems
  • Provide guardrails (CLAUDE.md files) as complexity grows
  • Promotion path: when an app proves value, we deploy to cloud and set up infrastructure
  • Cross-team elevation: if one team's app benefits others, we broaden scope firm-wide
  • Build use-case-driven micro agents in AdventGPT for productivity, insights, and workflow automation
  • For workflows that touch multiple systems and datasets, requiring significant engineering effort
  • End-to-end AI products that are long-term intelligence assets for the firm
  • Our team assists with the design, implementation, and deployment
  • With the operational benefit of monitoring, error recovery, and logging
  • Run tailored trainings alongside the ChatGPT Enterprise rollout to help with adoption
  • Light-touch coaching for AI novices, not just quick tips. Teaching prompting from first principles
  • Evolving AI Corner for online tutorials and recorded webinars teaching new and advanced features in AdventGPT
Data assets

We have both proprietary and 3rd-party data assets, and we are actively working to make them available across all endpoints.

AdventDB centralized data layer connecting proprietary and third-party data sources to firm AI tools

⁺ Plus Outlook email and calendar, with more sources coming online each quarter.

By the way

Most top use cases from the AI competition are already live.

Build a public information book? company-research. Recruit advisors? expert-network-builder. Kathy Fang's digital twin? In development now.

THE REPOSITORY
We are centralizing every AI artifact the firm has produced (prompts, agents, workflows, and skills) into one repository. Since the firm's first prompt library in early 2023, it has grown into a significant and valuable asset. The Mind is how we make it accessible to everyone.
▸ Open the training repository
The Advent Mind
ACCESS

One email.
The whole firm.

01 Send an email in plain English.
No command syntax. Start with "Mind," and everything after is natural language.
02 The Mind routes to the right capability.
Skill, agent, workflow, or prompt. The user doesn't decide; the system does.
03 Deliverable lands in your inbox.
Advent-branded Word, Excel, PowerPoint, or HTML. Ready for the deal team or IC.
The Advent Mind
Launching now

Your 24h
assistant.

One email. No login. No new tool. Just ask.

▾ What landed back
How it routes
Send an email in plain English. The Mind routes to the right capability: skill, agent, workflow, or prompt. Deliverable lands back in your inbox in Advent brand style.
Why it takes a few minutes
It may take 15 minutes to respond. We are being deliberate. The email form factor rewards thoroughness over speed. Rapid, interactive work: web UI, Slack app. Longer-form work: email. Just like you would with an assistant.
A real example · April 2026

One question. One workbook back.

A partner asked the Mind to map warm-intro paths into Lululemon via Advent's APN network. Here is how it played out.

1The ask · 11:01 PM
The ask: warm-intro paths into Lululemon
2The acknowledgment · 11:02 PM
The acknowledgment with work plan
3The response · 11:13 PM
The response with APN warm-intro deliverable
⬇ Download Lululemon_APN_Warm_Intro_Paths.pptx 9 slides · Executive summary, leadership coverage, per-leader path deep-dives, recommended actions
Skills · Live today

40+ skills across the full PE lifecycle.

Click any skill name to launch it in Claude with a prefilled prompt.

▸ Live setup · pause here
Get every partner into Claude, install the skills, and turn on the connectors so they can run any of the below from their own login.
The journey

From scattered prompts to a shared firm brain.

In early 2023, we published the firm's first AI prompt library. Since then, we took on the responsibility of maintaining an inventory of the best AI use cases across the firm, and we ran an AI use case competition to surface the most promising ideas.

Every quarter we added a new layer. Prompts became micro agents and workflows, then came the skills paradigm. Today, the Advent Mind brings it all together: one system that routes your question to the right capability.

2023
First prompt library published
2024
AdventGPT launched · Workflow automation
2025
Knowledge agents · ChatGPT Enterprise
2026
Claude Enterprise · The Advent Mind goes live
Skills · Pipeline

New skills scoped for the next month.

AdventLabs is pursuing each of these aggressively. Scaling the deal team's productivity is a major goal for the firm, and these skills are the next tranche of capability the deal teams are asking for right now.

SOURCING · 7 NEW
whitespace-mapacquisition-pattern-trackersponsor-watchlistsubscale-champion-finderadjacency-mapperfragmented-market-detectortarget-universe-builder
SCREENING · 7 NEW
red-flag-scannermoat-evidence-graderpricing-power-detectormanagement-reputation-scanconcentration-risk-spotterprecedent-transactionspublic-comps-builder
DILIGENCE · 9 NEW
customer-voice-synthesizerdiligence-question-genfinancial-recast-builderrevenue-build-enginewhat-if-scenario-enginecohort-vintage-builderarr-bridge-builderunit-economics-calcability-to-pay-calc
IC / APPROVAL · 8 NEW
ic-pushback-simulatorrisk-prioritizerdownside-case-narratorconviction-gap-finderthesis-pressure-testerdeal-breaker-checklistvaluation-triangulationsensitivity-lab
PORTFOLIO · 3 NEW
portfolio-news-radarvalue-creation-scannerpositioning-refresh
EXIT · 5+ NEW
equity-story-builderexit-timing-analyzersynergy-angle-generatorbuyer-objection-predictorexit-readiness-checker
How to use AI

From simple to expert.

The same question, asked four ways. Click any send button to fire the prompt at Claude.

Simple
"Just an LLM, no?"
You don't know what each tool can really do. Wrong context, lazy prompt, wrong model. The answer comes back vague and you walk away frustrated.
what markets could orveon expand into?
Aspiring
Better prompts, better answers.
You learn each model's strengths and write structured prompts with context, constraints, and format. Sharper answers. Still copying the same workflow over and over.
You are advising Advent International on Orveon, a beauty/cosmetics holding company in our portfolio (bareMinerals, BUXOM, Laura Mercier). Identify the 5 most promising adjacent markets Orveon could expand into. By adjacent, I mean channel synergies and brands with similar positioning (not much more premium, not much less premium). For each, give: market size, growth rate, key competitors, why Orveon has the right to win, the most likely mode of entry (organic, M&A, partnership), and one risk. Format as a ranked table. Keep it tight.
Rising star
Spot the pattern. Build a skill.
You realize you keep doing the same analysis for every portco. You codify it as a skill, share it with the team, and the pattern compounds across the firm.
I keep getting asked "where could this portco grow next?" for different portfolio companies. Create a new skill called portco-adjacency-expansion that systematically evaluates adjacency expansion for any portco. The skill should: (1) define the company's core business, (2) generate adjacencies across product, customer, geographic, channel, value-chain, and business-model vectors, (3) screen each on attractiveness, right-to-win, and financial impact, (4) shortlist 3 to 5 priorities with two-pagers, (5) suggest bolt-on M&A targets, (6) produce an Advent-branded Excel and HTML report. Trigger on phrases like "adjacencies for X", "where can X expand", "growth opportunities for X". Apply Advent brand identity throughout.
Expert
Use AI to write the prompt.
Ask the model to think like a top PE professional and draft the prompt that will create the skill. THEN feed that prompt back in. The AI writing the AI.
Step 1: ask AI how a top PE pro would write this skill-creation prompt. Step 2: take that output and feed it to the skill-creator.
What it can do · Examples you can run

Classical deal team work, in minutes.

Each card below is a real example. Click the paper-plane to send the request to Mind from your inbox.

NETWORK
APN warm-intro paths
Maps Advent People Network connections to surface the warmest path to a target. Works for one name or hundreds in bulk. Advent-branded deck with ranked paths.
▾ ResultEmailPowerPoint
DILIGENCE
LBO model from a CIM Beta
Extracts the financial package, validates figures, and populates a full PE-grade multi-tab model with sources, debt schedule, and returns analysis. Still maturing; flag any rough edges.
▾ ResultEmailExcel
QA
Model review and deck audit
Reads a financial model cell by cell, or a deck slide by slide. Checks for circular refs, broken formulas, hardcoded overrides, balance sheet imbalances, and cross-slide contradictions. Severity-ranked findings.
▾ ResultEmail
RESEARCH
Sector primer or company profile
Institutional-quality research report with market sizing, competitive dynamics, and valuation context. Advent-branded HTML, ready for the deal team.
▾ ResultEmailPrimer
SCREENING
Comps and precedents CapIQ data soon
Trading comps, precedent transactions, calendarized multiples. The analysis a deal team would build, delivered in minutes. Advent-branded Excel.
▾ ResultEmailExcel
PRODUCTION
Advent-branded documents
Company profiles, stock charts, working group slides, executive briefs, deal memos. Output in PDF, Word, or Excel, all in the firm's visual identity. PowerPoint output is in beta.
▾ ResultEmailPDF
Live exercise · 1 of 2

Build your own skill.

Pick something you do often that the firm would benefit from doing the same way every time. We will brainstorm together, then each of you will pick one of the three approaches below and build it live.

A

Type /skill-creator in Claude and describe the skill in plain English. The base Anthropic skill-creator (not the AdventLabs modified version) will interview you, scaffold the SKILL.md file, and walk you through testing.

▸ Fastest path
B

Run the task once end-to-end with Claude. Then ask it to generalize what just happened into a reusable skill: capture the inputs, the methodology, the output format. The example becomes the spec.

▸ Most concrete
C

Ask Claude to think like a top PE professional and write the prompt that would create this skill. Then feed that prompt back to skill-creator. The AI writing the AI. See From simple to expert for the full demo.

▸ Brainstorm prompts
What recurring task burns hours every week? What analysis do you redo for every deal? What output do you wish came back already in the firm's format? That's a skill.
Live exercise · 2 of 2

Read the whole VDR. Right now.

Same task, two paths. We will run them side by side and look at the difference together. The point is not which one wins; it is to feel where Claude alone hits a wall, and where AdventGPT picks up.

A
Claude · directly
Point Claude at the VDR.

Open the VDR in SharePoint, share the folder URL with Claude, and ask it to summarize the contents and surface the most important findings. It will try. Watch what happens when context windows meet thousands of files.

B
AdventGPT · VDR mapper
Use the purpose-built agent.

Send the same VDR to the VDR Mapper agent. It runs a cheaper model 1,000× across every file in parallel, returns a structured map of the entire data room, and surfaces the documents that matter, in minutes, at roughly 100× lower cost.

▸ Discuss
What did you see? Coverage, depth, speed, cost: where does each break down? This is exactly the gap the next slide is about.
The vision

Each sector. Its own Mind.

Over time, the Mind specializes. Each sector team gets its own mailbox carrying its deal history, sector-specific skills, and the team's preferred analytical voice. CC the right one and the answer comes back already steeped in context.

consumer@adventinternational.com
Knows
Orveon, Sauer Brands, Salt & Stone, Parfums de Marly, Zimmerman, Olaplex
Skills
amazon-review-analyzer, shelf-velocity-tracker, dtc-funnel-decomposer, retail-account-mapper
Voice
Brand-fluent, retail-savvy, trend-aware. Talks SKUs, channels, and elasticity.
psg@adventinternational.com
Knows
Value creation levers, historical effectiveness of each play, talent database for operating roles
Skills
vcp-stress-tester, 100-day-planner, benchmark-engine, synergy-quantifier
Voice
Consultant-fluent, slide-precise. Frames problems in 2×2s. Quantifies before opining.
healthcare@adventinternational.com
Knows
AccentCare, BioDuro, Suven Pharma, Definitive Healthcare
Skills
regulatory-pathway-mapper, pharma-pipeline-analyzer, clinical-trial-tracker, payer-channel-mapper
Voice
Reg-fluent, payer-savvy, clinical-aware. Knows when to call the experts.
tech@adventinternational.com
Knows
SonarSource, Wiz, BigID, Cyware, Automic, Tinexta
Skills
gen-ai-defensibility, rule-of-40-checker, net-retention-analyzer, open-source-mapper
Voice
ARR-fluent, SaaS-native, AI-defensibility-obsessed. Reads roadmaps for breakfast.
bfs@adventinternational.com
Knows
EBANX, MangoPay, Thredd, Yes Bank, Worldpay
Skills
customer-cube, excel-dcf-modeler, market-primer, leadership-mapper
Voice
Take-rate-fluent, regulatory-aware. Knows the difference between rails and risk.
industrial@adventinternational.com
Knows
LayerZero, Imperial Dade, Tigre, Aareon-2, DFM Foods, Wagas
Skills
aftermarket-mix-analyzer, capex-payback-modeler, supply-chain-stress-tester, industrial-cycle-mapper
Voice
Capex-aware, supply-chain-literate. Knows aftermarket from OE in two questions.
CONTEXT PERSISTS
Every sector Mind keeps the context of its conversations: deals, preferences, history. Nothing is shared with others unless they are cc'd. CC two sector Minds on the same email when a question crosses them, and you get both perspectives back in one thread.
How we are disseminating

A curriculum the firm can grow into.

This month: 4 group sessions and individual interactions with 30+ people across AdventLabs and IT, from power users like Dov and Eli to people still figuring out where to start.

50+
Modules
~5h
Full curriculum
40+
Codified skills
1-3min
Video per module
01 · Foundations
7 modules: mental model, prompting, which tool for which job
02 · Excel & data
7 modules: clean data, market maps, variance, model audits
03 · PowerPoint & docs
6 modules: branded slides, tearsheets, deck QA, redlines
04 · Productivity
6 modules: inbox, calendar, meeting prep, status summaries
∞ · Skills chapter
Replicating the mind of an investment professional
A SMALL ASK
We are reaching out to your teams and offices in the coming weeks to schedule trainings. Your support in landing the dates and rallying attendance would mean a great deal.
"
AI should increase our accuracy and speed in analysis, and increase the quality and quantity of research available for every decision we make. With AI tools, teams can apply judgement to higher quality inputs and drive increased deployment.

As a result, more time should be available for applying judgment and for portfolio management. The goal would be to unlock our ability to drive breakout returns with greater frequency, at a minimum due to the time created, but ideally through AI supported value creation as well.
TRICIA GLYNN · ADVENT
THE GOAL
Codify the knowledge of the best deal team.
Augment the deal team. Scale the firm's judgment.
mind@adventinternational.com
One email. The whole firm.
Questions?
ISSUED BY Advent Labs · Front Office Technology & Innovation · May 2026
⟨ bringyourown.app ⟩