A8 Presale Summary App
A purpose-built workspace where AI synthesizes deal intelligence from multiple sources into a single, evolving document -- with human-in-the-loop review on every change.
The Business Problem
Aptitude 8 is a HubSpot consulting firm where deals range from one-week engagements to six-month enterprise implementations. Every deal generates a massive volume of intelligence -- discovery calls, stakeholder emails, requirements documents, qualification notes -- spread across HubSpot, Avoma, Google Docs, Slack, email, and individual reps' heads.
This creates cascading problems across the revenue team:
Information Lives in People, Not Systems
| Problem | What AEs Feel | What Collaborators Feel | Impact on Revenue |
|---|---|---|---|
| Tool fragmentation | "I'm always browser-hopping" | Tools don't talk to each other | Inefficiency at scale |
| Information decay | "I forgot what they said" | "What did sales learn?" | Deals stall or die |
| Admin burden | "I don't have time for this" | Notes are incomplete | Lost capacity to be strategic |
| Handoff friction | Pressure to brain-dump | Starting from scratch | Project risk, rework, escalations |
| Weak discovery | Feeling unprepared | Mis-scoped projects | Lower win rates, deal quality |
| No synthesis | Can't find what I need | Missing the "so what" | Forecast uncertainty |
The Ever-Changing Context Problem
Deal cycles vary wildly, and when there are gaps, reps end up re-learning their own deals. Reading 2-5 call transcripts before a meeting isn't realistic. Critical context -- what matters, what changed, what's risky -- disappears between touchpoints. Picking a deal back up means asking "where did we leave off?" instead of selling.
The Tool Problem
Reps toggle between 8-10 tools constantly. Each captures a slice of deal intelligence, but nothing synthesizes the whole picture or tracks how deals evolve over time. Current AI tools help with one-off tasks, but they aren't designed for A8's workflow, don't maintain persistent deal context, and can't integrate with the systems reps already use.
The core tension: documentation competes with selling time. Reps are forced to choose between staying prepared and actually working deals.
The Solution
The A8 Presale Summary App is a purpose-built workspace where AI synthesizes deal intelligence from multiple sources into a single, evolving document -- with human-in-the-loop review on every change. Reps feed in transcripts, emails, and notes; the AI suggests targeted updates to a structured deal document; and reps accept, reject, or refine each suggestion individually.
The core innovation: Google Docs-style inline suggestions, powered by AI.
Unlike generic AI tools that produce monolithic output you either take or leave, the Presale Summary App gives reps per-change granularity. Accept the new stakeholder detail but reject the timeline update. Each suggestion is independent, editable, attributed to its source, and accompanied by the AI's rationale.
The result: deal context builds over time without the manual overhead. Reps stay in control. The document stays current. And the choice between selling and documenting disappears.
Tool Consolidation, Not Addition
The app replaces three tools AEs currently juggle:
| Replaced Tool | What It Did | What the App Does Better |
|---|---|---|
| Google Docs | Separate deal notes documents | Structured, AI-assisted deal doc that updates itself |
| Avoma (for review) | Jump into Avoma to review past calls | Transcripts imported directly; searchable within the app |
| ChatGPT / External AI | Copy-paste context for ad hoc help | Built-in AI that knows your deal, your sources, and your document |
System of Record vs. System of Context
The app doesn't compete with HubSpot -- it complements it. HubSpot is the system of record (where is this deal?). The Presale Summary App is the system of context (what do we know about it?). They feed each other. Deal properties, notes, and emails flow from HubSpot into the app. MEDDPICC qualification status and AI-generated summaries sync back.
What We Built
Seven-Tab Document Architecture
The app organizes deal intelligence across specialized tabs, each serving a distinct purpose in the sales lifecycle:
| Tab | Purpose | Key Capabilities |
|---|---|---|
| Agendas | Meeting preparation | AI-generated discovery plans, manual agenda creation, date-based sorting |
| Deal Context | Internal deal intelligence (13 sections) | Stakeholders, timeline, risks, upsell opportunities, technical requirements -- all AI-updated |
| Deal Desk | Client-facing content (4 sections) | Executive summary, pain points, goals, outcomes -- copy-to-clipboard for quick sharing |
| MEDDPICC | Sales qualification (8 elements) | Color-coded status tracking, auto-analyzed from sources, HubSpot sync |
| Handoff | Combined read-only view | All sections merged into a single handoff document for delivery teams |
| Change Summary | Source-to-document audit trail | AI-tracked changes per source, chronological feed of what changed and why |
| Enablements | Generated deliverables | Follow-up emails, objection handling, company research, and more |
AI-Powered Document Processing
When a rep adds a new source (transcript, email, notes), the AI processes it in three phases:
- Phase 1: Generates suggestions for Deal Context sections (13 sections, parallel)
- Phase 2: Generates suggestions for Deal Desk sections (4 sections, parallel)
- Phase 3: Auto-analyzes MEDDPICC qualification (8 elements, auto-queued)
Processing happens in the background. Reps can navigate, edit other sections, or work on enablements while the AI works. Notifications alert them when suggestions are ready.
Suggestion types:
- Additions (green) -- new information to incorporate
- Modifications (orange) -- updates to existing content, showing old vs. new
- Deletions (red strikethrough) -- content that may no longer be accurate
Batch Accept All / Reject All controls per tab enable rapid review when suggestions are high-quality.
Eight Enablement Types
The app generates standalone deliverables from deal context and sources:
- Follow-up Email -- Send-ready next-steps email personalized with deal context
- Discovery Plan -- 2-4 AI-generated meeting agendas from prior conversation analysis
- Agenda Email -- Follow-up email with meeting context and prep
- Objection Handling -- Extracts objections with coaching recommendations
- Process Overview -- Synthesizes the client's buying and approval process
- HubSpot Entity Structure -- Proposed CRM objects/properties from discovery
- Company Overview Research -- Web-search-enabled company research brief
- Company News & Signals -- Recent news, acquisitions, and buying indicators
Enablements 7 and 8 use Claude's web search tool to pull real-time external data -- the first pattern in the app for live research beyond deal sources.
Semantic Search: Ask Deal Assistant
Reps can ask natural-language questions across all their deal sources:
- Ask Source: Quick Q&A against a single transcript or document
- Ask Deal Assistant: Semantic search across all embedded sources using vector similarity (OpenAI embeddings + pgvector)
Instead of re-reading transcripts to find "what did the CFO say about budget?", reps ask the question and get attributed answers with source citations. Responses stream in real-time.
MEDDPICC Sales Qualification
The app tracks all eight MEDDPICC elements (Metrics, Economic Buyer, Decision Criteria, Decision Process, Paper Process, Identify Pain, Champion, Competition) with:
- Color-coded status: Unknown (red) to Partially Identified (yellow) to Identified (green)
- AI-powered analysis: Automatically assesses qualification evidence from processed sources
- HubSpot sync: Push MEDDPICC statuses and AI-generated summaries back to HubSpot deal properties
The AI distinguishes genuine evidence-based qualification from mere mentions, giving reps and leadership an honest picture of deal health.
Integrations
| Integration | Direction | What Flows |
|---|---|---|
| HubSpot | Bi-directional | Deal validation, notes/emails as sources, MEDDPICC status sync, next steps |
| Avoma | Inbound | Call transcript import, AI-generated call insights (summaries, action items, takeaways) |
| Claude API | Core engine | Document processing, MEDDPICC analysis, enablement generation, Ask Deal |
| OpenAI | Embeddings | Source chunking and vector embedding for semantic search |
Measured and Expected Benefits
Time Savings
| Activity | Before (Manual) | After (With App) | Estimated Savings |
|---|---|---|---|
| Post-call documentation | 15-30 min writing notes per call | 2-5 min reviewing AI suggestions | ~80% reduction |
| Deal prep / re-learning | 20-45 min reading past transcripts | Ask Deal question, get instant answer | ~75% reduction |
| MEDDPICC qualification | Manual assessment across scattered notes | Auto-analyzed from sources with evidence | ~70% reduction |
| Handoff document creation | 1-2 hours compiling from multiple tools | Always current, one-click handoff tab | ~90% reduction |
| Follow-up email drafting | 10-20 min per email | Generated from deal context in seconds | ~85% reduction |
| Meeting agenda prep | 15-30 min reviewing context, drafting agenda | AI-generated discovery plan from prior calls | ~75% reduction |
Conservative estimate: 3-5 hours saved per rep per week across documentation, deal prep, qualification, and enablement generation -- time redirected to actual selling.
Quality Improvements
- Richer deal documents: AI catches details humans skim over in transcripts -- stakeholder titles, specific pain points, exact timelines
- Consistent qualification: Every deal gets the same rigorous MEDDPICC analysis, not just the ones reps have time for
- Better handoffs: Delivery teams start with comprehensive, current context instead of incomplete brain-dumps
- Institutional knowledge: Deal intelligence lives in the system, not in people's heads -- survives PTO, role changes, and deal pauses
- Audit trail: Change Summary tab provides a chronological record of what changed and why, per source
Revenue Impact
| Lever | How the App Helps | Expected Impact |
|---|---|---|
| Win rate | Better-prepared reps, deeper discovery, stronger qualification | Improved deal quality and close rates |
| Deal velocity | Less time on admin, faster handoffs, quicker re-engagement after pauses | Shorter sales cycles |
| Forecast accuracy | Documented MEDDPICC evidence vs. gut feel | More reliable pipeline predictions |
| Expansion revenue | Captured upsell opportunities and technical requirements persist | Better positioned for follow-on work |
| Delivery success | Comprehensive handoffs reduce mis-scoping and rework | Lower project risk, fewer escalations |
Current Status: Live in Beta
The app is deployed on DigitalOcean with Docker, running with Google SSO authentication and Supabase as the database layer. Two AEs are actively using it on real deals.
What's been built:
- 124+ completed work items shipped
- 7 document tabs with 17+ structured sections
- 8 enablement types (including web-search-enabled research)
- 3-phase AI processing pipeline with background processing
- Full Avoma and HubSpot integrations
- Semantic search across all deal sources
- MEDDPICC qualification with HubSpot sync
- Auto-save with real-time persistence to Supabase
- Row-level security on all tables
- Light/dark theme with A8 design system
Technical Architecture
| Layer | Technology |
|---|---|
| Frontend | React 19, TypeScript, Vite |
| Rich Text Editor | TipTap v3 (per-section instances) |
| Backend | Express.js API server |
| AI Processing | Claude API via @anthropic-ai/sdk (Sonnet 4.5 / Opus 4.5) |
| Embeddings | OpenAI text-embedding-3-small + pgvector |
| Database | Supabase (PostgreSQL + pgvector + RLS) |
| Authentication | Google SSO via Supabase Auth |
| Deployment | Docker (client: nginx, server: Node.js) on DigitalOcean |
| Styling | CSS custom properties (A8 design system tokens) |
Key architectural decisions:
- Per-section TipTap editors (not a single monolithic editor) enable granular suggestion targeting and section-level locking during processing
- Three-phase processing pipeline allows background AI work while reps continue editing
- Modular prompt architecture with separated prompt templates per section type, enablement type, and MEDDPICC -- easy to iterate and A/B test
- Embedding pipeline chunks sources (~500 tokens, 100-token overlap) for semantic search across all deal intelligence
What's Next
| Priority | Initiative | Impact |
|---|---|---|
| 1 | Incorporate beta feedback | Refine UX based on real usage patterns |
| 2 | Full AE rollout (8 reps) | Scale validated benefits across the team |
| 3 | SA Scope Builder app | Dedicated workspace for Solutions Architects with shared deal visibility |
| 4 | Deeper HubSpot integration | More data flowing both directions, automated triggers |
| 5 | Cross-app deal context | AE presale summaries inform SA scoping decisions, reducing handoff friction further |