AI across the Seller Connect Engine

Where AI already lives in the engine, where it naturally fits next, and how it connects to Ed's buyer outreach automation ask. 22 touchpoints across 11 of 14 epics.

Already specced and building In PRDs with prompts written
E3S3.2
Voice screening agent
Retell AI conducts structured 4-block interview via phone. AU accents, emotional sensitivity, follow-ups. Evaluation prototype.
E3S3.4
Open-text LLM evaluation
LLM evaluates freetext answers for coherence, red flags, sentiment. Feeds scoring dimensions (motivation, financial clarity).
E3S3.5
TL;DR generation
Anthropic SDK generates anonymized broker-facing summary. 60-second read. Prompt V1 written. Model + version tracked.
E3S3.6
Exception triage assist
AI-generated risk summaries for quarantined leads. Internal team reviews in Boss Mode. Human-in-the-loop.
V1 natural insertions Low effort, ships with prototype
E2S2.3
Bucket suggestion from ABN
ABR returns ANZSIC. AI maps to bucket AND suggests plain-language description seller can confirm/edit. Reduces friction.
E4S4.6
Match explanation for brokers
"Matched because: hospitality specialty, Sydney metro, strong conversion." One line on lead card. Builds trust in the algorithm.
E4S4.7
Match quality prediction
After 50 leads, LLM analyzes outcomes and suggests ranking weight adjustments. Boss Mode recommendation, human approves.
E14S14.2
Personalized email copy
"Hi Sarah, we've reviewed your cafe in Bondi" not "Dear Seller, your submission has been processed." Warmth at scale.
V1+ layers Medium effort, needs 50+ leads first
E4V2
Semantic matching
Ed's direction: AI-driven matching using business descriptions, keywords, industry data. Layers on top of bucket hard-filter.
E5
Dynamic credit pricing
Lead quality + industry demand signals feed pricing recommendation. Internal dashboard, not auto-pricing. Revenue optimization.
E5
Gaming / fraud detection
Pattern detection on credit purchases without accepts, synthetic profiles, rapid-fire lead gen. Flags for internal review.
E12
Predictive funnel insights
LLM analyzes Mixpanel data, surfaces weekly summaries. "Gate 2 drop-off up 15%, correlates with ABN lookup latency." Darko's domain.
Future layers Data model supports from day one, build when data exists
E6
Call transcription + summary
Transcribe masked broker-seller calls. LLM generates summary, flags action items, detects sentiment.
E6
Safety net monitoring
Post-call sentiment analysis. Flag distressed sellers or aggressive brokers for internal review.
E7
Smart document routing
ASIC entity type + AI determines correct signatories for complex structures (trusts, partnerships).
E7
Clause analysis
Flag non-standard amendments to commission addendum. "Seller wants anti-circumvention reduced to 6 months. Risk: high."
E8
Document classification
Auto-tag uploaded financials, leases, P&Ls, agreements. Organizes the data room without manual sorting.
E8
Financial data extraction
Extract revenue, profit, EBITDA from uploaded P&Ls and BAS. Cross-validate against seller's screening claims.
E10
Referral quality scoring
Which accountants send the best leads? Score by conversion, quality, deal outcomes. Surface top performers.
E11
Buyer thesis matching
Semantic match between buyer acquisition criteria and off-market seller profiles. Core value prop of the off-market pool.
Ed's buyer outreach automation Adjacent to SCE, shares 60% of tech stack
1
AI reads IM
LLM ingests Information Memorandum. IMs already in Resolve's data room.
2
Industry research bot
Trends, M&A transactions, dynamics. Shares enrichment layer with E4 semantic matching.
3
Buyer archetypes
AI-generated strategic buyer personas from research. Feeds off-market pool matching (E11).
4
Buyer list build + enrichment
Database enrichment + web scraping. Grows Pool B for deal sheet distribution.
5
Outbound comms drafting
Personalized email/SMS with A/B testing. Same LLM infra as E14 emails and E3 TL;DR.
6
Attribution tracking
Campaign tracking, engagement, 1-4 week attribution window. Same Mixpanel layer as E12.
Deal sheets are the productized version of steps 4-6. They distribute listings to buyer databases with tracking. The outreach pipeline is the AI-heavy version that generates the buyer list and copy. Parallel paths to the same outcome.

Data moat: Every AI feature feeds from the same data layer. Screening responses train better scoring. Match outcomes train better matching. Verified sale data from the buyer-credit contribution becomes the industry's best dataset on AU deal outcomes. By month 12, a compounding advantage no competitor can replicate.

22 AI touchpoints across 11 of 14 epics. 4 already specced with prompts written. 4 low-effort V1 additions. 4 medium-effort V1+ layers after 50 leads. 8 future layers the data model supports from day one. Ed's 6-step buyer outreach pipeline shares 60% of the tech stack we're already building.