Thunder Graph — Getting started

The big picture

Every mandate in Thunder Graph moves through the same lifecycle. You scan a company to see who would realistically fund or buy it, convert it to a client once economics are signed, optimize and shortlist the target list, run a campaign of personalized outreach, and drive replies down the funnel to a close. A parallel Origination engine surfaces companies that look like they are about to need a banker, and feeds them into the very same pipeline.

The philosophy in one paragraph. Thunder Graph is built on a single division of labor: the machine does the drudgery — researching thousands of firms, verifying mandates against real evidence, drafting every email, triaging every reply — and a human makes every judgment call. Nothing is ever contacted, sent, or committed without your explicit click. There is no auto-send anywhere in the system. Every outbound message sits in an approval queue until you approve it, and every approval is logged with your name in the audit trail. You are never cleaning up after an over-eager robot; you are editing drafts from a tireless analyst.

Find → New scan (Introduce a company)

The scan is where every deal starts — usually while the prospect is still on the phone. It builds a company profile, then searches the entire graph of investors and acquirers for realistic counterparties.

1. Open New scan and stay on the Introduce a company tab.
2. Enter the company — name or domain. Existing network records are matched automatically; unknown companies get a fresh record.
3. Pick the intent: Raise capital (then choose financing type: venture, buyout, or debt) or Be acquired. The intent drives everything downstream — which firms qualify, which mandate lane is verified, which reality test applies.
4. Paste any context you have: call notes, a URL, a deck link — anything. More context means a sharper profile, but even a bare name works.
5. Click Build profile. The AI assembles sector, model, scale, and geography, then proposes lifecycle stages — confirm or correct them before proceeding.
6. Click Find intros. In under a minute you get a ranked candidate list drawn from the whole graph.

What the matcher reads — and what each thing is worth

The context box is not a formality. On a sale mandate the strategic scorer has about 138 points to award, and the single most valuable thing you can give it is a company name. Categories describe a shape; names identify a buyer. Everything below is extracted automatically from whatever you paste — call notes, a transcript, a one-pager — so write it the way you would say it out loud.

Reading the results: the three tiers

Mandate verification pills (arrive ~1 minute later)

Ranking is fast; verification is deeper. About a minute after results land, each candidate picks up a mandate pill showing whether the firm's own stated thesis plus reality data (what they actually invest in or acquire) support contacting them for this specific deal:

• ✓ mandate  Verified. The firm's own thesis and its observed behavior both support this deal type. Safe to prioritize.
• ✗ growth equity only — no debt  Excluded, with a reason. Verification found a disqualifier; the candidate is auto-excluded and the reason is shown. You can override with a 👍 if you know better.
• ? unverified  Unknown. Not enough evidence either way. Treat as tier-2 diligence work, not a rejection.

Lanes. Verification runs in the lane that matches your intent, because the reality test differs: a majority sale checks for control-deal appetite and strategic acquirers; a venture round checks stage and check size; a buyout checks for control transactions at your EBITDA scale; debt checks for actual lending activity, not equity headlines. The same firm can be ✓ in one lane and ✗ in another — that is by design.

Votes: 👍 and 👎

Votes survive re-runs. Re-scan the same company next month with new intake data and every 👍/👎 you cast is still applied. You never re-litigate a decision you already made.

The refine bar

Above the results sits a plain-English refine bar. Type an instruction; the list re-ranks or filters instantly:

Real example: on a live debt deal, that one sentence removed 166 of 200 candidates — the profile had tripped a property-adjacent keyword and flooded the list with real-estate lenders. One sentence, fixed.

Strategic buyers (sale mandates)

When the intent is Be acquired or a majority sale, a dedicated Strategic buyers section appears below the financial sponsors: operating companies with a rationale to own the asset — product adjacency, geographic expansion, roll-up activity — each with the reasoning spelled out. Curate them exactly like investors: pills, votes, shortlist.

Search my network

Not everything needs a full scan. Search my network answers real questions, asked the way you'd say them, against Thunder's own graph — perfect mid-call, when a prospect asks "do you actually know anyone in our space?"

1. Open Search my network from the main navigation.
2. Ask a full question: "Who do I know in the Bay Area that invests in seed stage companies?", "What firms focus on search funds and entrepreneurship through acquisition?", "Which PE firms writing $5–20m checks have acquired logistics companies?" The AI understands location (city/metro), stage, sectors, check size, deal types, past investments, niche theses and more.
3. An "Understood as" line shows exactly how your question was interpreted (click details for the raw filters). If it misread you, just rephrase.
4. Results come back in two sections — People (who you actually know, warmest first, matched on their own location before their firm's HQ) and Firms (warm relationships first, each with a "Why" explaining the match). Click any row to open the full network record (Section 11).

Convert to client (Deal room)

A scan becomes a mandate in the Deal room tab. Converting does two things: it records the economics of the engagement, and it opens deep intake — the structured facts that turn a good target list into a great one.

1. The economics form

1. Retainer — monthly amount, if any.
2. Success fee — percentage and structure.
3. Exclusivity — yes/no and term.
4. Client contact for reports — the person at the client who receives progress updates. Set this now; Section 8's client updates need a recipient.

2. Deep intake — each field sharpens matching

Add fields as you learn them — none are required at scan stage, and every one you fill re-sharpens the list. Skip freely early; return often.

3. Supporting documents

Upload the deck and financials in the Deal room. The AI reads them and merges the facts into the company profile — revenue, margins, growth, customers. Outreach and reports use derived summaries only; raw files are never shared with anyone outside Thunder.

Target list curation

Between "the machine found 200 candidates" and "we contacted anyone" sits curation — your judgment pass over every card. Here is the anatomy of a candidate card:

• Fit score — the multi-component ranking score (sector, stage, scale, thesis, behavior).
• 🎯 badge — slam-dunk tier: every criterion confirmed (Section 2).
• Mandate pill — the verification verdict with its reason.
• Warmth + who can intro — the strongest relationship path from anyone at Thunder to the firm.
• Economics fit — check size, geography, and structure vs. this mandate's intake data.

Your three moves per card

1. Shortlist Tier 1 — first outreach wave.
2. Shortlist Tier 2 — the bench; contacted after tier 1 or on a slow week.
3. Dismiss with a reason — pick why (wrong sector / too small / conflict / relationship…). Reasons train the algorithm — a precise dismiss is a gift to every future deal.

Bulk buttons speed up the obvious calls: Shortlist all slam dunks and Shortlist all warm, then prune the exceptions by hand.

Campaign

Once a shortlist is committed, the campaign engine takes over the drudgery of outreach. The deal page is tabbed — Overview · Targets · Outreach · Funnel (Transaction is its own 💰 action) — and the campaign lives across the Outreach and Funnel tabs.

1. Sample → approve the templates. In the Outreach tab, all three emails of the sequence are drafted against your warmest target — the wave's templates. Edit any of them directly, or give plain-English feedback and the AI revises the whole sequence coherently ("the EBITDA figure is wrong" fixes every email; "make email 3 warmer" touches only email 3). Approve, and every contact gets personalized versions of all three.
2. Every contact gets a 3-email sequence. Per contact: email 1 (the intro), email 2 (sends 6 business days after email 1 — one new concrete angle, never a bare bump) and email 3 (sends 9 days after email 2 — a warm last touch), each personalized from your approved templates. All three are editable until the moment each sends, from the Outreach tab or a funnel card's 📧 Campaign modal.
3. Spotted a mistake after drafting? Don't fix 26 emails by hand — hit ✎ Refine templates & regenerate all (Review & start step). It reopens the three templates for the same edit-and-feedback loop, and re-approving regenerates every unsent email in the wave from the corrected templates. Sent emails are never touched.
4. Start the campaigns. Tick targets on the Funnel board and hit ▶ Start N (or Start all ready in Outreach). Email 1 joins the send queue; 2 and 3 follow automatically on cadence — and stop instantly if the contact replies or is passed.
5. The send queue paces everything. Sends go out only Mon–Fri 8am–6pm and Sunday 2–10pm ET, one email every 8 minutes per mailbox — the steady, human pattern that keeps outreach out of spam folders (the lemlist playbook). Each queued email shows its expected slot ("Tue 10:40 AM ET"); ⏸ Pause pulls a campaign back out, and Send now still exists for one-offs.
6. Replies auto-triage. A reply flips the candidate contacted → responded automatically (and cancels its remaining follow-ups). Judgment moves (interested? needs NDA? soft pass?) surface as one-click confirms. Interested and meeting-request replies arrive with a threaded draft response already written — your click to send.
7. Meetings advance the funnel. A booked meeting auto-advances the candidate's stage — no manual bookkeeping. The Funnel board itself is drag-and-drop: pull a card into any column to move its stage.

The funnel ladder

Portfolio

The Portfolio view is the partner-level cockpit across every live mandate:

• Book totals — mandates, total deal value, fee value at stake.
• Health per mandate — green (on pace) / amber (stalling — e.g. no funnel movement in two weeks) / red (at risk — silence, slipped process, unhappy client).
• Weighted fee pipeline — each mandate's fee × its stage-based probability, summed: the honest revenue forecast.
• Conversion by firm type — response and meeting rates per counterparty type (growth PE vs. strategic vs. lender…), so lists tilt toward what actually converts.
• Learning loop — monthly the system proposes ranking-weight adjustments from accumulated votes, dismiss reasons and outcomes. Proposals are human-adopted: nothing changes until a person accepts them. Template A/B response rates live here too.

Client reports

1. Progress report. Every deal has a per-mandate HTML progress report — funnel state, outreach counts, responses, next steps. Open it from the deal, use your browser's print-to-PDF for a client-ready document.
2. Draft client update. Click Draft client update and Opus writes a progress email to the mandate's client contact (set in the economics form, Section 4). The draft lands in the same approval queue as outreach — you edit and click send.
3. Mondays auto-draft. Each Monday, updates for active mandates are drafted automatically and wait in the queue. Drafted — never sent — until you approve.

Origination

Origination is the inbound half of the machine: it continuously scores companies in and around Thunder's network for mandate-prospect likelihood, and shows its work. Signals include:

• Warm path — someone at Thunder already knows the founder or the board.
• Raise window — time since last round suggests they're due.
• Scale — revenue/headcount in Thunder's $5M–$100M sweet spot.
• News intent — coverage hinting at a sale, raise, or strategic review.
• SEC Form D — fresh filings showing capital in motion.
• Podcast guests — founders who appeared on the podcast: pre-warmed, already in the graph.

Every prospect card carries legible reasons — the exact signals and their sources, not a black-box score. Your moves per card:

1. Qualify — worth a partner conversation; keeps it hot on the list.
2. Decline — with a reason; tunes the scorer like a dismiss does (Section 5).
3. Convert — creates a prospect mandate that flows into the same Deal room as any scanned company: scan, economics, intake, shortlist, campaign.

Daily rhythm — a day in the life

Network records

Underneath everything sit the Companies and Contacts records — 360° views of every entity in the graph: profile, relationships and warmth, deal history, meetings, emails, votes, news.

• Fixing data. See something wrong? Edit it on the record. Manual edits survive the nightly rebuild — your correction is pinned, not overwritten by the next automated refresh.
• Hide / delete. Hide removes a record from search and matching but keeps history; delete is for true junk. Prefer hide.
• Inven enrichment. Records are enriched from Inven's 20M+ private-company database — financial scale, deal history, contacts — merged automatically into the profile.
• Pulse. Each record has a live news feed; Pulse items are the same signals Origination and campaign drafts cite.

Going further

You now know the whole loop: scan → convert → curate → commit → campaign → close, with Origination feeding the top and Portfolio watching the whole board. Two last things:

• Release notes — the platform ships constantly; skim the notes to catch new capabilities.
• Every knob is tunable. Ranking weights, tier thresholds, bump timing, draft tone, report cadence — if this guide said "the system does X" and you want Y, ask Claude in the app, in plain English. That is the intended interface.