graph8 · The Onboarding Report
Internal · 2026
Contents
Onboarding Research · 2026

The Onboarding Report

A research-grade view on perfect onboarding for graph8 — mapped to the 1,819-feature inventory, written from the consultant chair after working onboarding for Clay, Apollo, ZoomInfo, Revo, and 30+ other GTM platforms.

Authorgraph8 onboarding research
Pairs withonboarding-agent-value-first-playbook PRD
StatusStrategy reference
Reading time22 min

Figure 1 · The thesis, in one picture

Iceberg diagram: 30 features surfaced above the waterline, 1,789 hidden below. REVEAL · WAIT · EARN Intent + ICP double whammy Bridge linkup · WhatsApp/Slack 1-page account research brief Sequence drafted to hot contact Displacement: "shopping for X" 30 features surfaced in the first 5 minutes named, tactile, demonstrated 1,789 features hidden earned over months of use settings · admin · roadmap · help · API The job is to hide 1,790 and reveal 30.

Part One

The Manifesto

Onboarding is the only moment in a SaaS product's lifecycle where the user has neither learned the tool nor decided whether to keep paying. Every other interaction is conditioned by prior context. Onboarding is unconditioned — and so it's the single most leveraged surface in the company.

For a decade the industry has gotten this wrong in the same way. Onboarding has been treated as a tour ("here are our features") or a setup wizard ("complete your profile"). Both are conditioned on the user doing work for the product. Both fail at scale, because they ask for trust before they earn it.

The 2026 standard is different. Onboarding is now an agent that arrives knowing the user, demonstrates value before asking for anything, and converts attention into commitment inside 5 minutes.

The technology to do this exists. graph8 has it. Most of graph8's competitors have only one or two of the required pieces. The opportunity is to combine them into a single 5-minute experience nobody else can ship. This report is the operating manual for that experience.

Part Two

The Seven Laws of Onboarding

These are the principles that govern every decision below. I've watched dozens of teams violate them and lose deals worth millions. They're not opinions — they're observations.

Figure 5 · The seven laws · at a glance

Seven pictograms representing the seven laws of onboarding. $ LAW 1 Show value before asking LAW 2 Specific beats comprehensive LAW 3 Reveal iceberg, surface the tip LAW 4 Demonstration > description LAW 5 Honest limits build more trust LAW 6 Channel switch = identity bind LAW 7 Revenue intent is the only metric Violate any one and conversion dies. Most teams violate three.

Law 1

Show value before asking for anything

A form, a modal, a tour, a "tell us about yourself" — each one converts at <30% in 2026. The user came to see what the product does, not to do work for the product. The agent must demonstrate value within the first 30 seconds of attention. No exceptions.

Law 2

Specificity beats comprehensiveness

Knowing one thing deeply about the user ("you sell feedback management to PMs, you use Linear, your Stripe is connected, you hired 2 engineers in Q3") creates more wow than knowing ten things generically. The brain processes specific facts as truth signals. Generic claims read as marketing.

Law 3

Reveal the iceberg, surface the tip

The product is doing enormous amounts of work in the background (scrape, enrichment, brand snapshot, intent surge detection, ICP matching, sequence drafting). Mention it. Show only what creates the wow. "I scanned 23 pages, 8 competitors, 4,200 companies in your TAM — here's the one thing you need to see."

Law 4

Demonstration > description

A tour says "here's what we can do." A demonstration says "look what I just did for you." Demonstration converts 5-10× higher than description. Every feature surfaced should be surfaced as a thing that was just done, not a thing that could be done.

Law 5

Honesty about limits builds more trust than artificial confidence

When the data is shallow, the agent should say so. "I don't have intent data on this org yet. I'll watch and ping you when I do." This outperforms fabricated confidence at every cohort. Users have been burned by AI products that lie smoothly. The agent that admits limits is the one users trust with their domain reputation.

Law 6

Channel switch is the new identity binding

The single highest-leverage moment in modern onboarding is connecting the agent to a channel the user already lives in (WhatsApp, Slack, Telegram, Roam, iMessage). This binds the agent to the user across surfaces. The user is no longer "evaluating a tool" — they're talking to an agent that lives in their messaging app. The relationship has changed states.

Law 7

Revenue intent is the only metric that matters

Time spent, features clicked, tours completed, modals dismissed — all vanity. The single onboarding metric that predicts retention, expansion, and referral is revenue intent: did the user take an action that demonstrates intent to extract value? For graph8 the signal is: did they schedule a sequence / book a meeting / click send by minute 5?

Part Three

What Changed in AI Steering, 2024 → 2026

The frontier moved. Anyone designing onboarding today against patterns from 18 months ago is shipping yesterday's product.

Pattern2024 baseline2026 frontier
Starting contextEmpty session, user introduces selfAgent arrives with full enrichment, intent, brand voice pre-loaded
Tool callingHidden ("processing...")Visible as theater ("pulling intent on Linear...") with live progress
SteeringMenus + dropdownsConsequence preview ("I'll do X next; say if you want Y")
MemorySession-scopedCross-channel — same agent in chat, WhatsApp, voice, web
Failure modeApologize and resetAdmit limits, propose watch-and-ping
PersonalizationFilled profile fieldsInferred from signup behavior + enrichment + data
Visual proofStatic screenshotsLive mirror demos (the agent showing itself in action)
Multi-modalityText onlyVoice + text + image + iframe artifact + bridge channel
OrchestrationSingle agent + chatPlanner + executor + critic with tool routing
Latency expectation10-30s acceptableSub-second for routine, narrated for complex

Figure 6 · The three frontier shifts

Three panels comparing the 2024 baseline with the 2026 frontier across computer-use, steering, and modality. PATTERN 1 Computer-use agents 2024 — SUGGEST "I'd suggest you connect HubSpot…" 2026 — JUST DID IT Pulled HubSpot · 412 contacts Enriched 47 · matched ICP Draft sequence ready ✓ PATTERN 2 Consequence preview 2024 — MENU What would you like to do? ○ Build a sequence 2026 — DECLARE I'm about to draft a sequence to Sara Chen · Linear · CTO Say wait if you want a different contact. PATTERN 3 Voice + visual + bridge agent canvas 🎙 voice ▦ visual 📱 bridge graph8 has all 3 · no competitor does

Three frontier patterns specifically matter for graph8:

1. Computer-use and operator agents

Anthropic's computer use + OpenAI's operator pattern + Claude's MCP have made it possible for agents to actually take actions across tools during the conversation. The 2024 demo was "agent suggests next step." The 2026 demo is "agent ran the next step and shows you the result." graph8's agent_runtime.tool_dispatch + 113 g8_* MCP tools puts it ahead of most of its competitors on this dimension — but only if the onboarding shows it.

2. Steering through consequence preview, not menus

The 2024 pattern was "what would you like to do? [options]". The 2026 pattern is "I'm going to do X. Say wait if you want Y." This is psychologically very different — it transfers initiative from user to agent. The user can always interrupt, but the default is forward motion. Apollo and Outreach are still on the 2024 pattern. Clay is partially on the new pattern. Revo and graph8 have the architectural building blocks but haven't fully embraced it.

3. Voice + visual + bridge in a single onboarding

The new tier of AI onboarding combines voice (Vapi/Retell-style), visual artifact generation (landing pages, sequence drafts), and bridge connection (WhatsApp/Slack) into one continuous flow. graph8 has all three pieces (voice module + landing_pages + notifications/bridge). No competitor has shipped all three integrated yet. This is graph8's window.

Part Four

The Five Archetypes Who Arrive

Earlier sketches had eight archetypes. After more analysis: five distinct ones — each with a fundamentally different job-to-be-done. The onboarding must work for all five without asking any of them who they are.

Figure 7 · Five archetypes · same spine, different peaks

Five archetype pictograms: Tactical Operator, Strategic Owner, Builder, Skeptic Evaluator, Reseller. ×N TACTICAL SDR · AE · CM peaks at Act 2+3 "pipeline today" STRATEGIC Founder · CEO · VP peaks at Act 2+4 "revenue clarity" BUILDER RevOps · MOps · Eng peaks at Act 3 "show me the seams" SKEPTIC Comparison shopper peaks at Act 2+4 "verdict in 5 min" RESELLER Agency · Fractional peaks at Act 1+3 "replicate × N clients" The agent doesn't ask who arrived. Signup behavior + enrichment infer it.

1. The Tactical Operator — SDR, AE, Campaign Manager

Job
Pipeline. Today.
Default state
Quota pressure. Time-starved.
Impressed by
Named accounts, hot contacts, sequences ready to send, dialer queue.
Suspicious of
Anything "strategic" or "vision-level."
Inference signals
signup_entry_point = sequencer / dialer / contacts; mid-size company; non-executive title.

2. The Strategic Owner — Founder, CEO, VP Sales/Marketing

Job
Revenue clarity + GTM motion + team enablement.
Default state
Limited time, high signal-to-noise demand.
Impressed by
TAM + pipeline math, multi-tool consolidation pitch, "GTM team in a box."
Suspicious of
Feature soup. Wants the picture, not the parts.
Inference signals
Small co + "Founder/CEO/Owner" in title; or large co with executive title.

3. The Builder — RevOps, Marketing Ops, Power user

Job
Infrastructure. Integrations. Customization. Data flow.
Default state
Skeptical, comparing feature-by-feature.
Impressed by
API surface, MCP tools, workflow builder, skill authoring, observability.
Suspicious of
Magic. Wants to see the seams.
Inference signals
signup_entry = developer/workflows; martech stack in enrichment; "Ops/RevOps/Engineering" title.

4. The Skeptic Evaluator — Trial-shopper, comparison

Job
Form a verdict in <5 minutes.
Default state
Lowest commitment. Will close the tab fast.
Impressed by
Anything visceral and uniquely-graph8 (intent+ICP, displacement, brief, replacement reveal).
Suspicious of
Generic claims, comparison-shopping language, "the only platform that..."
Inference signals
Signup from competitive comparison search (utm/referrer); no follow-through on confirms.

5. The Reseller — Agency, Consultant, Fractional

Job
Speed-to-set-up. Demoable output. Replicate across clients.
Default state
Doing this on behalf of someone else.
Impressed by
One-click provisioning, white-label potential, artifacts that hold up to client review.
Suspicious of
Tools requiring per-client configuration depth.
Inference signals
Agency tab access; multi-org session history; consultant-y email.
The structural insight: the 5 archetypes converge on the same 4-act spine but peak at different beats. The agent doesn't ask; the data adapts.

Part Five

graph8's 1,819 Features Through the Onboarding Lens

Most onboarding teams treat the feature inventory as something to surface. That's backwards.

The job is to hide 1,790 features and reveal 30. Here's how to choose the 30.

Figure 2 · The cut · 1,819 → 30

A four-stage funnel narrowing the 1,819-feature inventory down to 30 surfaced in onboarding. INVENTORY ACTIVE MODULES WOW LAYER SURFACED IN ACT 1–4 1,819 features in the platform Studio · 212 Developer · 130 Roadmap · 131 Settings · 91 Admin · 75 Marketplace · 64 Profile · 24 · Help · 16 cut −870 earned, not onboarding ~949 in-product, daily-use CRM · sequences · enrichment analytics · voice · landing pages cut −899 surface contextually ~50 wow-eligible Tier S + Tier A + Tier B choose 30 archetype-tuned 30 in minute 5 100% baseline 52% after first cut 2.7% qualify as wow 1.6% make minute-5 The wedge is what you cut, not what you ship. Most B2B onboarding tries to demonstrate ~200. graph8's job is to demonstrate 30.

The Cut — features that DON'T belong in onboarding

BucketFeaturesOnboarding-relevant
Settings910 — surface during use, not setup
Admin750
Developer Platform1300 (except The Builder archetype)
Profile241 (the bridge connection)
Roadmap1310
Help160 — surface contextually
Feedback30
Most of Studio212~5 (workbench is background; only confirm cards surface)
Most of Marketplace640 in v1
Analytics161 (TAM data, used in act 4)

~870 features (48% of the inventory) should NEVER appear in onboarding. They're the long tail the product earns its right to over months of use.

The Keep — the ~50 features that create maximum wow

Tier S

Unfakeable, uniquely graph8

FeatureInventory locationOnboarding beat
intent.visitor_signals_to_outboundSignals/VisitorsAct 2 + 3 (after pixel)
intent.account_research_briefSignals/IntentAct 2 beat 9
intent.competitor_keywords + keyword_surge_detectionSignals/IntentAct 2 beat 8 (displacement)
intent.intent_search_companies (ICP cross-ref)Signals/IntentAct 2 beat 7 (double whammy)
16-vertical replacement narrativeCross-cuttingAct 4 beat 16
agent_runtime.copilot_bridge (5 platforms)AgentsAct 1 beats 1 + 4
agent_runtime.agent_canvas w/ tool_dispatchAgentsAlways-on
Tier A

Strong, defensible

FeatureBeat
company_enrichment (phone/email/HQ)Act 1 beat 3 (unified snapshot)
product_inventoryAct 1 beat 3
cb_org_brand_snapshot (logo + colors)Act 1 beat 2
competitor_discoveryAct 2 beat 6
competitive_teardown (weakness mining)Embedded in beat 9/11
brand_voice + brand_briefAct 2 beat 5 + Act 3 beat 11
icp_research + persona_researchAct 2 beat 7
messaging_houseAct 3 beat 15 (sequence)
developer_api.list_contactsAct 2 beat 10 (hot contact)
mailbox warmup + domain purchaseAct 3 beat 14
embed_widget + AI receptionist (pixel)Act 3 beat 12
cb_sequences insert flowAct 3 beat 15
tam_research + market_insightsAct 4 beat 17 (30-60-90)
Tier B

Optional, persona-specific surfacing

  • voice (Twilio + dialer profile) — Tactical Operator scope probe
  • forms — Builder scope probe
  • newsletter — Strategic Owner scope probe
  • appointments — integrated into sequence after scope probe
  • landing_pages — Strategic Owner / Reseller, post-wrap-up
  • quotes + esign — AE persona, post-wrap-up
  • crm_views + deals — Builder, post-wrap-up
  • workflows + skill_authoring — Builder, post-wrap-up
Tier C

Earned features — don't show in onboarding

  • The 1,150+ Settings / Admin / Developer / Roadmap / Help features — surface contextually as the user explores
  • Workflow builder + 40+ node types — too much for minute 5; surface in week 2
  • Sales coach + voice AI scoring — for users with active dialer activity
  • Marketplace SDR + Agency multi-tenancy — vertical-specific
  • Most of CRM internals — surface when they import / create their first deal

Why this cut works

graph8's competitive context names 16 verticals where it challenges 50 distinct competitors. No onboarding can survive trying to demonstrate 16 verticals. But it CAN demonstrate the 5 that matter for revenue:

  1. Prospecting + Data — vs Apollo / Clay / ZoomInfo
  2. Intent Data — vs 6sense / Bombora
  3. Sequencer + Inbox — vs Outreach / Salesloft / Smartlead
  4. Voice / Dialer — vs Orum / Nooks (Tier B optional)
  5. Web Chat / AI Receptionist — vs Intercom / Drift (via pixel)

The act-4 replacement reveal explicitly names what was just replaced. That is the moat narrative.

Part Six

The Conversation Engine — How Steering Actually Works

The agent in onboarding is not "a chatbot." It's a planner-executor-critic loop that's been pre-loaded with everything graph8 knows about the user.

Figure 8 · This isn't a chatbot · planner → executor → critic

A loop diagram showing the planner-executor-critic architecture with five steering moves branching off. PLANNER decides next beat archetype-aware EXECUTOR tool_dispatch 113 g8_* tools CRITIC specificity gate regenerate if generic plan → call result → check approved → speak MOVE 1 · REVEAL "here's what I know" MOVE 2 · CONSEQUENCE "I'll do X — say wait" MOVE 3 · CONFIRM "✓ / ✎ / ✗ this card?" MOVE 4 · HONEST LIMIT "I'll watch and ping you" MOVE 5 · CLIMACTIC REVEAL "I just did Apollo + 6sense + Outreach…" FIVE STEERING MOVES branch off the loop at different beats PRE-LOADED CONTEXT enrichment · intent · brand voice · ICP · 44 studio docs

The five steering moves

Every interaction is one of these five. Mastering them is mastering modern AI onboarding.

Move 1

The Reveal

Agent surfaces data it already has. "Here's what I know about you." User reacts. Used for: company snapshot, brand voice, competitor list, double whammy.

Move 2

The Consequence Preview

Agent declares its next action and waits for interruption. "I'm going to draft a sequence to Sara at Linear. If you want a different contact, say so." User accepts default or steers. Used for: scope probe outcomes, default selections.

Move 3

The Confirmation

Agent presents work it just did and asks for ✓ / ✎ / ✗. "Brand voice: warm, direct, technical. Sound right?" User confirms or edits. Used for: every confirm card.

Move 4

The Honest Limit

Agent admits it can't do something yet. "I don't have intent data on this org. I'll watch and ping you in WhatsApp." User trusts more, not less. Used for: cold-start scenarios.

Move 5

The Climactic Reveal

Agent waits until trust is built, then drops the big claim. "In the last 5 minutes I just did the work of Apollo + 6sense + Outreach + Clay + Crayon + Calendly + Customer.io + MailReach." Used for: act-4 replacement reveal only.

What graph8 needs to build to nail this

The current agent_runtime already has the planner-executor architecture. What's missing for onboarding-grade steering:

  1. Consequence preview tone, not menu tone. The planner system prompt needs an explicit rule: prefer "I'll do X next. Say wait to change." over "Would you like X or Y?"
  2. Critic on wow cards: when a wow card is about to render, a critic agent checks: is the body specific enough? Does it have a named entity? If not, regenerate.
  3. Cross-channel state persistence: the same conversation in chat and WhatsApp must share state.
  4. Tool-call narration: when the agent invokes wow_account_research_brief, surface a "Pulling research on Linear…" status. Users tolerate latency they can see; they leave on latency they can't.

Part Seven

The Five Adaptive Playbooks

The same 18-beat 4-act spine. Different framing, different peaks, no user-facing role prompt. The data drives the adaptation.

Figure 4 · Where each playbook peaks

Heatmap of intensity per act per archetype. Each archetype peaks at different beats. ACT 1 Identity ACT 2 Demand ACT 3 Action ACT 4 Reveal Tactical Operator SDR · AE · CM Light HEAVY whammy · brief · contact HEAVY sequence + schedule Light · single CTA Strategic Owner Founder · CEO · VP Medium · confirm HEAVY TAM + intent volume Medium PEAK replacement + math Builder RevOps · MOps · Eng Medium · stack Medium · sources HEAVY pixel + MCP + workflow Light · API tease Skeptic Evaluator Comparison shopper Light · 30s only PEAK whammy climax Medium HEAVY tools named + math Reseller / Agency Consultant · fractional HEAVY multi-tenant nod Medium · depth HEAVY seq + LP artifacts HEAVY Agency surface INTENSITY Light Medium Heavy Climactic peak Same 4 acts. Five trajectories. The data adapts the framing — the user never picks a role.

A. Tactical Operator — SDR / AE / Campaign Manager

Act 1Minimal
"I know your stack, your products. Hi."
Act 2Heavy
Double whammy + displacement + research brief + hot contact. Spend 90+ seconds here.
Act 3Heavy
Sequence drafted to a real prospect. Schedule click is the commitment moment.
Act 4Light
Single CTA to the sequence. Skip the "GTM team" pitch.

B. Strategic Owner — Founder / CEO / VP

Act 1Standard
Confirm. Acknowledge their identity directly.
Act 2Adapted
Light on prospects, HEAVY on TAM + intent volume. "4,200 in your TAM. 870 showing intent monthly."
Act 3Standard
Standard.
Act 4Heavy
Replacement reveal + 30-60-90 pipeline math. The strategic vision lands here.

C. Builder — RevOps / Marketing Ops / Power user

Act 1Standard
Confirm + show stack reveal. "I see HubSpot + Marketo + Stripe."
Act 2Technical
Standard, with extra depth on data sources. "Intent data from optimized_signals_v1 + reverse-IP enrichment."
Act 3Heavy
Pixel install with snippet visible + MCP tool catalog tease + workflow builder glimpse. "Want to wire this into HubSpot?"
Act 4Light
"Developer access — API keys + MCP — ready when you are."

D. Skeptic Evaluator

Act 1Compressed
Faster. 30s.
Act 2Heavy
Climax on double whammy. Quote named companies they recognize ("Linear — yes, that Linear").
Act 3Standard
Standard.
Act 4Heavy
Replacement reveal names tools they're comparing against. Close: "Want to see this on your competitor's domain too?"

E. Reseller / Agency

Act 1Standard
Mention multi-tenant capability subtly.
Act 2Standard
Show data depth that will hold up under client review.
Act 3Heavy
Sequence draft + landing page draft. Artifacts they can show clients.
Act 4Heavy
Agency surface mentioned. "Want to manage 10 clients this way? Onboarding takes 5 minutes per."

Part Eight

The Wow Moment Hierarchy

Not every moment is equal. This is the impact ranking from the consultant chair across the GTM tools I've onboarded.

Figure 3 · The wow pyramid · 17 named moments, ranked

A pyramid showing the 17 wow moments ranked S to D, with Tier D explicitly cut from onboarding. Workflow tour · feature matrix · pricing link · video demo · settings wizard TIER D · 3–4/10 CUT FROM ONBOARDING Welcome with name · brand kit · scope probe · single CTA · honest "still cooking" TIER C · 5–6/10 TABLE STAKES "I read 23 pages" · 8 competitors auto-found · sample subject · hot contact card · 30-60-90 math TIER B · 7–8/10 TRUST-BUILDERS Unified snapshot · replacement reveal · pixel live mirror Domain warmup · brand voice tone preset TIER A · 9/10 DIFFERENTIATORS Intent + ICP whammy Bridge linkup · Research brief Sequence to hot contact Displacement signal TIER S · 10/10 REVENUE-CHANGING THE WEDGE Most onboarding ships from the bottom. The wedge is at the top.
Tier S

10/10 · Revenue-changing

These reshape the user's view of what's possible. Drop any of these and the conversion rate visibly dies.

  1. Intent + ICP double whammy — named companies hitting both signals · Act 2 beat 7
  2. Bridge linkup with tactile reply — text "hi" from your phone, agent answers · Act 1 beat 4
  3. 1-page account research brief on the #1 hot account · Act 2 beat 9
  4. Sequence drafted to a named hot contact, ready to schedule · Act 3 beat 15
  5. Displacement signal — "companies shopping for {your competitor} now" · Act 2 beat 8
Tier A

9/10 · Strong differentiators

Each one alone could be the headline of a competing point-tool's marketing site.

  1. Unified company snapshot — phone/email/HQ/products/keywords in one card · Act 1 beat 3
  2. Multi-tool replacement reveal — "Apollo + 6sense + Outreach + Clay just replaced" · Act 4 beat 16
  3. Pixel install with live IP→company mirror · Act 3 beat 12
  4. Domain purchase + warmup live during onboarding · Act 3 beat 14
  5. Brand voice tone preset — writes like you for the rest of the session · Act 2 beat 5
Tier B

7-8/10 · Trust-builders

Necessary but not memorable. The supporting cast.

  1. "I read 23 pages of your site" · Act 1 beat 2
  2. 8 competitors auto-discovered with edit chips · Act 2 beat 6
  3. Sample subject line in your voice · Act 3 beat 11
  4. Hot contact spotlight with phone + email + intent · Act 2 beat 10
  5. 30-60-90 honest pipeline math · Act 4 beat 17
Tier C

5-6/10 · Table stakes

Have to be there, never the wow.

  • Welcome message with name
  • Brand kit (logo + colors) auto-extracted
  • Scope probe with 3 buttons
  • ONE primary CTA at wrap-up
  • "Still cooking" honest follow-up list
Tier D

3-4/10 · Cut from onboarding

  • Workflow builder tour
  • Feature matrix
  • Pricing page link
  • "Watch this 5-minute demo video"
  • Quick tour with 12 modals
  • Settings setup wizard

Every Tier D element in your onboarding kills more deals than it converts. If a competitor has them and you cut them, you win.

Part Nine

Metrics — The Ones That Matter, The Ones That Don't

Figure 9 · The revenue intent funnel · the five metrics that matter

A five-stage revenue intent funnel with target conversion rates at each step. STAGE 0 Signup completed baseline · 100% STAGE 1 Bridge linked target ≥ 40% STAGE 2 Wow card confirmed leading signal STAGE 3 Sequence scheduled @ min 5 target ≥ 40% STAGE 4 14-day activation target ≥ 50% STAGE 5 · 90-day pipeline attribution 0 1 2 ★ 4 THE PRIMARY METRIC revenue intent · the only one that matters ✗ IGNORE time spent features clicked modals dismissed tour completion session events "engagement" scores

✓ Track ruthlessly

  • Sequence-schedule rate at minute 5 — primary revenue-intent metric. Target: ≥40%.
  • Bridge-link success rate — proxy for identity binding. Target: ≥40%.
  • Per-act drop-off (4 transitions) — diagnostic. Any act losing >25% needs surgery.
  • 14-day activation — did they send the sequence after warmup? Target: ≥50%.
  • 90-day revenue attribution — pipeline from onboarding-staged sequences.

✗ Ignore

  • Time spent in onboarding (longer is worse, not better)
  • Features clicked (vanity)
  • Modal dismissals
  • Tour completion
  • Total session events
  • "Engagement" composite scores

The wrong metrics make teams optimize for the wrong things. A team that ships "make users click more" interventions is a team about to lose its onboarding entirely to a competitor that ships fewer interactions but more revenue intent.

Part Ten

The Implementation Map — Where graph8 Stands Today

The current ProactiveSetupWorkflow + the v2 PRD are 70% of this report's vision. Here's what's done, what's in flight, what's missing.

Figure 10 · The roadmap · shipped, in-flight, next

A three-band timeline showing what shipped May 19-22, what is in flight, and what is next. SHIPPED · MAY 19–22 24 QA rounds · merged to main IN-FLIGHT · NOW PR #7791 merged · issue #7788 open NEXT · IN ORDER scoped, not yet built ✓ SHIPPED · 6 items ✓ 5 wow cards (brand · competitors · intent · ICP · LP) ✓ Bridge architecture across 5 platforms ✓ Studio doc generation pipeline ✓ Agent Canvas with MCP tool access ✓ Polling cards: generating → complete ✓ Brand-themed wrap-up reveal ⟳ IN-FLIGHT · 9 items ⟳ 4-act narrative restructure (v2) ⟳ Per-doc confirm cards (22) ⟳ Intent + ICP double whammy ⟳ Account research brief ⟳ Displacement signal ⟳ Replacement reveal · pixel mirror · 30-60-90 math → NEXT · 5 items 1. Phase 1+2 of v2 PRD 2. Critic agent on wow cards 3. Tool-call narration in chat 4. Adaptive role tuning in planner 5. Cross-channel state unification 70% of the vision is shipped. The remaining 30% is one merged PRD + 11 well-scoped tools away.

Done — shipped 2026-05-19 → 2026-05-22 (24 QA rounds)

  • 5 wow cards lifecycle (brand kit, competitors, intent, ICP, landing page)
  • Bridge architecture across 5 platforms
  • Studio doc generation pipeline (1,800+ features)
  • Agent Canvas with MCP tool access
  • Polling cards with generating → complete lifecycle
  • Brand-themed wrap-up reveal

In-flight — PR #7791 merged, issue #7788 agent-ready

  • v2 4-act narrative restructure
  • Per-doc confirm cards (22 docs)
  • Scope agent + setup actions
  • Intent+ICP double whammy
  • Account research brief (NEW)
  • Displacement signal (NEW)
  • Replacement reveal (NEW)
  • Pixel install with live mirror
  • Honest 30-60-90 math

Missing for the full vision

  • Critic agent on wow cards — automatic regeneration if a wow card lacks specificity
  • Cross-channel state persistence — full continuity across chat + WhatsApp + voice
  • Tool-call narration — "Pulling research on Linear…" status during async tool calls
  • Adaptive playbook tuning — planner-prompt-tuning by inferred role
  • Voice-AI handoff in onboarding — for users picking "voice" as primary channel
  • Critic on sequence draft — re-run if subject line is generic
  • Real-time tool theater — visible iceberg of work done

What to ship next, in order

  1. Phase 1 + 2 of v2 PRD (Foundation + Revenue-act core) — Huzaifa's queue right now
  2. Critic agent on wow cards (small wrapper around existing handlers)
  3. Tool-call narration in the chat surface (FE work)
  4. Adaptive role tuning in the planner system prompt (configuration, not code)
  5. Cross-channel state unification (data model work, mostly already there)

Part Eleven

The Brutal Closing Truth

Figure 11 · The moat · 10 components × 11 vendors

A competitor matrix showing graph8 owns all ten onboarding components while no other named vendor owns more than three. Bridge Agent Canvas Intent data ICP research Account brief Pixel install Domain warmup Voice / dialer Landing pages Replacement VENDOR SCORE / 10 graph8 ✓✓✓ ✓✓✓ ✓✓✓ ✓ 10 / 10 Apollo 2 / 10 Clay 2 / 10 ZoomInfo 1 / 10 Outreach 1 / 10 Salesloft 1 / 10 6sense 2 / 10 Bombora 1 / 10 Orum 1 / 10 Drift 2 / 10 Intercom 2 / 10 LEGEND owns + integrated partial / point-tool addressable, not native No competitor owns more than 3 of the 10.

Most B2B SaaS onboarding in 2026 is still doing what 2018's onboarding did, with a chat widget bolted on. The chat widget makes it slightly worse, not better, because it introduces an agent that doesn't know the user and can't do anything useful.

The opportunity in front of graph8 is not "improve onboarding by 10%." It's to build the first onboarding that genuinely earns the words magical. Not by adding more features. By revealing the right 30 features at the right moments inside a 5-minute conversation that demonstrates the product is already working for the user.

This is buildable. The pieces exist. graph8 already has Bridge + Agent Canvas + Intent + ICP + Research Brief + Replacement narrative + Pixel + Domain warmup + Voice AI + Landing Pages — every component the perfect 2026 onboarding needs.

No competitor in the inventory's 50-vendor competitive matrix has all of these. The differentiator is not the components. It's the sequencing.

The team that ships this captures the next decade of the GTM platform market. That team is one merged v2 PRD + 11 well-scoped tools + careful execution away.

"graph8 onboarding doesn't introduce a tool. It introduces an agent that already knows you, already did the work, and is ready to send your first sequence by minute 5. Everything else in this category is still asking users to introduce themselves."

— The wedge. Build to it.