What a Customer Journey Mapping Tool actually does
Ingests call-center transcripts, support tickets, NPS verbatim, and session data, then clusters them into emergent journey stages, personas, and pain points with AI-generated stakeholder narratives.
An AI journey mapping platform processes unstructured CX data (call transcripts, support tickets, survey verbatim, session-replay events) and uses embedding-based clustering (text-embedding-3-large + HDBSCAN/UMAP) to discover emergent journey stages without predefined assumptions. Claude Sonnet 4.6 then drafts narrative descriptions of each persona and pain point for stakeholder presentations. This is insight and visualization — not action-taking (see the orchestration brief for that distinction).
The 2026 market context: Smaply, UXPressia, and Custellence are established journey mapping SaaS platforms but none have white-label reseller programs. The 'AI angle' in 2026 is not the map itself — it's the automation of the analysis phase. Traditionally, a CX consultant spent 2–3 weeks manually coding transcripts and interviews to generate journey maps. With this stack, that coding phase takes hours, freeing consultant time for synthesis and client engagement. The platform's value proposition is 'reduce journey-map discovery from 3 weeks to 3 days' — not 'replace the CX consultant.'
AI capabilities involved
Touchpoint clustering from unstructured CX data
Emergent persona generation from behavioral clusters
Pain-point narrative drafting for presentations
Verbatim sentiment and topic tagging
Journey diagram generation (Mermaid/SVG)
Who uses this
- Boutique CX strategy consultancies serving 5–20 enterprise clients per year with journey-mapping engagements
- In-house CX teams at large enterprises wanting to automate the data-collection-to-insights phase of journey projects
- Marketing and growth agencies offering customer-experience audits as a service
- Research operations teams at B2C brands with high call-center and NPS data volume
SaaS alternatives on the market
Real products you can sign up for today — with current 2026 pricing, honest pros and cons.
UXPressia
CX teams wanting a collaborative visual mapping tool for manually-crafted journey maps.
Free plan (1 project, 1 user)
$36/user/mo (Pro)
Pros
- +Best-in-class visual journey map builder with drag-and-drop interface.
- +Collaborative — multiple stakeholders can view and comment on maps.
- +Persona and journey map templates reduce setup time.
Cons
- −No white-label — UXPressia branding throughout.
- −No AI-automated analysis — data must be manually entered into the map.
- −Per-seat pricing grows with team size.
Smaply
Service designers and CX consultants doing manual journey mapping with client stakeholders.
Free trial
$39/user/mo
Pros
- +Strong journey map and persona visualization.
- +Good for service blueprint creation alongside journey maps.
- +Export to PDF and PowerPoint.
Cons
- −No white-label.
- −No AI-automated data analysis.
- −Similar price point to UXPressia without clear differentiation.
Quadient
Large enterprise CX teams needing journey mapping integrated with omnichannel communication platforms.
None
Enterprise, quote-based
Pros
- +Enterprise-grade journey orchestration features beyond mapping.
- +Partial API access for data integration.
- +Multi-channel customer communication alongside journey visualization.
Cons
- −No white-label reseller program.
- −Enterprise pricing — not accessible for boutique consultancies.
- −Overkill for pure journey mapping use cases.
The AI stack
The journey mapping stack is embedding-heavy: the quality of persona clustering depends on embedding quality and clustering parameter tuning. Spend more time on the pgvector schema and HDBSCAN parameters than on LLM prompt engineering — the clustering determines map quality; the LLM narrativizes what clustering reveals.
Text embedding and clustering
Converts transcript segments, ticket summaries, and survey verbatim into vectors, then clusters them into emergent journey stages and touchpoints
text-embedding-3-large
$0.13 / M tokensPrimary embedding for clustering — quality matters here as it determines persona granularity
text-embedding-3-small
$0.02 / M tokensInitial exploration and large-volume batch embedding where cost matters
Our pick: text-embedding-3-large for final production embedding; text-embedding-3-small for initial exploration and prompt iteration. The $0.11/M premium is worth it for client deliverable quality.
Verbatim tag classification
Tags each transcript segment with journey stage (awareness, consideration, onboarding, support, churn risk), sentiment, and topic before clustering
Claude Haiku 4.5
$1 / $5 per M tokensAll pre-embedding classification — cheap enough to run on every transcript segment
DeepSeek V4 Flash
$0.14 / $0.28 per M tokensInternal data science pipelines processing very large anonymized datasets
Our pick: Claude Haiku 4.5 for all verbatim tagging — the data-residency assurance justifies the cost premium over DeepSeek when processing real customer PII.
Persona narrative generation
Generates narrative descriptions of each emergent persona cluster for stakeholder presentations
Claude Sonnet 4.6
$3 / $15 per M tokensAll persona narrative generation — quality directly affects client deliverable impact
GPT-5.4
$2.50 / $15 per M tokensOpenAI-stack teams or where tight cost control is needed at scale
Our pick: Claude Sonnet 4.6 for all persona narratives. The $0.20–$0.50 total cost for a 5-persona journey map narrative is negligible against a $20K consulting engagement.
Reference architecture
Journey mapping has a batch-analysis pipeline shape: all data is ingested, tagged, embedded, clustered, and narrativized before the client sees results. There is no real-time component. The hardest challenge is clustering quality — HDBSCAN parameter tuning (min_cluster_size, min_samples, epsilon) for CX verbatim requires domain expertise and iteration.
CX data ingestion (transcripts, tickets, NPS verbatim, session events)
CSV upload or API connector (Zendesk, Salesforce, Qualtrics) + SupabaseRaw data ingested into raw_cx_data table. Source type tagged (call, email, chat, NPS, session). Client project isolated by project_id.
Segmentation into analysis units
Trigger.dev batch job + text chunkingLong call transcripts split into 200–500 word segments at natural break points (speaker turns for transcripts, paragraphs for tickets). Each segment becomes an analysis unit in cx_segments table.
Classification: journey stage, sentiment, topic
Trigger.dev + Claude Haiku 4.5Haiku classifies each segment: journey_stage (awareness/consideration/onboarding/activation/support/churning), sentiment, primary_topic. Stored in segment_classifications. Cost: ~$0.001 per segment.
Embedding generation
Trigger.dev + text-embedding-3-largeEach segment embedded. Vectors stored in Supabase pgvector segments_embeddings table. Cost: ~$0.001 per 100 segments.
Clustering (HDBSCAN via Python service)
Python Edge Function or Modal serverless + HDBSCANRuns HDBSCAN on the embedding vectors to discover natural clusters. Cluster labels assigned to each segment. Noise points (segments that don't fit any cluster) flagged for review. Number of clusters typically 5–15 for a typical CX dataset.
Cluster labeling and persona narrative generation
Trigger.dev + Claude Sonnet 4.6For each cluster: sample 20 representative segments, pass to Sonnet with prompt: 'These are customer verbatims from the same cluster. Name this cluster (5 words), describe the customer type (100 words), and describe their key pain point (100 words).' Cost: ~$0.10 per persona narrative.
Journey map visualization rendered for client review
Next.js dashboard with custom Recharts/SVG journey mapJourney stages displayed as horizontal flow. Personas displayed as swim lanes per stage. Pain points shown as callout cards. Client can rename clusters, reorder stages, and export as PDF.
Estimated cost per request
~$0.001 per segment classified; ~$0.001 per 100 segments embedded; ~$0.10 per persona narrative. Total AI cost for a 10,000-segment CX dataset producing 10 personas: ~$25.
Cost calculator
Drag the sliders to model your actual usage. The numbers update in real time so you can stress-test economics before writing a single line of code.
Cost model for a CX consultancy running 10 journey-mapping engagements per year, each processing ~5,000 CX data segments.
Estimated monthly cost
$75.50
≈ $906 per year
Calculator notes
- Total AI cost per 5,000-segment engagement: ~$5 classification + $0.50 embedding + $1 narratives = ~$6.50 per engagement.
- Against a $20K+ consulting engagement fee, AI cost is ~0.03% of revenue — the build and infrastructure investment dwarfs API cost.
- Clustering computation (HDBSCAN via Modal) adds ~$0.10–$0.50 per engagement depending on dataset size.
- PDF export of journey maps may require a headless Chrome service or Puppeteer — budget $10–$20/mo for a Browserless.io subscription.
Build it yourself with vibe-coding tools
10-day MVP delivers: CSV upload of CX verbatim, AI classification of journey stage and sentiment, embedding-based clustering, and Sonnet-generated persona narratives displayed in a branded dashboard. Production clustering quality requires Python iteration beyond the initial build.
Time to MVP
10 days (classification + embedding + narrative MVP); +2–3 weeks for HDBSCAN clustering quality
Total cost to MVP
$25 Lovable Pro + $40 Anthropic credits + $0.50 initial embedding batch = ~$65 total
You'll need
Starter prompt
Build a white-label AI customer journey mapping tool using Vite + React + TypeScript + Tailwind CSS with Supabase backend. Consulting clients upload CX data and get AI-generated journey maps. Core features: 1. Project workspace: create a named project (client name, project type: B2C/B2B). All data scoped to project. 2. CSV upload: accept CSV with columns: text, source (call|chat|email|nps|survey), date, optional_customer_id. Insert into cx_segments table. 3. Classification batch: a Supabase Edge Function iterates cx_segments (unclassified), calls Claude Haiku 4.5 with structured output: {journey_stage: awareness|consideration|onboarding|activation|support|churn_risk, sentiment: positive|neutral|negative, primary_topic: string, confidence: 0-100}. Store in segment_classifications. 4. Embedding (simplified for MVP): Edge Function calls OpenAI text-embedding-3-large for each unembedded segment. Store vector in segments_embeddings (using pgvector). 5. Simplified clustering (for MVP): use pgvector cosine similarity to find the 3 nearest neighbors of each segment. Group by journey_stage + primary_topic combination as a proxy for personas. Production will use Python HDBSCAN. 6. Persona view: display one card per topic cluster, showing: cluster name (top keyword), segment count, sentiment distribution (bar), 5 sample quotes. 7. Narrative generation: 'Generate Narratives' button calls Sonnet 4.6 Edge Function with the top 10 segments per cluster. Returns: {persona_name, persona_description: 100 words, key_pain_point: 50 words}. Display as persona cards. 8. Journey map view: horizontal flow showing journey_stage columns with persona count per stage. Database: projects, cx_segments, segment_classifications, segments_embeddings (pgvector), personas, persona_narratives.
Paste this into Lovable
Follow-up prompts (run in order)
- 1
Replace simplified clustering with Python HDBSCAN: create a Modal serverless Python function that reads embeddings from Supabase pgvector, runs HDBSCAN (min_cluster_size=15, min_samples=5), and writes cluster_label back to segments_embeddings. Trigger via a Supabase Edge Function after all embeddings are complete. This gives production-quality clusters.
- 2
Add call-transcript connector: integrate with Gong or Chorus API (or Zoom Recordings API for smaller clients) to automatically import call transcripts into cx_segments without manual CSV upload. Speaker diarization via Deepgram Nova-3 separates customer vs. agent speech — only customer speech segments should go into the analysis.
- 3
Add journey map PDF export: use Puppeteer via Browserless.io to render the journey map dashboard as a high-resolution PDF. Include client logo, brand colors, and project name in the header. Add a page per persona with their quote samples and narrative. Export as branded client deliverable.
- 4
Add cross-channel attribution overlay: add a second data type 'session_events' (pages visited, time spent, funnel steps completed). Join session events to cx_segments by customer_id where available. Display a heatmap of digital touchpoints alongside each journey stage in the map view.
Expected output
A working branded tool where you upload 500 customer verbatims, see them classified by journey stage, grouped into rough persona clusters, and read Sonnet-generated 100-word persona descriptions. Suitable for a client demo conversation.
Known gotchas
- !Supabase pgvector cosine similarity clustering (used in the MVP) produces rough groupings based on single nearest-neighbor relationships — it's not HDBSCAN and won't produce clean journey personas. Production requires a real clustering algorithm via Python/Modal.
- !HDBSCAN parameter tuning is iterative: min_cluster_size too high merges distinct personas; too low creates noise. Expect 1–2 weeks of parameter iteration with a real CX dataset before cluster quality is client-presentable.
- !Two-party consent for call-center recordings: many US states require all-party consent for phone recording. Before ingesting call transcripts, confirm that the client's call recordings were collected with proper consent disclosures.
- !Lovable's Edge Function 30-second timeout means embedding 5,000 segments in one call will fail. Use Trigger.dev with batches of 100 segments per job iteration.
- !Cluster naming via Sonnet is an art: Sonnet may produce generic cluster names ('dissatisfied customers') rather than insightful ones ('first-time buyers overwhelmed by onboarding'). Build cluster-naming as an interactive process where the consultant can edit Sonnet's suggestion.
- !EU GDPR verbatim processing: customer verbatim from EU users may be subject to GDPR. Anonymize or pseudonymize customer identifiers before ingesting into the AI pipeline. Do not pass customer names, emails, or account IDs into the LLM context.
Compliance & risk reality check
Journey mapping tools face two primary compliance areas: call-center recording consent and GDPR data-processing rights for EU verbatim.
Two-party consent for call recordings
12 US states (California, Connecticut, Florida, Illinois, Maryland, Massachusetts, Michigan, Montana, Nevada, New Hampshire, Oregon, Washington) require all-party consent for telephone recording. Processing call transcripts from these states without confirmed all-party consent creates legal exposure for the client and potentially for the platform.
Mitigation: Require clients to confirm that all call recordings were collected with proper consent disclosures before uploading transcripts. Include this as a terms-of-service obligation. For California specifically, consent must be 'explicit' — a general 'calls may be recorded for quality purposes' disclosure may not be sufficient for AI analysis use.
GDPR — EU customer verbatim as personal data
Customer verbatim text from EU citizens may be personal data under GDPR if it is linked to an identifiable individual. Processing this data through an AI API (including Anthropic's API) requires a lawful basis and must be disclosed in the privacy notice. EU data subjects have the right to erasure — if they request deletion, their verbatim must be removed from the platform.
Mitigation: Anonymize or pseudonymize customer verbatim before ingestion where possible. Remove or hash customer identifiers (names, emails, account IDs) before passing text to the AI API. Document the lawful basis for processing (legitimate interest in improving CX analytics). Sign DPAs with Supabase and Anthropic.
AI training opt-out — CX verbatim is proprietary
Customer call transcripts, NPS verbatim, and support tickets may contain competitively sensitive information about a client's products, customers, and operations. Consumer-tier AI accounts may train on user data. API-tier accounts do not.
Mitigation: Always use API-tier Anthropic and OpenAI for processing client CX data. Include contractual commitments to clients specifying data handling. Never paste client data into consumer AI interfaces.
Build vs buy: the real math
10–14 weeks
Custom build time
$25,000–$45,000
One-time investment
5–8 engagements (at $5K+ saved analysis time per engagement)
Breakeven vs buying
A CX consultancy running 10 journey-mapping engagements per year, each currently requiring 3 weeks of analyst time at $150/hr for 120 hours = $18K in labor per engagement. Automating the discovery phase from 3 weeks to 3 days saves ~80 hours per engagement = $12K in labor cost saved. At 10 engagements/year, the custom build saves $120K/year in analyst time. Against a $35K build cost, payback is achieved in 3–4 months of operations. The ROI case is compelling — if the consultancy commits to using the platform for their full engagement volume.
Skip the DIY — RapidDev builds the production version
A Lovable MVP gets you a demo. Production needs auth that doesn't leak data, AI calls that don't bankrupt you, observability when models drift, and code you can audit. That's what we ship.
Discovery call (free)
30 minWe map your exact Customer Journey Mapping Tool use case: who uses it, target volume, AI model choice, integrations, compliance scope. You get a detailed scope document and fixed-price quote within 48 hours.
AI-accelerated build
10–14 weeksOur engineers use Claude Code, Lovable, and custom tooling to ship 3–5x faster than agencies. You see weekly progress in a staging environment — not a black box.
Launch + handoff
1 weekWe deploy to your infrastructure, transfer the GitHub repo, set up CI/CD and monitoring, and train your team. You own 100% of the source code, prompts, and model configurations.
What you get
Timeline
10–14 weeks
Investment
$25,000–$45,000
vs SaaS
ROI in 5–8 engagements (at $5K+ saved analysis time per engagement)
30-min call. Fixed-price quote within 48 hours. No commitment.
Frequently asked questions
How much does it cost to build a white-label AI journey mapping tool?
A Lovable MVP costs $25 tools + $40 API credits + $0.50 embedding batch = ~$65 to build. AI cost per 5,000-segment engagement is ~$6.50. RapidDev builds production-grade platforms with real HDBSCAN clustering, call-transcript connectors, and PDF export in the $25K–$45K range.
Is there a white-label journey mapping SaaS I can resell?
No honest one exists. Smaply and UXPressia are direct subscription products with no reseller programs. Custellence and Quadient are similar. The only path to a branded AI journey mapping product is a custom build — which is also why no one else has one yet.
What's the difference between journey mapping and journey orchestration?
Journey mapping is an insight and visualization product — it analyzes historical CX data (transcripts, tickets, NPS) to reveal how customers actually experience your product, then visualizes this as personas and journey stages. Journey orchestration is an action product — it triggers real-time next-best-action (email, SMS, push, chat) based on where each customer is in their journey right now. Both require different architecture: mapping is batch analytics; orchestration is real-time streaming.
How do I handle the HDBSCAN clustering quality problem?
Expect to spend 1–2 weeks tuning HDBSCAN parameters on your first real CX dataset before clusters are client-presentable. Start with min_cluster_size=15, min_samples=5, epsilon=0.5 for typical NPS verbatim datasets. Reduce min_cluster_size if your dataset is small (<2,000 segments) or you want more granular personas. Increase epsilon if too many segments are classified as noise. Build a parameter-tuning UI so consultants can adjust without engineering help.
Can I use this platform for call-center transcript analysis in California?
Yes, but confirm that the call recordings were collected with explicit all-party consent under California's two-party consent law (CIPA, Penal Code §632). 'Your call may be recorded for quality purposes' may not be sufficient — California courts have been inconsistent on whether this constitutes proper consent for AI analysis. Have the client's legal team confirm consent adequacy before ingesting California call recordings.
Can RapidDev build a white-label journey mapping platform for my CX consultancy?
Yes — RapidDev has shipped 600+ applications and can build a branded journey mapping platform with HDBSCAN clustering, call-transcript connectors (Gong, Zoom), Sonnet-generated persona narratives, and PDF export. Standard builds run $25K–$45K and ship in 10–14 weeks. Book a free 30-minute consultation to scope your specific data sources and client-deliverable format requirements.
Want the production version?
- Delivered in 10–14 weeks
- You own 100% of the code
- AI cost monitoring built in
30-min call. No commitment.
