Research model
Three-phase interview systemThe source separates problem discovery, silent usability testing, and outcome-focused customer-success interviews rather than mixing every question into one call.
Case-backedGrow & Operate a Small BusinessMixedHigh
Fastlane Case Study: Customer Research for an AI Short-Form Marketing Product
A source-backed Fastlane case showing how a small SaaS team used discovery, usability, and success interviews to focus an AI short-form marketing product, then used consented evidence, Claude, and MCP-connected data for roadmap decisions.
How Fastlane turned customer conversations into product decisions
Fastlane’s case suggests a disciplined loop: ask about past behavior before pitching, separate discovery from usability and success calls, recruit users where they already interact with the product, capture consented evidence, combine qualitative themes with usage signals, and prioritize the customers receiving the clearest value.
Save this nine-stage loop for interview design, ethical recruitment, consented capture, research coding, AI-assisted synthesis, usage segmentation, roadmap decisions, and stop rules.
- 01. The source describes three call phases: discovery, usability testing, and customer success.
- 02. Fastlane offered seven extra days of access through an in-product call-booking prompt and used Calendly for scheduling.
- 03. An unnamed AI note-taker captured calls, notes were stored in Notion, and Claude was used for synthesis.
- 04. A non-technical co-founder reportedly used Claude Code and MCP-connected data to build an internal customer-intelligence dashboard.
- 05. The claimed 2,000 calls, $69K software MRR, $1M whole-business ARR, and 1,000 paid users are unaudited Founder-reported figures.
Results and operating figures on this page are attributed to the cited source case. AI LinkBase has not independently audited them, and they are not a forecast for another product.
Discovery gate
Ask about the last real attempt, cost, workaround, and consequence of doing nothing
Usability gate
Watch the participant complete a task without rescuing them
Success gate
Find who receives repeat value and what changed for them
AI gate
Every summary must link back to consented source evidence
Roadmap gate
Prioritize recurring value signals, not the loudest request or an opaque score
Prices reviewed August 31, 2026 · Full Stack reviewed August 31, 2026
Decision profile
Architecture Snapshot
These labels describe how the complete workflow operates—not the license or deployment mode of every individual component.
Recruitment
Network + social + product promptThe team used personal contacts, X direct messages, Reddit and social lead magnets, waitlist email, and an in-product booking button.
Incentive
Seven extra days of accessUsers were offered a bounded product-access extension for booking; the source does not disclose redemption, bias, or abuse controls.
Capture
AI note-taker + NotionThe note-taking vendor was not named, so privacy, consent, retention, and accuracy must be evaluated before selection.
AI role
Synthesis + dashboard implementationClaude was used to analyze the call repository, while Claude Code and MCP-connected data reportedly helped a non-technical founder build the internal dashboard.
Decision model
Segment by motive, usage, and realized valueThe source describes signup reason, business type, subscription tenure, product usage, customer outcomes, and an internal customer-love score.
Historical delivery stack
Convex + Vercel + Railway + Clerk + Resend + Axiom + AI APIsThese are founder-named product tools, not a universal recommendation for the customer-research loop.
Human control
RequiredAI can organize and retrieve evidence, but participant handling, score design, causal interpretation, and roadmap choices require named human owners.
Setup difficulty
HighThe process combines research operations, consent, qualitative coding, analytics identity, access control, and an internal application.
Source case
Case Snapshot
Fastlane is an AI short-form marketing platform for products and small businesses. Co-founder Gaurav described an earlier horizontal product, Cassius AI, that attempted SEO, LLM optimization, Reddit engagement, and short-form content at once. Repeated customer conversations led the team to focus on short-form content, simplify the interface into a swipe-based review flow, and build an internal customer-intelligence system that connected interview evidence with customer and usage data. This page reconstructs that operating loop without treating the reported revenue or call volume as proof of causality.
- Project
- Fastlane
- Interviewed founder
- Gaurav
- Named co-founder
- Jock
- Predecessor
- Cassius AI, a broad marketing tool
- Public launch
- March 23, 2026
- Interview publication
- August 30, 2026
- Metric period
- Not disclosed; do not infer from publication date
- Timing classification
- Current operating case for a roughly five-month-old public product
- Core customer
- Solopreneurs marketing mobile apps and SaaS products
- Research cadence
- Around 20 calls per week, Founder-reported
- Research phases
- Discovery, usability, and customer success
- Decision layer
- Customer evidence + usage segmentation + human roadmap judgment
Evidence
Reported Results
Every figure keeps its evidence label so the historical case cannot be confused with an independently verified benchmark.
| Metric | Reported value | Context | Evidence |
|---|---|---|---|
| Software monthly recurring revenue | Around $69,000 | Founder-described software product level; homepage card says $80K/mo and the exact metric date is not disclosed | Founder-reported |
| Whole-business annual recurring revenue | More than $1 million | Founder says this covers the entire business; composition, date, and audit trail are not disclosed | Founder-reported |
| Paid users | More than 1,000 | Founder-described level; plan mix, churn, trials, and date are not disclosed | Founder-reported |
| Paid-beta annual recurring revenue | Around $16,000 | Founder-described starting point before the public launch; cohort and date are not disclosed | Founder-reported |
| Customer calls | Around 2,000 | Founder-reported total; does not transparently reconcile with the stated approximately 20-call weekly cadence | Founder-reported |
| Call cadence | Around 20 per week | Founder-described cadence for Gaurav and Jock since customer discovery began | Founder-reported |
| Vertical-beta waitlist | Around 2,000 people | Founder-described list size; not a count of activated or paying users | Founder-reported |
Cost control
Cost Snapshot
Fixed software is separated from variable usage, advertising, store fees, refunds, taxes, and labor. The largest cost is not necessarily a SaaS subscription.
Manual discovery pilot
Estimated$0 software floor + participant and founder time
- ✓Use one Calendly Free event type or a direct calendar link for interviews.
- ✓Store consented notes in a one-owner Notion Free workspace or a local spreadsheet.
- ✓Code five to ten interviews manually before adding AI synthesis.
- ✓Use a fixed interview guide and evidence tags so later automation has clean inputs.
Not included
Video conferencing · Participant incentives · Transcription · Research labor · Legal/privacy review · Data storage and backup
Two-founder repeatable loop
EstimatedAbout $60/month at listed annual-equivalent plan prices + transcription
- ✓Two Calendly Standard seats model to $20/month at the listed $10/seat/month annual rate.
- ✓Two Notion Plus seats model to $20/month using the current $10/member/month public price view.
- ✓One Claude Pro seat is $20/month and includes Claude Code.
- ✓Select an AI note-taker only after consent, retention, export, and no-training terms are verified.
Not included
Tax · Monthly-billing premiums · Additional Claude seats · Meeting platform · Participant incentives · Custom dashboard · Product analytics · Founder time
Case-inspired power-user layer
EstimatedAbout $240/month + transcription + custom-app infrastructure
- ✓Replace the modeled Claude Pro seat with the founder-named Claude Max 20x plan at the current $200/month price.
- ✓Keep two Calendly Standard and two Notion Plus seats as the scheduling and repository baseline.
- ✓Start MCP connections read-only and expose only the tables and fields needed for an approved research question.
- ✓Add a custom dashboard only after the manual coding scheme and identity joins have been validated.
Not included
Fastlane's historical product stack · AI note-taker plan · Convex, Vercel, Railway, Clerk, Resend, Axiom, and model API usage · Security review · Engineering labor · Participant compensation
Historical Fastlane product stack
EstimatedVariable and not reconstructable from public evidence
- ✓The founder named Convex, Vercel, Railway, Clerk, Resend, Axiom, OpenAI, Claude, and a garbled image/video vendor.
- ✓No plan names, invoices, usage volumes, storage, generation seconds, email volume, or observability volume were disclosed.
- ✓Current vendor pricing must not be projected backward as Fastlane's historical operating cost.
Not included
Labor · Model training or inference volume · Support · Payments · Marketing · Agency operations · Profit and margin
Operating system
How the Workflow Works
Keep the evidence gates in order: define the risky assumptions, recruit relevant people without buying compliments, ask about real past behavior, observe the product silently, capture only consented data, code the raw evidence before summarizing it, connect themes to usage and outcomes, and let a named human make the roadmap decision.
01
Define the decision and risk
Name the product decision the research must inform and the assumption that could make the build fail.
Components
Decision owner + hypothesis + target segment + stop condition
Output
One bounded research question and decision deadline
↓
02
Run problem discovery
Learn how people handled the problem before seeing the proposed solution.
Components
The Mom Test-style guide + personal network + targeted outreach
Output
Evidence about last behavior, workaround, cost, urgency, and consequence
↓
03
Recruit relevant testers
Invite people already showing the target behavior instead of optimizing for generic volume.
Components
X direct messages + Reddit/social lead magnets + waitlist email + in-product prompt
Output
A consent-ready interview queue with source and segment tags
↓
04
Offer a bounded incentive
Reduce scheduling friction without buying positive feedback.
Components
Seven-day access extension as source inspiration + eligibility and abuse rules + Calendly
Output
Booked sessions with incentive and consent recorded separately
↓
05
Observe usability silently
See where the product fails without teaching participants the intended path.
Components
Task script + screen share + facilitator silence + event markers
Output
Timestamped friction, errors, workarounds, and successful paths
↓
06
Capture and normalize evidence
Create a searchable research repository without losing provenance.
Components
Approved AI note-taker or manual notes + Notion schema + participant IDs + redaction
Output
Consented transcripts or notes linked to segment, task, date, and evidence tags
↓
07
Synthesize with traceable AI
Use Claude to cluster repeated signals while keeping every conclusion auditable.
Components
Claude + fixed taxonomy + source citations + human review
Output
Themes, counterexamples, confidence, and unanswered questions linked to calls
↓
08
Connect usage and outcomes
Distinguish curiosity and feature requests from repeat product value.
Components
Read-only product analytics + subscription status + business type + signup motive + outcome evidence
Output
Segments showing activation, retention, value, and contradictory cases
↓
09
Decide, test, and close the loop
Make one reversible roadmap decision, tell the team why, and measure the next cohort.
Components
Human decision owner + customer-intelligence dashboard + release note + follow-up interviews
Output
Build, reject, narrow, or investigate decision with evidence and recheck date
Tool-by-tool guide
What Every Tool Does in This Workflow
Each component has one job. Original-case tools are kept separate from optional affiliate and expert-execution routes so the historical record stays accurate.
The Mom Test
Design discoveryA question framework that emphasizes real past behavior and concrete costs over compliments or hypothetical intent.
- ✓Avoid pitching during discovery
- ✓Ask about the last real attempt
- ✓Separate opinions from evidence
Founder-named framework; the source summarizes several example questions.
Review The Mom TestCalendly
Schedule interviewsCalendly exposes available times and converts a product or outreach prompt into a booked call.
- ✓Offer interview slots
- ✓Record booking metadata
- ✓Send reminders and rescheduling links
Founder-named historical scheduler; current prices are reference-only.
Review Calendly pricingAI note-taker
Capture callsAn approved meeting assistant can record or transcribe a consented call and produce a first-pass summary.
- ✓Capture only with informed consent
- ✓Export the raw record and timestamps
- ✓Apply retention and deletion rules
The founder did not name the provider. No Current Primary is selected.
Notion
Store research evidenceNotion provides pages and databases for call notes, participant metadata, research tags, and links back to source evidence.
- ✓Maintain the interview repository
- ✓Normalize tags and properties
- ✓Restrict access to sensitive notes
Founder-named historical repository; AI LinkBase has a current Notion tool page.
Review NotionClaude
Synthesize evidenceClaude can classify and summarize a bounded set of redacted notes, surface repeated themes, and list counterexamples for a human reviewer.
- ✓Cluster evidence
- ✓Preserve source references
- ✓State uncertainty and contradictions
Founder-named for the customer-intelligence system; current plan choice depends on workload.
Review ClaudeClaude Code + MCP
Build the internal decision layerClaude Code can help implement a dashboard and query approved tools or databases through MCP connections.
- ✓Build reviewed dashboard code
- ✓Use least-privilege connectors
- ✓Keep changes and queries auditable
Founder says a non-technical co-founder built the dashboard with Claude Code and MCP-connected data; exact servers and permissions were not disclosed.
Review Claude Code MCP docsCustomer-intelligence dashboard
Support product decisionsA custom view joins interview themes with customer, subscription, usage, and outcome fields so a human can inspect segments.
- ✓Show evidence by segment
- ✓Expose source links and counterexamples
- ✓Record decision rationale and review dates
Custom internal Fastlane system; the scoring algorithm and schema are not public.
Product analytics
Measure behaviorAn approved analytics layer supplies activation, feature use, retention, and outcome signals without pretending that events explain customer intent on their own.
- ✓Track named events
- ✓Join only approved identifiers
- ✓Reconcile qualitative and behavioral evidence
The transcript does not clearly name the provider. Do not infer PostHog from the garbled wording.
Portfolio
Original Tools vs Current Portfolio
Original Tools preserve what the case reported. Current Portfolio reflects present workflow fit. A Required slot can remain empty when evidence is not strong enough to name a Current Primary.
S1FrameRequiredSelected
Customer-conversation method
Original case
The Mom Test principles
Current Primary
The framework matches the source and helps keep discovery focused on past behavior rather than compliments.
S2RecruitRequiredSelected
Interview acquisition
Original case
Personal networkX direct messagesReddit threadsSocial lead magnetsWaitlist emailIn-product call button
Current Primary
Selection still under review
The reusable unit is a tagged source mix, not a single acquisition tool.
S3IncentivizeOptionalUnder review
Participant exchange
Original case
Seven days of extra product access
Current Primary
Selection still under review
A bounded access extension can increase response, but the team must measure selection bias and avoid conditioning access on praise.
Important condition: Show the incentive before consent, keep it independent of sentiment, and define eligibility and abuse rules.
S4ScheduleRecommendedSelected
Call booking
Original case
Calendly
Current Primary
Calendly is founder-named and its current Free plan is sufficient for a one-event pilot; Standard adds repeatable reminders and event types.
S5CaptureOptionalPending benchmark
Meeting transcription
Original case
Unnamed AI note-taker
Current Primary
No Current Primary Yet
The historical role is clear but the vendor is not. Selection depends on consent, accuracy, export, retention, residency, and model-training terms.
Important condition: Default to manual notes when a participant declines recording or the data boundary is unclear.
S6OrganizeRequiredSelected
Research repository
Original case
Notion
Current Primary
A structured database can preserve participant, segment, phase, evidence, and decision links; access controls and exports still need ownership.
S7SynthesizeRecommendedSelected
AI analysis
Original case
Claude
Current Primary
Claude can assist with bounded synthesis when notes are minimized, sources remain linked, and a human reviews every conclusion.
S8OperationalizeOptionalUnder review
Internal customer-intelligence app
Original case
Claude CodeMCP-connected dataCustom Fastlane dashboard
Current Primary
Build the custom layer only after the manual taxonomy and joins are stable; begin with read-only, field-limited access.
Important condition: Require connector trust review, prompt-injection controls, audit logs, and a rollback owner.
S9MeasureRequiredPending benchmark
Product and subscription evidence
Original case
Provider not clearly namedSubscription and customer fields in the internal dashboard
Current Primary
No Current Primary Yet
The workflow needs behavioral and outcome evidence, but the source does not identify enough implementation detail to recommend a provider.
S10Deliver productOptionalPending benchmark
Historical SaaS stack
Original case
ConvexVercelRailwayClerkResendAxiomOpenAIClaudeUnclear image/video vendor
Current Primary
No Current Primary Yet
This is Fastlane's founder-reported product stack, not the minimum customer-research stack and not a current blanket recommendation.
Replacement choices
Alternatives & Trade-offs
These are comparison directions, not automatic recommendations. Replacing one component can change integrations, operating burden, attribution quality, or the economics of the whole workflow.
| Slot | Lower-cost / direct path | Specialized path | Replacement trade-off |
|---|---|---|---|
| Scheduling | Use Calendly Free or a direct calendar invitation | Use Cal.com when self-hosting, routing, or an existing scheduling workflow justifies it | More scheduling flexibility adds configuration, identity, and event-sync work. |
| Call capture | Take structured manual notes and mark timestamps | Use a research repository or meeting-intelligence tool with verified consent and retention controls | Automation saves review time but adds privacy, transcription-error, and vendor-lock-in risk. |
| Research repository | Use a spreadsheet with stable participant and evidence IDs | Use Notion or a dedicated qualitative-research platform | Flexible tools are faster to start; specialized systems improve coding and governance but require process discipline. |
| Synthesis | Code themes manually and compare reviewers | Use Claude on a redacted, source-linked evidence set | AI accelerates clustering but can flatten minority evidence or invent certainty. |
| Decision dashboard | Use filtered Notion database views | Build a read-only internal app after the schema stabilizes | Custom dashboards improve joins and speed but create engineering, access-control, and score-governance obligations. |
| Product analytics | Use a minimal first-party event table | Use a dedicated analytics product with cohort and identity controls | More instrumentation creates more decision context and more opportunities for bad joins, surveillance, and vanity metrics. |
Transferability
What to Copy—and What Not to Assume
What is transferable
- ✓Separate problem discovery, usability, and customer-success calls because each phase asks a different question.
- ✓Ask about the last real behavior, workaround, cost, and consequence before describing the solution.
- ✓Recruit users at moments of demonstrated intent, then tag the recruitment source to expose selection bias.
- ✓Keep incentives bounded and independent of whether feedback is positive.
- ✓During usability sessions, watch first and rescue only after the observation is complete.
- ✓Preserve raw notes, timestamps, counterexamples, and participant context before asking AI for themes.
- ✓Connect qualitative evidence to activation, retention, and outcomes without treating correlation as causation.
- ✓Use read-only, least-privilege MCP connections before granting any write capability.
- ✓Make one reversible roadmap decision per evidence cycle and record the recheck date.
What may not transfer
- !Fastlane's claimed 2,000-call volume is not a universal requirement and is not independently verified.
- !The reported $69K software MRR, $1M whole-business ARR, and launch virality are not forecasts for another team.
- !A seven-day access extension may attract low-intent users or violate another product's economics or terms.
- !Fastlane's swipe-based interface follows its specific short-form content workflow and should not be copied without usability evidence.
- !The internal customer-love score is undocumented and cannot be transferred as a validated formula.
- !The founder-named product delivery stack is not necessary for a small research pilot.
- !Customer conversations can inform channel and product choices, but the source does not prove they caused the reported growth.
Guardrails
Risks and Stop Conditions
Customer calls can contain personal data, confidential business information, access tokens, health or financial details, and unverified opinions. AI synthesis and MCP-connected dashboards add leakage, hallucination, prompt-injection, and over-automation risk. The workflow is publishable only with participant notice, data minimization, scoped access, human review, and traceable links from conclusions back to raw evidence.
Critical
Recording or transcribing calls without informed consent can create privacy, contractual, and trust failures.
Mitigation: Provide clear notice, capture affirmative consent, support no-recording participation, minimize fields, and publish retention and deletion ownership.
Critical
MCP-connected AI can expose customer data or act on malicious instructions embedded in external content.
Mitigation: Trust-review every server, start read-only, restrict tables and fields, redact secrets, require approval for actions, and retain query and access logs.
High
AI summaries can hallucinate consensus, erase minority evidence, or detach conclusions from source calls.
Mitigation: Require source links, counterexamples, confidence labels, reviewer sampling, and direct evidence checks before decisions.
High
An access incentive can bias the sample toward people seeking free usage rather than durable value.
Mitigation: Tag incentivized participants, compare them with organic and paying cohorts, and never condition the reward on positive feedback.
High
A customer-love score can hide subjective weights and turn a convenience metric into an authority signal.
Mitigation: Publish the inputs internally, test sensitivity, show raw component values, and keep a named human accountable for the decision.
High
Joining interview, identity, subscription, and product data can create an excessive customer-surveillance profile.
Mitigation: Use purpose limitation, pseudonymous research IDs, minimal joins, role-based access, expiration, and a documented lawful basis.
High
Power-user interviews can overfit the roadmap and exclude new, struggling, or churned customers.
Mitigation: Sample across activation states, include churned and failed users, report segment sizes, and preserve contradictory evidence.
High
High interview volume can consume the team's build and support capacity without improving decisions.
Mitigation: Set saturation and decision thresholds, stop collecting redundant evidence, and measure which research questions changed an action.
Stop when: Stop recording, connecting, or automating when participants have not given informed consent, sensitive fields cannot be minimized, an MCP server requires write access without a justified task, AI summaries cannot be traced back to call evidence, customer segments are too small or biased to support a decision, or the team is using a score to replace human product judgment.
Frequently Asked Questions
What is the core Fastlane workflow?+
It is a three-phase research loop: discover real problems, observe usability without rescuing the user, then interview successful or high-usage customers to understand realized value. Notes feed a structured repository and are combined with behavioral evidence for human roadmap decisions.
Did customer calls cause Fastlane to reach $69K per month?+
The source presents calls as important to product and marketing choices, but it does not isolate causality. Launch attention, distribution, timing, product quality, pricing, and other factors may also explain the reported result.
Did Fastlane really complete 2,000 customer calls?+
That is the founder's claim. It is not independently verified, and the public source does not transparently reconcile the total with its stated approximately 20-call weekly cadence. Treat it as Founder-reported, not a benchmark.
When were the $69K MRR and 1,000 paid-user figures measured?+
The exact recording and dashboard date is not disclosed. The article was published August 30, but its 'just over two months' wording does not align with the March 23 launch date, so the publication date must not be used as the metric date.
What changed because of the interviews?+
The founder describes narrowing a broad marketing product to short-form content, identifying usability friction, adopting a swipe-based review interface, segmenting customers by signup motive and usage, and prioritizing customers reporting the clearest outcomes.
How many calls should a small team run?+
The case does not establish a universal number. Start with a bounded segment and question, stop when new calls are repeating existing evidence or the decision can be tested, and reopen research when the next cohort contradicts the result.
Do I need an AI note-taker?+
No. Structured manual notes are safer when consent or vendor terms are unclear. Add an AI note-taker only when recording notice, export, retention, deletion, residency, model-training, and accuracy requirements are acceptable.
Should Claude be allowed to write to customer systems through MCP?+
Not by default. Begin with read-only, field-limited access for a named research question. Add write actions only after separate authorization, approval gates, audit logging, rollback, and prompt-injection controls are in place.
Was PostHog part of the original stack?+
The transcript wording is unclear and may be ordinary language or a transcription error. This candidate does not infer PostHog or any other analytics vendor without a human audio check.
Was fal.ai the image and video provider?+
The public transcript renders the vendor as 'Fowl AI.' That is not enough evidence to normalize the name. A human must check the original audio before any vendor is named as historical fact.
Can I copy Fastlane's full technical stack?+
You can study the roles, but the case does not prove that the same vendors or architecture are right for another product. Start with the research loop, then choose current tools against data sensitivity, usage, team skill, and cost.
Is Fastlane still operating?+
Yes as of this review: its official site is live, publishes current product and pricing information, and has dated first-party content in August 2026. That verifies current operation, not the unaudited revenue claims.
Transparency
Sources & Methodology
AI LinkBase separates Starter Story publication metadata, direct founder statements, launch-platform corroboration, current Fastlane first-party product and policy claims, official vendor documentation, reconstructed workflow steps, and current recommendations. Financial results, paid-user counts, call volume, waitlist size, and historical tool use remain Founder-reported. The $69K interview figure conflicts with the $80K homepage card, the exact metric date is undisclosed, and the 2,000-call total does not transparently reconcile with the weekly cadence. Garbled vendor names are left unresolved. No historical cost, profit, margin, causal effect, or analytics vendor is inferred without evidence.
Starter Story founder interview ↗Starter Story homepage discovery capture source ↗Fastlane official website and pricing ↗Fastlane official privacy policy ↗Fastlane official terms ↗Fastlane dated operating evidence ↗Product Hunt Fastlane launch record ↗Gaurav Thapa launch announcement ↗The Mom Test official site ↗Calendly official pricing ↗Calendly official plan guide ↗Notion official pricing ↗Notion official plan guide ↗Claude official pricing ↗Claude Max plan pricing ↗Claude Code MCP documentation ↗Convex official documentation ↗Vercel platform documentation ↗Railway services documentation ↗Clerk official documentation ↗Resend email API documentation ↗Axiom observability documentation ↗
Source case: The Insanely Obvious Secret Behind This $69K/Month SaaS, Starter Story, August 30, 2026. Short excerpts are used only where necessary; this page is an original workflow analysis.