The SaaS Distribution Course #15: How Within Reached Nearly $20M ARR by Following One Conference Signal

TL;DR

Within, formerly Klarity, reportedly took a new enterprise product from zero to nearly $20 million in ARR in about 18 months. That result is founder-reported and unaudited. The useful fact is how the team found the new motion: at a roughly 500-person conference it already owned, customers ignored a year of core-product releases and kept asking to buy one small internal tool.

The team did not treat applause as validation. It offered a simple end-of-October deal, signed 12 to 14 customers in about five to five-and-a-half weeks, and reached roughly $350,000 ARR. Then it tested several possible markets, weighted feedback by actual customer behavior, and concentrated capital on events rather than Google ads. The predecessor finance product, itself around $10 million according to co-founder Andrew Antos, was later sold to an implementation partner so the company could focus on the faster motion.

The warning is important. No public source reveals event CAC, payback, sourced pipeline, retention, or expansion revenue. Do not copy “run a conference.” Copy the evidence chain: weighted conversations → bounded market bet → qualified comparison surface → paid proof → competing hypotheses → customer-evidence loop → channel concentration.

What you will build

You will build a signal-to-concentration distribution system for an early B2B SaaS. Its outputs are a weighted conversation ledger, one six-month market gate, a qualified comparison session, a paid-intent offer, a quarterly experiment portfolio, and a channel allocation rule. Every stage has a pass condition and a stop condition, so a weak market cannot hide behind busy sales activity.

Use this system if

• you can put 20 to 50 relevant buyers in the same room, call, community, or workshop;
• you have an adjacent tool, workflow, or offer to compare with the current product;
• the founder can still hear sales, usage, and support evidence directly;
• one customer can justify enough revenue to support manual discovery and onboarding;
• you are willing to stop a respectable motion when a much stronger one appears.

Do not use it if

• the “event” is a broad audience built for impressions rather than qualified decisions;
• you will count compliments, scans, or sign-ups as willingness to pay;
• every experiment lacks an owner, deadline, and explicit failure threshold;
• you cannot support existing customers while testing the adjacent motion;
• channel concentration would expose the company to one platform it cannot control.

Verified case snapshot

Stage	Evidence	Limit
2017–2020	AI contract review reached about $800K ARR after roughly two-and-a-half years	Founder-reported; a viable but slow motion
Finance bet	Goal: $250K ARR from at least eight customers in six months; result: about $340K and eight customers	Founder-reported
January 2022	TechCrunch reported 40+ customers and 9x ARR growth over nine months	Independent coverage of company-supplied metrics
Approx. 2024 conference	A small internal tool drew the strongest demand among roughly 500 customers and prospects	Attendance and reaction are founder-reported
First 5–5.5 weeks	12–14 deals and about $350K ARR after a deadline-based early offer	Founder-reported; pricing varied by deal
Next quarter	More than $1M ARR	Founder-reported and unaudited
About 18 months	New product close to $20M ARR; predecessor finance product sold after reaching about $10M	No independent verification found
August 2026	Klarity rebranded as Within around its company-brain positioning	Official company announcement

The model: a signal only matters when the next stage makes it costlier

Within's story can be misread as a dramatic pivot. The distribution system is more disciplined. Each stage asks the buyer for a stronger behavior than the stage before it.

1. A conversation reveals language, pain, and market resistance.
2. A bounded market bet asks several customers to pay within a fixed period.
3. A qualified event compares the current roadmap with an adjacent product in real time.
4. A deadline offer turns “interesting” into a commercial decision.
5. Usage and deal progression decide which market hypothesis deserves resources.
6. Concentrated channel spend creates more qualified conversations and restarts the loop.

This is causal because the output of one stage becomes the input of the next. The conference did not create a market from nothing. The finance customer base supplied the audience. The internal implementation work supplied the tool. Buyer questions supplied the initial message. Paid deals supplied the right to run broader market experiments.

Step 1: weight market feedback by behavior

Klarity began with AI-assisted contract review. Law firms made almost no progress because the product threatened billable work and the technology could not generalize across complex documents. In-house legal teams were better buyers, but the product still required too much implementation for a problem that was not urgent enough. After about two-and-a-half years, the business sat near $800,000 ARR.

The team stopped asking only how to improve the sales process. It asked which document-heavy function had a bigger recurring problem. Finance teams handled contracts, purchase orders, invoices, receipts, and order forms. The same technical capability could remove a larger amount of expensive review work.

Build a conversation ledger with these fields:

• buyer function and seniority;
• problem named before your pitch;
• current time, headcount, or compliance cost;
• next step reached and time to that step;
• product usage or paid status;
• reason the deal advanced, stalled, or died.

Give existing customers and fast-progressing opportunities more weight than friendly first meetings. Pass condition: 30 qualified conversations reveal one repeated problem, and at least five buyers have already spent money or headcount on it. Stop condition: the repeated language appears only after you introduce it.

Step 2: give the market bet a deadline and two thresholds

The finance hypothesis was not “finance seems promising.” The team allowed three months to build on its existing technology, then set a six-month commercial gate: at least $250,000 ARR from at least eight customers. Both dimensions mattered. One unusually large contract could not prove a repeatable market, while eight tiny pilots could not prove useful economics.

The result was roughly $340,000 ARR from eight enterprise customers in six months. That did not prove a final distribution system. It proved enough buyer density and contract value to keep learning. TechCrunch later reported that the finance product served more than 40 companies by January 2022, which independently corroborates the direction of that expansion.

Write your gate before building:

• Window: ___ weeks from usable product to decision.
• Minimum distinct customers: ___.
• Minimum recurring revenue: ___.
• Maximum founder delivery per customer: ___ hours.
• Minimum active usage after 30 days: ___.

Pass condition: clear both the customer-count and revenue thresholds without breaking the delivery cap. Stop condition: one heroic deal supplies most of the result, or paid customers do not use the product after onboarding.

Step 3: create a qualified comparison surface

The finance business grew partly through two large owned conferences each year. At one event, roughly 500 current and prospective customers watched the team launch major product work. One minor release was different: an internal tool that turned recordings of how people worked into implementation documentation in minutes.

The audience supplied a live comparison. Buyers could discuss a year of finance features or the small adjacent tool. They repeatedly asked how to buy the latter. That is more valuable than a post-event satisfaction score because the same qualified people saw both alternatives at once.

You do not need a 500-person conference. Run a 45-minute customer council, office-hours session, implementation workshop, or niche dinner. Show the current roadmap and one adjacent tool. Do not announce which one you prefer. Record the first unaided question after each demo.

Metric: qualified buyers who ask for access, pricing, or an implementation step without prompting. Pass condition: at least 20% of the qualified room asks for a concrete next step within 72 hours. Stop condition: interest exists only in a poll or feedback form.

Step 4: translate attention into paid proof

Within removed procurement and pricing complexity from the first decision. Buyers who committed by the end of October received unusually high volume for a low early price. The objective was not margin optimization. It was to discover whether the conference enthusiasm survived a buying decision.

Twelve to 14 deals closed in roughly five to five-and-a-half weeks. Their pricing was messy and had to be standardized later, but the signal was much faster than the finance motion: six weeks rather than six months to reach a similar ARR level.

Use a paid-intent offer with four lines:

1. the narrow outcome delivered in the next 30 days;
2. the fixed scope or usage allowance;
3. one price that avoids negotiation;
4. a real deadline tied to onboarding capacity.

Do not use a fake countdown or hide renewal economics. Pass condition: qualified buyers sign and begin implementation. Stop condition: discounts create contracts without usage, or each deal requires a new product.

Step 5: run four bets, not forty tactics

Early demand did not reveal the final market. The team tested compliance, bank risk, post-merger integration, ERP transformation, and AI transformation. It compared usage, deal size, progression, customer language, and renewal or expansion behavior. AI transformation became the strongest direction.

Within now uses 90-day cycles with four to five hypotheses. Each bet has an owner and a startup metric. Antos gives a 2026 example: the company's MCP bet needed 100 genuinely active users in 90 days, with “active” defined in advance. A small positive movement is not automatically success. His operating assumption is that nine of ten bets will fail and one should work much better.

Bet	Startup metric	Kill signal	Scale signal
New segment	Paid pilots and median ACV	Polite calls, no paid next step	Fast progression plus active use
Owned event	Qualified pipeline per attendee	Attendance without buying intent	Repeated access and pricing requests
New message	Qualified reply-to-meeting rate	Higher clicks, same opportunities	Buyer repeats the language internally
Adjacent product	Active customers inside 90 days	Demo excitement without retention	Usage and expansion across teams

Step 6: make customer evidence queryable

Within records customer conversations with consent, joins them to Salesforce progression, usage, Linear tickets, support email and calls, and stores the combined view in a Customer 360 layer. Teams query it for sentiment, repeated language, segment differences, and product friction. Feedback from an active customer or fast-moving deal receives more weight than a meeting that went nowhere.

The lesson is not to copy this stack on day one. Start with one row per account and five evidence types: words, behavior, commercial progress, support burden, and retention. Review it weekly. Promote a message only when the buyer's language and behavior agree.

Pass condition: every major positioning claim links to at least three accounts whose behavior supports it. Stop condition: the team quotes enthusiastic calls while ignoring stalled deals, low use, or support debt.

Step 7: concentrate resources and protect the transition

Within says it runs no Google ads and puts a large share of marketing budget into owned and third-party events because that channel works for its enterprise motion. Current company pages still show this behavior: Within promotes executive events and customer-led sessions such as its ServiceNow webinar and the Inflection event.

Concentration did not mean abandoning customers. The new product initially ran beside the finance business. When it overtook the predecessor in about a year, an implementation partner that already knew the product acquired the operation and took the supporting team. Transaction terms are not public, but the sequence shows the necessary constraint: protect customer service while moving resources.

Scale condition: one channel produces at least twice the qualified-pipeline yield of alternatives across two cycles. Diversification condition: add a second channel when the winner depends on a platform, partner, or audience you cannot control. Stop condition: concentration raises pipeline while delivery, retention, or customer trust falls.

What failed, and why

• Law firms: structural conflict and weak urgency. Better sales execution could not make the initial buyer care enough.
• In-house legal: some demand, but a small problem and difficult implementation capped velocity.
• Professional-services-heavy finance delivery: a real business that did not fit the founders' operating strengths.
• Most market experiments: deliberately disposable. Small incremental movement did not earn permanent resources.
• Default paid-search playbook: rejected because events produced a stronger enterprise distribution surface for this company.

These are different failure types. Legal was a market problem. Finance delivery created a founder-business fit constraint. Individual hypotheses were expected learning costs. Paid search was a capital-allocation choice. Diagnose the type before changing the tactic.

Your 30-day implementation plan

1. Days 1–5: export the last 30 customer and prospect conversations. Add progression, usage, support, and loss reason.
2. Days 6–8: rank three repeated problems by urgency, buyer seniority, existing spend, and delivery burden.
3. Days 9–11: write one six-month market gate with customer-count, revenue, usage, and delivery thresholds.
4. Days 12–16: build one adjacent proof that solves the most expensive repeated problem. Keep it small enough to demo live.
5. Days 17–19: invite 20–30 qualified accounts to a customer council or workshop. Separate customers, active pipeline, and cold prospects in the analysis.
6. Day 20: demo the current roadmap and adjacent proof. Record unaided questions and concrete next-step requests.
7. Days 21–23: send one transparent paid-intent offer with fixed scope, price, capacity, and deadline.
8. Days 24–27: define four 90-day bets. Give each one an owner, startup metric, kill threshold, and scale threshold.
9. Days 28–30: allocate the next cycle's time and budget to the strongest paid and usage evidence, not the loudest audience response.

Lead Scorer implementation

Lead Scorer can reproduce the research, qualification, and draft-review parts of this motion. It cannot decide product-market fit, run an event, activate a campaign without approval, or replace product usage data. Use it when your comparison surface is a LinkedIn event, a post, a Sales Navigator search, or a narrow company list. Do not use this workflow when you cannot define the buyer or when the event audience is broad and unqualified.

Phase 1: separate evidence before enrichment

Invoke the icp-offer-context skill to record the product, narrow buyer, trigger, exclusions, and proof you may claim. Create separate lists for existing customers, active opportunities, event registrants, attendees, and cold lookalikes. If the signal lives on LinkedIn, use the signal-audiences skill and the MCP action create_audience_source, then sync_audience_source. Never mix customers and cold prospects into one score or outreach sequence.

Copyable prompt: Capture this event or post as an audience source. Keep registrants, attendees, and visible engagers separate. Deduplicate them against customers and active opportunities. Do not enrich or contact anyone yet.

Output: clean signal lists with origin and date. Pass condition: every person has a traceable source and belongs to only one outreach state. Stop condition: the source cannot distinguish qualified participation from passive reach.

Phase 2: score before spending contact credits

Use icp-scoring-rubric to define a weighted 1–10 rubric. Require fit, a current trigger, and one behavior connected to the event or problem. Run get_leads_pending_scoring and record decisions through submit_lead_score. Keep a suggested threshold of 8/10 for personal research, but calibrate it against five known-good and five known-bad accounts before applying it.

Copyable prompt: Score this audience against the event's narrow job. Require company fit, role authority, a dated trigger, and a relevant action. Return keep, hold, or reject with the evidence. Do not enrich rejected or held leads.

Only for keepers, use signal-research-dossier and then contact-discovery. The corresponding MCP actions can include enrich_company, enrich_leads, and find_lead_contact_info. Preserve the platform's credit confirmation gates. Pass condition: every keeper has two verified, dated signals or is honestly marked unresearchable. Stop condition: contact finding begins before fit and evidence are confirmed.

Phase 3: draft by evidence state, then require review

Create one draft campaign for the qualified cohort with create_campaign. Use ai-authored-campaign or cold-email-first-touch to draft a personal first touch, then generate_campaign_drafts or write_campaign_drafts as the execution action. The message should reference the person's actual event question or observed problem, offer one bounded next step, and avoid claiming that attendance proves intent.

Copyable prompt: Draft one message per qualified lead. Use only verified evidence from the event and dossier. Offer a 20-minute comparison session with one clear outcome. Leave every action in draft and require human review before activation.

Run outreach-qa-audit before review. A human must approve claims, recipients, senders, sequence, and timing. Lead Scorer may draft; it must not send or activate this campaign automatically. Pass condition: each message contains one verified signal and one ask. Stop condition: the copy treats a generic event interaction as personal intent.

Phase 4: turn objections into the next experiment

Use reply-triage to classify interested, timing, wrong person, objection, or no. Save repeated objections as research or content-source items rather than forcing a reply. The daily-topic-briefs skill can turn the strongest repeated problem into useful evidence for the next customer council. Capture the people who engage with that evidence as a new, separately dated signal audience. This recreates the outer loop: conversation evidence changes the next message, audience, and product experiment.

Saveable checklist

• Weight feedback by paid status, usage, progression, and support evidence.
• Write the market gate before building the next version.
• Require both customer-count and revenue thresholds.
• Compare the current product with one adjacent proof in a qualified room.
• Convert interest through a transparent, bounded paid offer.
• Run four or five hypotheses with explicit kill metrics.
• Promote buyer language only when behavior agrees.
• Concentrate on a measured channel winner, then protect delivery and platform risk.

Sources and limits

The main source is the complete 49-minute Product Market Fit Show interview with Andrew Antos, published August 31, 2026. The transcript was audited from character zero through character 56,135 in eight contiguous windows. Podscan's public page supplied episode ID ep_wabyjg8nwolyr3vd, but its authenticated MCP was unavailable in this run.

TechCrunch independently corroborates the earlier finance-market direction, funding, customer count, and reported growth, but not the 2026 revenue outcome. Within's official rebrand announcement confirms the current name and positioning. Current event pages establish that owned events remain part of the company's go-to-market surface, not that events caused all revenue.

The nearly-$20-million ARR result, $10-million predecessor size, $350,000 launch run rate, deal count, and timing remain founder-reported and unaudited. The sources disclose no CAC, event ROI, gross margin, churn, net retention, payback, or channel-sourced revenue. Treat 30 conversations, 100 customers, and nine failed bets out of ten as Antos's operating heuristics, not universal laws.