The First 100 Customers Course #41: How Ask-AI Won Its First 25 With Founder-Led Sales
A practical course on using enterprise interviews, a design partner, cross-buyer validation, and founder-led outreach to win early SaaS customers without mistaking one custom project for a market.
By Miljan @ Lead Scorer 17 min read The First 100 Customers Course · Lesson #41
TL;DR
Ask-AI got close to $1 million in annual recurring revenue with about 25 customers, according to founder Alon Talmor. He sold those customers himself through calls, emails, connections, and some help from an SDR service. Before that, he spent six months getting inside enterprises to hear a problem he could build for. The first useful prototype ran free with Yotpo for about a year. None of these figures is an audited cohort report; they come from Talmor's SaaS Club interview.
The lesson is a sequence: find a buyer with a costly, repeated problem; work beside one design partner; test the same problem with several other buyers; then make the founder own the sales conversation until the objection pattern is clear. The warning is equally important. A free design partner, an expression of willingness to pay, and a paying customer are three different events. Count them separately.

What you will build
By the end of this course, you will have a four-part early sales system: an interview list, a problem record, a design-partner agreement, and a validation queue outside that partner. You will also have a small founder-led outreach experiment with explicit pass and stop conditions. You can run the first pass in seven days. The goal is to decide whether to continue, narrow, or abandon a buyer/problem pair, not to manufacture a milestone from one encouraging meeting.
This edition is for a builder who can make a working prototype but has no reliable path to paid buyers. It is most useful when the product depends on access to a buyer's workflow or data. It is less useful when customers can evaluate a self-serve product independently in minutes. Ask-AI was selling into enterprises, and its founder had previous enterprise experience. Copy the testing logic, then adjust the access route and cycle length to your market.
Case snapshot: separate the two companies and stages
| Stage | What the source supports | What it does not prove |
|---|---|---|
| Discovery | Talmor says he talked to enterprises for six months before pitching investors. Yotpo's delivery leader described scattered knowledge across systems. | A fixed number of total interviews or a conversion rate. |
| Design partnership | A free Yotpo prototype was used internally for roughly a year. Talmor recalls a discussion of around $30,000 annual willingness to pay. | A signed $30,000 contract at the start of the pilot. |
| Early commercial motion | Talmor says he handled the first roughly 25 sales, using his own calls, emails and connections, plus an SDR service. | How many customers came from each route or what each paid. |
| Wider business | Ask-AI was close to $1 million ARR around that early cohort. A sales leader arrived toward the end of 2023. | Exactly $1 million, exactly 25 paying accounts, or a durable repeatable channel. |
| Later change | The company narrowed from a broad enterprise assistant toward technical support and became Mosaic AI. | That today's five-day pilot won the original 25. |
The investor's account independently names Yotpo and monday.com among the early customers it knew. It is a participant source, not an audit of the count. Mosaic's January 2026 rebrand note confirms the later change in positioning. The order matters: do not retrofit the current technical-support sales process onto the original Ask-AI acquisition story.
The model: discovery, proof, replication, payment
A design partner can tell you whether a workflow is painful and whether your prototype can touch the real data. It cannot tell you alone whether the market wants the same product. Talmor's useful move was to get close enough to see the workflow, then ask other enterprises whether the same failure was present. Only after that does founder-led selling have a proposition to test.
Think of the process as four gates. Discovery identifies an expensive repeated problem in the buyer's words. Proof checks that a narrow prototype changes the workflow. Replication checks that several buyers have the same problem without a bespoke implementation. Payment asks whether anyone commits budget on terms you can deliver again. Each gate produces a different artifact. Keep them separate in your CRM, even if one customer moves through all four.
For Ask-AI, the first named pain was fragmented company knowledge: support tickets, CRM, Confluence, Jira, Slack, and product data held answers that customer-facing teams could not assemble quickly. The founder's data background made this a plausible problem to attempt. That is a case-specific advantage. A builder without deep integration skills should pick a narrower slice instead of promising an enterprise-wide answer layer.
Step 1: enter with a problem hypothesis, not a demo
Talmor describes approaching roughly 10–20 enterprises to get two or three conversations during ideation. That is his recollection, not a measured campaign denominator. The practical point is that access was hard even with a strong background. He used friends of friends where possible and asked to understand the buyer's problem. One VP of delivery at Yotpo described knowledge scattered across systems and employees unable to answer customer questions. That conversation became a product hypothesis.
Write your hypothesis as a job and a failure: “The support lead needs to answer a complex account question, but the evidence is split across ticket history, CRM and product systems.” Name the person who owns the cost. Do not start with “companies need AI.” That phrase contains no buyer, budget, or workflow.
Build a list of 20 buyers in one segment. For each, record the trigger that makes the problem plausible, the system where it occurs, and one route to the person who owns it. Send a short request for a problem interview. In the call, ask for a recent example, the current workaround, time or money lost, who approves a change, and what they already tried. Ask permission to see an anonymized real case. A story about last week's incident is stronger than agreement with your pitch.
Output: ten documented invitations and at least two real conversations before you expand the list. Pass condition: two buyers independently describe the same costly workflow and can show a recent example. Stop condition: they praise your concept but cannot name a recent incident, owner, or consequence. If the first ten invitations produce no calls, revise access and buyer title before changing the product.
Step 2: use one design partner to expose the work
Yotpo tested an early prototype internally without paying for about a year, Talmor says. The VP discussed a possible annual value around $30,000. Treat that statement as a pricing signal, not a booked contract. The free work bought access to the workflow, system relationships and failure cases. It also created a dangerous temptation: one large partner can turn your roadmap into its private procurement list.
Define the partnership on one page. State the exact workflow, the user's current baseline, the data you may access, the prototype's narrow job, the review cadence, and the date at which you will ask for a paid decision. Write down who can approve the purchase. If you give free access, be explicit about why: to learn a costly workflow that no synthetic demo can show. Free access without a learning question is a subsidy.
Set a weekly evidence review. For each attempted case, record whether the prototype found the right information, where it failed, whether a human trusted the output, and what would have happened without it. Collect the raw examples before you build more connectors. The first useful metric may be “cases in which the user found a correct answer faster,” not an impressive aggregate AI score.
Output: a narrow pilot log and a dated commercial decision point. Pass condition: real users return with new cases and the buyer agrees on how success will be judged. Stop condition: every improvement is a one-off request that no second customer needs. Do not turn a verbal price discussion into revenue in your report.
Step 3: test the same pain outside the partner
Talmor says you must talk to five others quickly after a design partner. His reason is concrete: a very large enterprise may offer a large contract while demanding a custom subsidiary. The next customer may need none of the features you built. He preferred mid-market buyers when possible because access and cycles were simpler. That preference is a founder judgment, not a universal rule; some products require the largest enterprises.
Create a validation queue of five other organizations with the same buyer role. Present the problem statement, not the partner's feature list. Ask each buyer for a recent instance, its cost, what the partner prototype would miss, and whether their existing systems could support a similar solution. Code answers as “same core pain,” “adjacent pain,” or “different job.” A product wedge exists when the common job survives contact with different buyers.
Do a deliberately small cross-buyer test. For example, if three of five buyers independently show the same workflow and at least two will let you inspect a real case, continue. These are proposed decision thresholds for your experiment, not Mosaic's historical numbers. If only the design partner has the problem, narrow the product to that account or stop. If everyone has the problem but no one owns a budget, change buyer or economics before building more.
Output: a five-row comparison with current workaround, cost owner, systems, urgency and required customization. Pass condition: the same job appears without prompting in multiple accounts. Stop condition: each account asks for a distinct product. The point is to protect a repeatable product from an attractive bespoke project.
Step 4: let the founder own the first sales cycle
Talmor says the first roughly 25 customers came from his own calls, emails and connections, assisted by an SDR service. He ran the demos. A sales leader joined later, when the business was approaching seven-figure ARR. The source does not break out channel attribution, so you cannot say that cold email alone won 25 customers. You can say the founder stayed close to the objections and buying process.
Run one explicit prospecting experiment. Separate warm introductions from unfamiliar target buyers. Send each group the same problem hypothesis and a small ask: a short discussion of one recent incident. Track invitations, replies, qualified conversations, case access, pilot offers and paid commitments by cohort. Keep the copy restrained enough that a reply means interest in the problem rather than curiosity about a dramatic claim.
After every call, capture the exact objection and the buyer's language. Was the problem real but low priority? Was IT access the obstacle? Did the buyer want a different outcome? Which evidence changed their mind? The founder should update the offer and demo from these notes before delegating the learning loop. An SDR service can help open doors; it cannot decide whether you have a wedge.
Output: two small cohort funnels and an objection ledger. Pass condition: buyers outside your warm circle advance to a real case or commercial discussion. Stop condition: only friendly contacts move, or each sale requires a fresh product. The first sale is proof of willingness to try; the next independent sale is a better test of repeatability.
Step 5: sell an outcome you can measure
Mosaic AI's later technical-support motion illustrates an upgrade to the early model. After ChatGPT made the broad assistant position vulnerable, the company focused more deeply on technical support. Talmor describes a diagnostic using roughly 1,000 tickets, then a short pilot in which an internal assistant helps a support engineer resolve real cases. He prefers tickets handled per engineer per day to time to resolution, because engineers switch among cases and ticket volume may rise when customers trust support. He reports one customer's 80% weekly lift. That is one anecdote, not a forecast for your pilot.
The transferable move is to make the decision metric hard to game. Pick the unit that a buyer already tracks and that moves in the direction of business value. Before the pilot, agree on baseline, sample, reviewer, exclusion rules and the moment of decision. If your product saves time but creates review work elsewhere, count both. If the buyer proposes “it feels useful,” ask what observation would justify budget.
This later five-day pilot did not acquire Ask-AI's original 25 customers. It demonstrates how a founder can improve the sales proof after learning which buyer and workflow matter. Mosaic's current site shows the technical-support positioning, while its launch note describes a wider support, CX and GTM mandate. Keep the stage and wording clear when you borrow the method.
What failed, and what remains unproven
The broad enterprise assistant became a vulnerable position when general AI tools improved. Talmor did not describe a sudden replacement of every customer or a clean overnight pivot. He describes gradually building more for support teams, then reducing investment in the general assistant. The failure was strategic positioning under changing competition, not proof that the original customers were fake.
Several attractive claims are missing their denominators. We do not know exact contact volume for the first 25, conversion by channel, pilot-to-paid rate, retention, or net revenue by cohort. We cannot infer that a free year is optimal, that $30,000 was the first contract, or that a five-day pilot works for every enterprise. The company reports later customer and revenue scale, but this course teaches the earlier founder-led mechanism. Leave the unavailable numbers blank in your own worksheet rather than borrowing someone else's outcome.
Your seven-day implementation plan
Day 1: choose one buyer role and one costly repeated workflow. Write the problem hypothesis in one sentence. Build a 20-account list with owner and access route. Day 2: send ten problem-interview requests, split warm and unfamiliar accounts. Log the wording and the source of each contact. Day 3: hold the first conversations. Ask for recent incidents and current workarounds. Record exact language, budget owner and urgency.
Day 4: choose one willing design partner only if there is real case access. Write a narrow pilot agreement with a review date and commercial decision point. Day 5: compare the problem with five other buyers; use the same questions and code the responses without smoothing differences. Day 6: run a small demo or manual simulation against an actual case. Record time saved, errors, and trust breaks. Day 7: decide: proceed with the shared wedge, narrow to a different buyer, or stop. Write down the evidence that triggered the decision.
Use this worksheet for every account: buyer role; recent incident; current workaround; annual or monthly cost; existing tools; who signs; prototype job; success metric; next commitment; evidence URL or note. Leave a field empty when you do not know. That visible blank is more useful than a guessed score.
Lead Scorer implementation
Use Lead Scorer to keep the two cohorts and the gate evidence separate. Make one list for warm introductions and another for unfamiliar target accounts. Score on observable fit: buyer role, company stage, workflow complexity and a dated trigger. A fit score should prioritize research and review; it is not evidence that the account wants to buy.
Enrich only the leads that pass the fit gate. Attach source URLs for role and trigger, and flag uncertain facts. Draft a campaign for the problem-interview ask, then review the wording and recipient list before anything is sent. Keep the first message focused on the buyer's work, not on Mosaic's story or a promise of AI savings. Log responses under the relevant account and note whether they give a real example, request a demo, or decline.
After five to ten conversations, turn recurring objections into a short content note or a revised question. If buyers dispute the cost, gather baseline evidence. If they already built an internal tool, ask what it does not measure. If only warm contacts advance, inspect the ICP and access route before declaring product-market fit. Human approval remains the gate for outbound messages; the software helps organize and draft the work.
Checklist and sources
- Name one buyer and one costly repeated job.
- Separate interview, design partner, payment intent and signed payment.
- Test the same pain with five buyers outside the first partner.
- Split warm and unfamiliar acquisition cohorts.
- Keep the founder close to demos and objections until the pattern is legible.
- Agree on a metric and decision date before the pilot.
- Report unknown conversion and retention numbers as unknown.
Sources: SaaS Club's interview and full transcript, Mosaic AI's dated rebrand note, current product description, and Leaders Fund's investment account. Customer count, first-25 channel mix and early ARR are founder reported. The investor account corroborates named early customers but does not verify those totals. This article does not claim that the company reached its first 100 through the early Ask-AI motion.
Frequently asked questions
Did Ask-AI reach 100 customers through founder-led sales?
No. Founder Alon Talmor described about 25 early customers sold by him before a sales leader joined. The later company, Mosaic AI, reports roughly 100 customers, but that is a different stage.
Was the first design partner a paying customer?
Talmor says the Yotpo prototype was free for about a year. A discussed annual willingness to pay was a signal, not proof of a signed contract at that point.
Did the five-day pilot win the first 25 customers?
No. The five-day diagnostic pilot belongs to Mosaic AI's later technical-support focus. The first 25 came from founder calls, emails, connections, and an SDR service.