The First 100 Customers Course #39: Dover Sold 10 Customers Before It Had a Product
A sourced course on Dover's first ten paid buyers: manual recruiting, a narrow automation wedge, failed cold outreach, and specific customer introductions.
By Lead Scorer 18 min read The First 100 Customers Course · Lesson #39
TL;DR
Dover's first ten buyers paid for recruiting help before the startup had a product name, website, or automated workflow. Co-founder George Carollo says his team found them mainly among founder friends, then reviewed resumes, sourced candidates, and ran first interviews themselves. The founder's SaaS Club interview provides the early sequence. It does not provide contracts, buyer identities, or a conversion funnel, so the count remains founder reported.
The reproducible mechanism is paid manual delivery before selective automation. Charge for one painful job, observe where human time goes, automate only a repeated bottleneck, and ask successful buyers for specific introductions. Dover also gives an important warning: after its first 15–20 buyers, the team tried roughly 500–1,000 cold emails and outsourced content. Neither was an effective next channel in the founder's telling. This is a course in earning and learning from the first ten payments, not a promise that manual work always scales.
What you will build
By the end, you will have a one-page paid-service offer, a ten-account buyer list, a delivery log that separates repeatable work from expert judgement, and a rule for when to automate. You will also have a referral request that names two or three relevant people rather than asking a customer to “spread the word.” The output is a small, falsifiable sales system. It is suited to a founder with technical capacity but no public audience, provided the first promised result can be delivered safely and manually.
Do not use this exercise to disguise bespoke consulting as a scalable SaaS. Set a maximum number of delivery hours and a narrow outcome before taking payment. If every buyer needs a different service, or if the result depends on professional qualifications you do not hold, stop. The company example is a recruiting operation in 2019–2021; today's Dover offers a different recruiting marketplace. The historical sequence is the teaching object, not a description of its current product.
Case snapshot: keep the milestones separate
| Stage | What the sources support | Limit |
|---|---|---|
| Early 2019 | Three repeat founders began tackling recruiting. | Company retrospective confirms founding period. |
| First ten paid buyers | Primarily friends who ran companies; the team delivered recruiting manually. | Founder recollection; no customer ledger. |
| About 15 buyers | The work was still highly manual; the Dover name arrived around this stage. | Approximate timing. |
| Late 2019 to early 2020 | Roughly 15–20 buyers; cold email and outsourced content were tested. | Not part of the first-ten acquisition. |
| 2020 pause | COVID reduced hiring; Dover paused some billing and focused on product. | Enabled partly by about $3m of funding. |
| Later scale | Founder reported 5x growth and over $5m ARR; independent reporting later confirmed over 100 customers. | Growth attribution is not independently proven. |
The distinction matters. The first ten were won through trust and paid service. The later numbers reflect product development, market recovery, referrals, and renewed sales together. TechCrunch independently reported a $20 million Series A and more than 100 customers in July 2021, but did not audit Dover's first-ten funnel. A separate Mixergy interview supports the manual-service start; it is still the same founder speaking.
The model: sell the outcome, then discover the software
The sequence is a loop, not a slogan. Founder access produced an initial conversation. A paid promise forced the team to deliver actual candidate flow, not a friendly product demo. Delivery exposed where recruiters spent time: reviewing resumes, finding people, contacting them, scheduling, and screening. The team could then pick one frequent and measurable bottleneck for software. Improved execution made an introduction easier to request. More buyers supplied more workflow observations.
Every link can break. A friend may buy out of goodwill and never repeat. Manual delivery may require too much time. An automated step may improve operator efficiency without improving the buyer's outcome. A satisfied customer may know nobody in the same market. Test each link with a separate measure. Paid conversion is evidence of a problem worth paying to solve; delivery quality and repeat use are evidence of the solution; introductions are evidence of trust. None alone proves a scalable acquisition channel.
Step 1: choose a job you can finish by hand
Dover did not begin by selling an end-to-end recruiting platform. The founders went to company founders they knew and offered to handle recruiting work. Carollo says Anvisha Pai sourced and screened candidates, he reviewed resumes, and Max Kolysh sold. That division matters more than the absence of a website: someone owned the sale, someone owned the work, and a buyer could judge whether a hiring pipeline improved. The company later described the underlying waste as operational work in its Series A announcement.
Write an offer in this form: “For [one buyer] trying to [one urgent job], I will deliver [observable result] within [time box]. It costs [price]. I will do [three bounded activities] and stop at [boundary].” A software founder might manually reconcile one data report each week, produce a monitored alert for one failure class, or assemble a qualified buyer list for one niche. Do not promise a broad platform, an outcome controlled by a third party, or unlimited support. Pick a job that lets the buyer inspect the result after one or two cycles.
Checkpoint: can you name the buyer, result, deadline, price, and the manual steps in one page? Ask five people with that exact problem for a paid trial. If nobody will commit after a clear offer and useful conversation, inspect the problem and buyer before writing software. A “great idea” compliment is not a pass condition.
Step 2: use trust to get the first conversation, then insist on payment
Carollo says the first customers were mostly the founders' friends who ran startups. Trust gave them permission to try a rough service. It did not, by itself, prove the service worked. The founders charged from the start because a free favor would blur the answer: was the buyer using this because it solved a costly problem, or because they liked the founders? The podcast does not disclose first-customer price points, so a copied Dover price would be fiction.
Make a list of ten people who already own the problem, including former colleagues, clients, operator peers, and second-degree introductions. Record why each has the problem now. Message each with one question about the job and one specific offer to do it. Ask for payment before describing a future product. Keep a row for introduction source, problem trigger, offer, quoted price, decision, delivery hours, and result. Separate a buyer who paid from someone who merely agreed to a call.
Decision gate: after ten qualified conversations, count paid commitments and the reasons for refusal. One payment is a useful start, not product-market fit. If all interest depends on a deep personal relationship and the buyers will not refer peers, note that constraint. Do not hide it by expanding the list or lowering the price without learning why.
Step 3: log the labor before automating it
Dover's first roughly 15 accounts stayed highly manual over approximately nine months, the founder says. Its first automation target was resume review, because that was a large unknown and a repeated task. Carollo says the team took about three months to make that review work in an automated way. He also describes incremental automation elsewhere. This is why “Dover wrote no code for nine months” would overstate the evidence. The practical point is that manual work remained close enough to the buyer to reveal the actual workflow.
For every delivery, record each task, elapsed time, input, output, error, and buyer correction. Mark whether the task requires expert judgement, can follow a stable rule, or is still undefined. Automate a step only after it has appeared repeatedly and its quality can be measured. Dover used hiring-manager approval of presented candidates as one quality signal. Your equivalent might be accepted records, resolved alerts, or approved draft messages. A speed gain that lowers acceptance is not progress.
Pass condition: at least three paid deliveries expose the same bottleneck and a measurable output. Stop condition: every delivery needs a custom workflow, or you cannot tell whether the automated output helps the customer. Keep that part manual and narrow the offer. The first feature should make the paid job cheaper or more reliable, not make a demo look complete.
Step 4: test a second channel without confusing it with first-ten proof
Around 15–20 buyers, Dover tried to grow beyond friends and friends of friends. The team sent cold email to recently funded companies and fellow YC companies. Carollo estimated the volume at roughly 500–1,000 messages and said it produced very little lift. He did not give a verified send export, response count, or conversion rate. He thought recruiting buyers were saturated with similar promises and that Dover's optimization offer required more trust and explanation. The team also hired a content firm, but the writers lacked the underlying recruiting data and market understanding required to write useful pieces.
Treat those as two channel tests, not proof that cold email or content never works. Before sending, state what the message will test: a specific buying trigger, one problem, and a clear next step. Track replies that name the problem, not only opens. If the market is crowded, send fewer, more contextual messages and compare them with warm introductions. Before outsourcing expert content, ask whether the writer can access the observations that make it useful. If not, the founder or operator must supply the substance first.
Decision gate: after a bounded test, keep the channel only if it produces qualified conversations at a cost and effort you can sustain. Dover did not have such evidence from its early experiments. Its lesson is to stop a weak channel deliberately rather than use the existence of activity as a substitute for buyer progress.
Step 5: ask for an introduction after a visible win
Dover's later growth is often summarized as word of mouth. The transcript is more specific. After the product improved, customers began recommending it organically, according to Carollo. The team also asked buyers who had just made a hire whether they knew a relevant hiring manager. Instead of demanding an open-ended referral, they suggested three or four possible people from the buyer's network. Carollo estimated this prompted approach accounted for perhaps 10–15% of user growth. That attribution was not independently audited.
Copy the timing and specificity, not the percentage. Wait for a real delivered result, then write: “We completed [specific result]. Would it be useful for [named peer one] or [named peer two] to see the same approach? An introduction is optional; I can send a two-sentence note you can edit.” Let the customer decline. A request immediately after payment, before a result, uses trust you have not earned. Track requested, accepted, introduced, conversed, and paid as separate stages. That protects you from counting a polite yes as new demand.
Pass condition: a buyer can explain the outcome in their own words and chooses to connect you with a peer who has the same problem. Stop condition: the buyer cannot name the value received or the referral request feels broader than the result. Return to delivery quality. A referral loop starts with a product or service worth discussing.
A seven-day first-buyer plan
- Day 1: define one buyer, one paid job, one completion signal, and a delivery-hour cap.
- Day 2: list ten problem owners and record the concrete trigger for each.
- Days 3–4: hold five conversations, quote a real price, and log refusals without softening them.
- Day 5: deliver the first paid job manually if a buyer commits; timestamp every task and correction.
- Day 6: review the delivery log. Name one repeated bottleneck and one buyer quality measure.
- Day 7: decide: repeat the offer, narrow it, or stop. Ask for a specific introduction only after a buyer has a visible win.
Your worksheet should have columns for account, buyer, trigger, source, paid commitment, promised result, actual result, delivery hours, repeated step, objection, and introduction permission. A zero in the paid column is honest information. If no buyer commits, interview the refusals and revise the job. If one commits but delivery exceeds the cap, fix scope before selling a second. If repeated work appears and quality holds, prototype that one automation. The seven-day plan ends with a decision, not a claim of ten customers.
Lead Scorer implementation: a researched first-buyer queue
Lead Scorer can help with the list and review loop around this motion. It cannot perform Dover's recruiting service or manufacture trust. Use it when the paid job has a clear B2B buyer and public signals can identify firms facing that job. Skip it when the only viable path is a private, regulated, or highly bespoke relationship. Keep friends, second-degree introductions, and cold prospects in separate lists so a later channel report does not assign network buyers to outbound.
Start with the ICP and offer context skill: state the buyer role, company stage, trigger, exclusions, paid job, and evidence you may claim. Then use the signal research dossier and daily vertical prospecting skills for a small, dated account set. An agent connected to Lead Scorer's MCP can create an audience source for a relevant public post, event, or search when the signal exists, but that is not a substitute for founder introductions. Ask the agent: “Find ten companies that recently showed [trigger]. Record the source date and why each fits [paid job]. Keep network and cold sources apart. Do not enrich contacts yet.” The output is a reviewable account queue.
Next, apply the ICP scoring rubric to each account. Set a minimum score before paying for enrichment or contact discovery, and require the agent to show the reason for every keep or reject. Use contact discovery only for qualified keepers and within the configured credit gate. The prompt is: “Score this ten-account queue against [rubric]. Enrich only accounts above [threshold] after the required credit confirmation. Return source URLs, unknowns, and a list of people worth a human review.” No model score proves a buyer will pay; it narrows research effort.
For cold prospects, use cold email first touch or an AI-authored campaign to create a draft, then the outreach QA audit to inspect claims and tone. Say: “Draft one short message per qualified account about [observed trigger] and one paid, bounded job. Cite the signal. Leave every message in draft for my approval.” The founder decides whether to send, controls sender health and credits, and runs the sale. A separate, personal introduction request goes to a successful buyer only after the result is verified; it should never be generated as a mass sequence.
Finally, feed actual replies and delivery objections back into Content Studio as source notes. Use reply triage to distinguish interest, timing, wrong person, objection, and no. Turn recurring objections into useful, sourced content, then capture and score people who visibly engage before drafting any follow-up. The pass condition is a queue with dated evidence, reviewed messages, paid commitments, and a delivery log. The stop condition is an agent proposing unsupported promises, sending without approval, spending beyond the credit gate, or treating an email reply as a customer. Keep those human gates explicit.
Checklist and evidence limits
- One buyer and one paid job are specified before building.
- A delivery-hour cap prevents the service from becoming unlimited consulting.
- Network, introductions, and cold outreach have separate source labels.
- The buyer's accepted result, not completed internal tasks, is the quality measure.
- Automation follows a repeated, measurable bottleneck.
- A referral is requested only after a visible result and is always optional.
- Channel claims are tied to recorded conversations and payments, never activity alone.
The primary source is the complete SaaS Club conversation with George Carollo. Dover's company retrospective confirms the founding team and original operational thesis. TechCrunch independently confirms later funding and a 100-plus customer scale. The founder's first-ten count, early channel mix, email volume, later revenue, and referral share have no independent cohort audit. The podcast intro says $4 million ARR, while George says Dover had just crossed $5 million during the interview. This course does not use either figure as first-ten proof. COVID recovery, product work, and resumed sales also overlap, so the later growth cannot be assigned to a single cause. Dover's current marketplace should not be mistaken for its early recruiting orchestration product.
Frequently asked questions
Did Dover get its first ten customers through cold email?
No. Co-founder George Carollo says the first ten paid buyers came mainly from the founders' personal networks. The later cold email experiment, estimated at 500–1,000 messages, produced little lift.
Was Dover completely without code for nine months?
No. The first roughly 15 customers were served very manually over about nine months. Carollo also describes incremental automation, including work on resume review. The precise point when meaningful code began is not established.
Should every SaaS founder start as a service business?
Only when the founder can deliver one paid outcome manually, learn from the work, and retain a path to software economics. If delivery requires regulated expertise or open-ended labor with no repeatable bottleneck, this is a poor fit.
Can Lead Scorer send referral requests automatically?
Lead Scorer can help research and qualify a narrow account set and draft reviewed messages. A customer introduction is a relationship request; the founder chooses whether and when to ask, and approves any outbound message.