Exploring the $47 Million YC Unicorn Harper: AI-Native Reconstruction of the $380 Billion Commercial Insurance Practice
Dakotah Rice
Harper, Founder
Original Statement
1. Entrepreneurial Background and Core Business Model Reconstruction
1. Generational Evolution of AI Business Paradigms
• From selling tools to becoming the business itself: Early AI startups mostly focused on selling "efficiency tools" to traditional industries, while the top AI-native startups are evolving to become the business itself, leveraging AI-native capabilities to capture market share and revenue from traditional giants.
• Reshaping the $380 billion U.S. commercial insurance market:
• Pain points for traditional brokers: Comparing quotes across insurance carriers usually takes weeks.
• Harper's efficiency breakthrough: Connecting over 160 insurance underwriters, compressing the inquiry, comparison, and issuance process for commercial insurance to 1-2 days or even hours (with a goal of same-day, second-level locking).
2. Financing and Capital Endorsement
• YC W25 batch star project: Rapidly exploded after graduating from Y Combinator's Winter 2025 batch.
• $47 million massive financing: Led by the well-known SaaS/enterprise service venture capital firm Emergence Capital, completing a combined seed and Series A round financing totaling $47 million.
• Demand capture business: Any entity in the U.S. is required to have commercial insurance, creating a naturally surplus market demand, with the key being whether the backend infrastructure can handle the massive influx of traffic at venture capital scale.
2. Geek-Driven Culture and Extreme Work Hours
1. Extreme schedules and "Gladiator" battle philosophy
• Arriving at the office at 5:15 AM: Sales, operations, and engineering teams report to work every morning at 5:15 AM, with a morning tactical alignment meeting at 6:30 AM, maintaining high-intensity operations throughout the day.
• Continuous fighting for 18-19 hours a day: Founder Dakota and the team have maintained 18-19 hours of high-intensity work for several years, solving organizational and sales funnel bottlenecks during the day, and entering "Night Shift" mode late at night, outlining the next day's action framework by 2-3 AM and distributing it to everyone.
• High-energy work habits: Executives avoid traditional breakfasts to prevent energy depletion, relying on C4 high-energy energy drinks to maintain high focus throughout the day.
2. Everyone as a Cursor and the evolution of the "Software Factory"
• Eliminating cross-department silos and implicit knowledge: All company phone recordings, emails, and messages are consolidated into a central knowledge base to avoid information fragmentation.
• Command interaction and AI collaboration ("Cursor Cursor Cursor"): The founder can invoke Cursor to record voice and generate analysis reports during meetings and daily management.
• Business operations (Ops) autonomously build applications: Engineers focus on creating the underlying "software factory," while operations team members directly use AI programming tools like Cursor to independently write internal tools and business systems, quickly addressing their own business pain points.
3. Sales Funnel Repair and Data Troubleshooting Details
1. Fixing the leaky funnel
• Deep diving into key conversion rate indicators (KPI):
• Rejecting mere faith in "First Pass Rate."
• Core indicators restructured to "actual quote return quantity and turnaround time" (Quote volume & turnaround speed), combined with follow-up frequency and underwriting accuracy.
• Month-end sprint strategy: For policies effective August 1, targeted combination sales for different vertical clients at month-end:
• Emphasizing "Inland Marine Insurance" (for tool and equipment protection) for contractors/renovation teams.
• Adding "Professional Liability" and "Workers' Comp" for high-revenue companies.
2. Production environment data parsing failures and on-site troubleshooting
• Data synchronization parsing pitfalls: Due to system changes, quote parsing errors occurred, with basic premiums and miscellaneous fees all classified as "Fees & Taxes" instead of being automatically split.
• Severe impact on conversion: When sending a $10,000 policy to a client, the interface displayed it as $10,000 pure tax fees, causing client misunderstandings; the founder intervened on-site, coordinating the engineering team for urgent fixes and requiring operations to manually verify policy entries.
3. Individual tracking of 38 stalled leads
• Microscopic review of every lead: The founder organized funnel penetration meetings every 1.5 hours, tracing each of the 38 high-value leads (e.g., a single lead costing $246, quoted at $2,000 but not paid as promised) to prevent lead sinking.
4. Voice AI Automation Full-Process Advancement
1. Extreme launch of multi-industry Voice AI agents in 2-3 days
• Rapidly expanding vertical categories: Launched three new vertical tracks over the weekend, including retail stores, smoke shops, and consultants, with plans to expand to 10 verticals that week.
• Comprehensive underwriting of four-in-one insurance types: Voice AI can simultaneously collect and complete information verification for general liability, cybersecurity, workers' compensation, and property insurance in one call.
2. Automated pre-identification (eliminating 6-7 minutes of redundant Q&A)
• Reverse lookup of company profile by phone number: Integrating Voice AI with the internal "Big Brother" data system allows instant pre-filling of basic information such as company name, registered address, business scale, etc., eliminating a lengthy 6-7 minute basic information inquiry.
3. Unlimited R&D budget tilt
• Prioritizing Voice AI lead testing as top-tier R&D investment: Regardless of single lead costs (even if a single lead costs $400-500), fully injecting test traffic to allow AI agents to run 50+ real-time calls concurrently.
• Completely decoupling manpower from business scale: Aiming to achieve 100% automation of all-channel inbound business by mid-next week, fully decoupling business scale from employee count.
5. Organizational Advancement: From "Founder Micromanagement" to "Systematic Delegation"
1. Beware of the founder becoming the company bottleneck
• From personal firefighting to establishing systematic delegation: The founder realizes that if all problems are personally solved, macro growth, strategic partnerships, and board preparations cannot be managed.
• A dual-track mechanism of "check and delegate": Clearly defining specific indicators for each business leader, but initially confirming system operation through 2-3 high-frequency checks before shifting focus to board reports and long-term strategy.
2. Founder dual-core collaboration mechanism
• "Double Tango" collaboration structure: Dakota focuses intensely on GTM, customer lifecycle, detail penetration, and sales rhythm; another co-founder hedges collaboration on technical architecture, leverage capability building, and backend resource support, ensuring team direction aligns with underlying computing power/products.
• Words to entrepreneurs: Entrepreneurship is "getting into the arena and fighting like hell," leading the team to continuously break barriers and achieve compound growth with extreme diligence, attention to detail, and a thirst for victory.
Video source: https://www.youtube.com/watch?v=5tjNU_hsERk
ABAB AI Insight
This episode is very worthwhile to do.
Because what Harper truly represents is not just "AI helping insurance companies improve efficiency," but a rapidly emerging new business paradigm:
AI is no longer just sold to traditional industries, but directly enters traditional industries and takes away their own revenue.
This is completely different from traditional SaaS.
The logic of traditional SaaS is:
"I sell software to insurance brokers to improve their efficiency."
Harper's logic is:
"Why should I only charge software fees? I directly become an insurance broker, reconstructing the entire cost structure with AI, and then earn brokerage commissions myself."
This is what Emergence Capital is currently emphasizing as AI-Native Services: not selling tools, but directly delivering outcomes, controlling customer relationships, and capturing the revenue pool of traditional service industries. Emergence clearly positions Harper as a representative company of this investment thesis.
────────────────
1. First, calibrate some very important facts
1. Harper cannot currently be called a "unicorn"
This is the most critical part that needs to be modified in your title.
What Harper announced in February 2026 is:
Cumulative financing of approximately $47 million in Seed + Series A.
The company's figure is $47M, and TechCrunch reports the precise number as approximately $46.8M, with Series A led by Emergence Capital.
However:
$47 million in financing ≠ company valuation of $1 billion.
Currently, there is no official announcement from Harper or mainstream financing reports confirming its valuation has reached $1B.
So the formal title should not write:
"$47 million YC unicorn Harper."
This is not valid in financial terms.
More accurately:
"YC AI insurance newcomer Harper with $47 million in financing"
or:
"YC star AI insurance company Harper."
────────────────
2. The second correction is the "$380 billion commercial insurance market"
This figure is not completely baseless, but it is no longer the market metric I would recommend using.
S&P Global statistics show that direct premiums written for U.S. commercial insurance lines alone will reach approximately $50.235 billion in 2024.
And YC/Harper emphasizes not the Premium Pool, but:
The annual commission revenue pool generated by commercial insurance brokers exceeds $100 billion.
These two figures must not be confused.
Because:
$500 billion+
is the premiums paid by insurance customers.
$100 billion+
is closer to the revenue pool that insurance distribution/brokers can compete for.
So the most professional way to analyze Harper should be divided into three layers:
Premium Pool → Brokerage Revenue Pool → Market that Harper can serve.
Instead of simply writing:
"This is a $380 billion market."
────────────────
3. The third very important correction: Harper is essentially a Brokerage, not an insurance underwriting company
This is key to understanding the business model.
Harper is:
AI-native commercial insurance brokerage.
That is:
AI-native commercial insurance brokerage.
It is not primarily a Carrier that bears insurance payout risks.
Harper's current role is more like:
Customer
↓
Harper
↓
Finding Carrier / Underwriter
↓
Getting Quote
↓
Customer purchases Policy
↓
Carrier bears insurance risk.
Harper earns:
Brokerage Commission / Fee.
Harper's own partner page also clearly states that its legal entity DBA is Harper Global Insurance Agency and earns brokerage commissions through referral policies.
This distinction is extremely important.
────────────────
4. Therefore, what Harper is truly trying to disrupt is not "insurance risk pricing," but "insurance distribution"
This is a completely different business.
The insurance industry can roughly be divided into:
Risk Manufacturing
and:
Risk Distribution.
Carriers are responsible for:
Underwriting.
Pricing.
Reserves.
Reinsurance.
Payouts.
Capital adequacy.
Brokers are responsible for:
Customers.
Demand analysis.
Filling Applications.
Finding Carriers.
Submission.
Quote.
Compare.
Bind.
Renewal.
COI.
Endorsement.
Claims Support.
Harper first attacks the second part:
Distribution Economics.
And this layer happens to have a lot of:
Emails,
PDFs,
Calls,
Forms,
Repetitive inputs,
Manual Follow-ups.
Very suitable for LLM + Agent.
────────────────
5. This is why commercial insurance may be one of the most perfect landing industries for AI
The industry most suitable for AI is not necessarily:
"The most technologically advanced industry."
Instead, it is often:
Information-rich, poorly structured processes, high labor costs, and valuable outcomes.
Commercial insurance fits almost all of these criteria.
A customer may involve:
Company type.
Address.
Payroll.
Revenue.
Employees.
Vehicles.
Property.
Claims History.
Subcontractors.
Cyber Exposure.
Equipment.
Industry risks.
Then different insurance companies:
Have different Applications.
Different underwriting appetites.
Different exclusions.
Different follow-up questions.
This creates huge:
Information Transformation Costs.
The previous solution was:
More people.
AI's solution is:
More Compute.
────────────────
6. The real value of Harper is turning "Human Judgment" into "Compute"
I believe this is the core layer of the entire episode.
Traditional Brokerage wants to grow:
1,000 customers
may need:
X Brokers.
5,000 customers:
More Brokers.
10,000 customers:
Continue hiring.
This is a:
Headcount-Coupled Growth.
Revenue is highly tied to employee count.
What Harper wants to do is:
1,000 customers:
20 people.
5,000 customers:
Maybe 30 people.
10,000 customers:
Does not need to become 200 people.
This becomes:
Compute-Coupled Growth.
YC currently describes Harper as a team of about 25 people; the company claims to have served over 5,000 businesses, while recruitment materials state it is growing at about 1,000 customers per month.
This is where VCs are truly excited.
────────────────
7. Why the economic value of "selling AI to insurance companies" versus "becoming the insurance business" may differ by dozens of times?
Assuming you develop an insurance AI SaaS.
You help a Brokerage:
Create $5M in profit annually.
You sell SaaS:
$200,000/year.
The customer gains:
$4.8M Value.
You gain:
$200K.
But if:
You become a Brokerage yourself,
Directly control the Customer.
The same automation capability may allow you to capture:
The entire Brokerage Revenue.
This is called:
Value Capture.
Creating value through technology is one thing.
How much value you can capture is another.
────────────────
8. This is a significant upgrade in business models emerging in AI entrepreneurship
First generation:
AI Tool
"Helps lawyers write documents."
Second generation:
AI Copilot
"Helps lawyers complete more work."
Third generation:
AI-Native Law Firm
"I directly become a law firm."
The same can happen in:
Accounting.
Recruitment.
Insurance.
Real estate brokerage.
Loan brokerage.
Logistics brokerage.
Property management.
Advertising companies.
Customer service BPO.
Security management.
Medical administration.
Emergence defines this type of company as AI-Native Services.
────────────────
9. Why this model may be larger than traditional SaaS?
Because SaaS competes for:
IT Budget.
AI-native services compete for:
Labor + Service Revenue.
Global labor costs far exceed the global SaaS market.
For example:
Traditional companies may only be willing to spend:
$20K a year on software.
But willing to spend:
$500K a year on services.
If AI can compress:
$500K service costs
to:
$100K,
AI companies have the opportunity to earn far more than ordinary SaaS seat pricing.
This is where the true magnitude of AI business lies.
────────────────
10. But "Software Margins in Services" cannot be just a marketing slogan
Many VCs will say:
AI can allow Services to achieve Software Margins.
Theoretically, this is correct.
But you must truly look at Unit Economics.
Harper's core revenue logic can be roughly understood as:
Premium Volume × Commission Rate + Fees
minus:
Lead CAC.
Sales.
Service Labor.
Licensed Brokers.
AI Infrastructure.
Compliance.
Customer Support.
Errors & Omissions.
Management.
Finally obtaining:
Contribution Margin.
Therefore, to judge whether Harper is successful in the future, one cannot only look at:
Customer numbers.
Must look at:
Revenue / Employee.
Contribution Margin / Account.
Automation Rate.
Retention.
────────────────
11. Here is a particularly important financial knowledge: Premium is not Harper's Revenue
For example:
A customer buys:
$20,000 Premium insurance.
This does not mean:
Harper earns $20,000.
Assuming the brokerage commission is only an example of:
15%.
Then Brokerage Revenue:
$3,000.
The Carrier receives the main Premium and bears the payout risk.
So in the future, when looking at any InsurTech data, one must distinguish between:
Premium Written
and:
Revenue.
YC early disclosed that Harper had reached over $6M in annualized premiums, which absolutely cannot be directly understood as $6M ARR.
This point is very important.
────────────────
12. This also makes the "$246 Lead → $2,000 Quote" in the video particularly worth analyzing
Ordinary people see:
Lead Cost:
$246.
Policy:
$2,000.
They might think:
Very profitable.
Wrong.
$2,000 is the Premium.
Assuming purely for example:
Commission 15%.
Harper's actual Revenue:
$300.
If CAC is already:
$246,
The first transaction has almost no profit.
But if this customer:
Continues to renew for 5 years,
The second year starts not needing to spend $246 again,
Also purchases:
Workers' Comp.
Property.
Cyber.
Auto.
Then:
LTV
may rapidly increase.
This is the real beauty of insurance Brokerage:
Renewal Economics.
────────────────
13. Therefore, one of the most valuable assets for insurance brokers is not the first transaction, but the Book of Business
Insurance usually renews annually.
A customer buys this year.
May continue next year.
Continue the year after.
Thus, quality Brokerages will continuously accumulate:
Recurring Commission Streams.
It is very similar to:
SaaS Subscriptions.
The difference is:
Subscription corresponds to Software.
Insurance Brokerage corresponds to:
Annual Renewal.
So what Harper should truly optimize is not just:
Customer Acquisition.
But also:
Customer Lifetime Value.
────────────────
14. This is why Cross-Sell is very important
A Contractor originally only has:
General Liability.
Later adds:
Workers' Comp.
Commercial Auto.
Inland Marine.
Umbrella.
The Brokerage Revenue contributed by a customer can significantly increase.
At the same time, the cost for customers to switch Brokers also begins to rise.
This is:
Wallet Share Expansion.
But high-quality Brokerages must be based on real risk needs, rather than selling insurance randomly for commissions.
Long-term trust is far more valuable than a one-time Upsell.
────────────────
15. The statement that "all U.S. businesses must buy commercial insurance" needs correction
It is inaccurate.
Not all businesses are legally required to purchase all commercial insurance.
Workers' Compensation is usually mandated by state law, but there are differences based on state, business size, and ownership structure; for example, the U.S. Department of Labor links the Workers' Comp system to state regulatory systems.
General Liability:
Is usually not uniformly mandated by law across the U.S.
However:
Landlords.
Banks.
General Contractors.
Clients.
Licensing Authorities.
Financing parties
may require COI or specific Coverage.
So the real advantage of commercial insurance is not:
"Legally mandated for everyone to purchase."
But rather:
Structural Demand.
────────────────
16. Commercial insurance is still a typical "High-Intent Demand"
What does this mean?
A boss suddenly needs:
COI
to start work tomorrow.
Or:
The bank requires insurance to close.
Or:
The client requires Cyber Coverage to sign a contract.
At this point, their willingness to purchase is extremely strong.
Not:
"I’m just browsing."
But rather:
I need coverage now.
Therefore, if Harper can reduce:
7 days
to:
1 day,
the speed itself is product value.
────────────────
17. Harper's publicly available data indeed supports this point
Harper claims to have served over 5,000 U.S. businesses and compressed the commercial insurance process that originally could take days or longer to about 24-48 hours; TechCrunch quotes Dakotah Rice stating that traditional Brokers often take 5-7 days, while they can achieve 1-2 days.
Note:
The real competitive advantage is not:
"AI is cool."
But rather:
Time-to-Coverage.
Customers do not care whether GPT, Claude, or a proprietary model is used in the backend.
Customers only care:
When will I get my insurance?
────────────────
18. This is also an important principle that AI companies should learn: sell Outcomes, not AI
A poor pitch:
"We use Agentic AI."
Customer:
So what?
A good pitch:
"Originally 7 days, now 24 hours."
Even better:
"If you don’t get COI today, you can’t start work tomorrow; we solve it today."
This is:
Economic Outcome.
Technology must translate into:
Time.
Revenue.
Cost.
Risk.
────────────────
19. Harper's choice of complex commercial insurance rather than the simplest insurance is also very smart
The more complex the risk:
The higher the labor costs.
The more fragmented the information.
The more complex the Carrier Matching.
The more valuable the Broker expertise.
The value created by automation is also higher.
Harper currently publicly emphasizes real operational industries such as manufacturing, healthcare, hospitality, transportation, and construction.
This has a greater:
Automation Surplus
than just doing:
The simplest small store General Liability.
────────────────
20. Especially the E&S market is very suitable for AI
Harper's public job descriptions describe the business as a commercial E&S brokerage.
E&S, or Excess & Surplus Lines, primarily serves non-standard risks that the admitted market finds difficult to underwrite:
Non-standard risks.
Special risks.
New risks.
High-capacity risks.
The NAIC clearly states that these types of risks often lack sufficient loss history, making traditional standardized pricing difficult, thus requiring a more flexible non-standard underwriting market.
This means:
The higher the complexity, the more valuable the manual information matching.
And AI excels at handling such complex information.
────────────────
21. Therefore, Harper Hub can truly be understood as an "insurance Operating System"
Harper has publicly introduced:
Harper Hub will centralize:
Customer data.
Underwriter data.
Deal history.
Then drive down:
Form filling.
Carrier matching.
Communication.
Voice AI.
If successful,
What it truly accumulates is not:
"A few AI Prompts."
But rather:
Operational Data Flywheel.
────────────────
22. This data flywheel may become Harper's true moat
Assuming:
The 1,000th customer.
The system learns:
Which risks suit which Carrier.
The 5,000th.
Knows:
Which issues most easily lead to Declines.
The 20,000th.
Knows:
How a certain type of Contractor:
Fills out Submission most effectively.
Which Underwriter:
Is more willing to accept what risks.
Which Supplementary Data can improve Quote Rates.
This forms:
Workflow Data.
The model itself may become increasingly commoditized.
However:
The data and feedback loops of real insurance transactions will not automatically commoditize.
────────────────
23. This is why "AI Wrapper" is completely different from a true AI company
Thin Wrapper:
Prompt
↓
LLM
↓
Output.
Very easy to replicate.
A true AI-native company:
Customer Data
↓
Workflow
↓
Human Feedback
↓
AI Agent
↓
Transaction
↓
Outcome
↓
Back to Data.
Once this closed loop is formed:
Competitors not only need to replicate your Software,
But also replicate:
Your experiential history.
────────────────
24. Why Voice AI may be one of Harper's most important products?
Harper's own recruitment materials even state:
"Voice is where the industry lives."
The company hopes to expand Voice AI from hundreds of calls a day to thousands, covering:
Lead Qualification.
Customer Service.
Underwriting.
Even 20-30 minute insurance Applications.
Why are calls so important?
Because the biggest Input Problem in commercial insurance is:
Customers are unwilling to fill out 50 fields.
────────────────
25. Essentially, Voice AI is a "Conversational ETL Machine"
Traditionally:
Customer speaks
↓
Person listens
↓
Person takes notes
↓
Person fills CRM
↓
Person fills ACORD / Carrier Forms.
AI:
Customer speaks
↓
Speech
↓
Structured Data
↓
Validation
↓
Insurance Fields
↓
Carrier Submission.
So the true value of Voice AI is not:
"AI will chat."
But rather:
Turning unstructured human information directly into structured business data.
────────────────
26. This is also why the "auto-pre-filling data by phone number" in the video is so important
If the system can already know:
70% of basic information through:
Phone Number.
Company Name.
Address.
Public Data.
Internal History.
Then the Voice Agent does not need to ask:
"What is your business name?"
"What is your address?"
"How long have you been operating?"
The entire Application:
20 minutes
may become:
8 minutes.
This is called:
Progressive Data Enrichment.
A truly good AI product does not make AI ask more questions.
But rather:
Makes customers answer fewer questions.
────────────────
27. But this also hides Harper's biggest risk: data errors
Assuming the system automatically thinks:
Customer Revenue:
$2M.
In reality:
$20M.
Or classifies:
Restaurant
as:
Nightclub.
These fields may directly affect:
Eligibility.
Premium.
Coverage.
Underwriting.
So the real difficulty for insurance AI has never been:
"Generating answers."
But rather:
"Guaranteeing Correct State."
────────────────
28. The "$10,000 all parsed as Fees & Taxes" in the video is a very typical AI-native operational risk
This issue appears to be:
Parsing Bug.
But in the financial and insurance system, it is different.
If the UI tells the customer:
Premium:
$0
Fees & Tax:
$10,000,
The customer may:
Refuse to pay.
Complain.
Misunderstand the product.
More seriously: errors can enter Binder,
Policy,
Accounting.
So the financial industry truly needs:
Exception Handling.
Not just Automation.
────────────────
29. Mature AI companies should not pursue "100% Automation"
But rather:
Maximum Safe Straight-Through Processing.
For example:
80% standard cases:
AI fully automatic.
15%:
AI + Human Check.
5% high risk:
Licensed Expert.
This is much more mature than:
"Everything 100% AI"
Harper itself also clearly states in its YC description that the AI system is learning from the expertise of Brokers, gradually moving towards high autonomy, rather than claiming that all insurance judgments are completely unmanned today.
────────────────
30. Voice AI also has a very practical issue: regulation
The FCC has clearly confirmed that AI-generated or simulated human voices fall under the regulatory scope of "artificial or prerecorded voice" in the TCPA, and corresponding telephone activities are subject to consent and other requirements.
The FTC's Telemarketing Sales Rule also has specific rules for:
Do Not Call,
Disclosure,
Payment Authorization
etc.
So the more successful Voice AI becomes:
Compliance Engineering
becomes increasingly important.
This is the huge difference between financial entrepreneurship and ordinary AI demos.
────────────────
31. Looking again at the most eye-catching part of the video: starting work at 5:15 AM
This is not entirely for show.
Harper's current recruitment page indeed states:
The team usually arrives at the office between 5:15-5:45 AM, and the office operates at high intensity until 7-9 PM; the Account Manager position even publicly lists a Monday-Friday 5 AM-8 PM schedule.
So:
This company indeed has an extreme:
Founder Intensity Culture.
────────────────
32. Why might this culture be effective in the early stages?
Because what startups lack is not:
Efficiency.
But rather:
Learning Speed.
One experiment a day.
A year:
250 times.
Three experiments a day.
A year:
750 times.
As long as the quality of learning does not decline,
After a few years, the knowledge difference is terrifying.
Therefore, Hyper-Intensity truly buys:
Compressed Calendar Time.
What others learn in a year,
Can be learned in 6 months.
────────────────
33. But do not mistake "18-hour workdays" as a business moat
This is a cold splash of water I want to pour on entrepreneurs.
Lack of sleep does not automatically lead to:
Good judgment.
Insurance is also:
Highly regulated,
High error costs,
High customer liability
industry.
If long-term extreme fatigue leads to:
Judgment errors.
Code Bugs.
Incorrect Coverage.
Compliance Failures.
Then speed becomes:
Negative Leverage.
Therefore, the real advanced goal is not:
"Everyone always works 18 hours."
But rather:
"Gradually solidifying Founder Heroics into Systems."
────────────────
34. The first phase of entrepreneurship can rely on heroism, but the second phase must rely on systems
Day 1:
Dakotah personally reviews 38 Deals.
Reasonable.
The 5,000th customer:
If the CEO still personally reviews every Deal,
It indicates the system is broken.
The real upgrade should be:
Founder identifies problems
↓
Establish Metrics
↓
Find Root Causes
↓
Establish Processes
↓
Automate
↓
Delegate
↓
Only handle Exceptions.
This is called:
Operational Scaling.
────────────────
35. The "38 stalled leads reviewed one by one" in the video is very similar to the Toyota production system
This is not mere Micromanagement.
The real question is:
Why:
38 Deals
are stuck here?
Each bottleneck may belong to:
Customer.
Sales.
Underwriting.
Carrier.
System.
Payment.
Follow-up.
If you only look at:
Total Conversion Rate,
You will never know the real Bottleneck.
So the Founder should early on:
Touch the Work.
────────────────
36. This is called Gemba thinking
Toyota has a famous management concept:
Gemba—go to where the work actually happens.
The CEO should not just look at:
Dashboard:
Conversion:
17%.
But ask:
Why didn’t that Plumbing Company pay yesterday?
Why:
There’s a Quote, but
The customer still doesn’t buy?
You must actually open:
Calls.
Emails.
CRM.
Quotes.
To discover:
It may not be that Sales is not working hard.
But rather:
The quote is too slow.
The page is broken.
The customer doesn’t know the next step.
AI filled in the wrong fields.
────────────────
37. So truly strong operations are not KPI management, but "metrics → raw facts"
In Harper's video:
First Pass Rate
looks good.
But:
What customers really care about is:
Quotes.
So better metrics may become:
Quotes per Account.
Time-to-First-Quote.
Time-to-Bind.
Quote-to-Bind Rate.
Rather than:
"Internal step completion rate."
This is a very important principle of corporate management:
Do not optimize internal metrics, optimize customer outcomes.
────────────────
38. This is also why the "Leaky Funnel" thinking is very suitable for AI-native services
The insurance sales Funnel can be written as:
Lead
↓
Contacted
↓
Qualified
↓
Application Complete
↓
Submission
↓
Quote
↓
Customer Selection
↓
Payment
↓
Bind
↓
Renewal.
At each stage:
There is Conversion.
If:
There are many Leads,
But few Binds,
It is not:
"Add more Leads."
But rather:
Find the Leak.
────────────────
39. Many startup companies' biggest mistake is to cover up product problems with traffic
Assuming:
1,000 Leads.
Only:
20 Binds.
CEO:
"Buy 5,000 Leads."
Wrong.
You should first clarify:
Why did 980 not convert.
Otherwise:
CAC will only explode.
This is completely aligned with the previous episode's discussion:
Retention Before Growth
Do not give a leaking bucket:
A larger faucet.
────────────────
40. The $400-$500 per Voice AI Lead in the video cannot simply be labeled as "burning money"
If the goal is:
Short-term Customer Acquisition,
This may be very expensive.
But if these calls are also generating:
Voice Training Data.
Objection Data.
Underwriting Data.
Failure Cases.
Conversation Labels.
Then part of the cost should actually be understood as:
R&D Spend.
That is to say:
They are not just "buying customers."
They are also:
Buying learning.
────────────────
41. But after the company matures, it must separate the two accounts
The first account:
Customer Acquisition Economics.
The second account:
Model Training Economics.
If you always use:
"We are still learning"
To explain:
$500 CAC,
Then the business model may never be established.
Ultimately, it must be seen that:
CAC ↓
Conversion ↑
Automation ↑
LTV ↑
Otherwise:
AI has not truly changed the economic model.
────────────────
42. The interesting part of "Everyone as Cursor" is not the tool itself
When Harper is currently recruiting for Forward Deployed Operations, it explicitly requires candidates to be able to:
Use SQL,
Python/TypeScript,
And actually use Cursor, Claude Code, and other AI Coding tools.
Cursor's recent activities have also allowed the Harper team to directly demonstrate how they use these tools in business.
This indicates that Harper is attempting a very interesting new organization:
Everyone becomes partially technical.
────────────────
43. This is a significant change in organizational design that may emerge in the AI era
Previously:
Operations:
Identifies problems.
↓
Writes Tickets.
↓
Product Manager.
↓
Engineering.
↓
Sprint.
↓
Launch two weeks later.
Now:
Operations:
Identifies problems.
↓
Cursor.
↓
Internal Tool on the same day.
Engineering team:
Responsible for Platform, Security, Architecture.
This is called:
Software Creation Decentralization.
────────────────
44. In the future, excellent companies may no longer be divided into "technical personnel" and "non-technical personnel"
But will become:
Builders
and:
Non-Builders.
Sales can Build.
Ops can Build.
Finance can Build.
Founder can Build.
Marketing can Build.
AI programming is turning:
Software Development
From:
A profession
Gradually into:
A general business capability.
This has a huge long-term impact.
────────────────
45. But "everyone can write code" also has new enterprise risks
If all employees can:
Randomly generate Apps,
Modify databases,
Deploy Automation,
It will create:
Shadow IT.
Security Holes.
Data Leaks.
Duplicate Logic.
Broken Workflows.
So a truly mature "software factory" must establish:
Permissions.
Testing.
Observability.
Version Control.
Production Gates.
Otherwise:
AI increases Build Velocity, but also increases Blast Radius.
────────────────
46. This is why the "production environment Parsing Bug" in the video is the most worth learning layer
AI makes development fast.
But:
Shipping Faster ≠ Shipping Safer.
In the past, a feature:
Took two weeks to develop.
Three days of QA.
Now:
Generated in 20 minutes.
So people may have the illusion:
"It should be in Production in 20 minutes."
On the contrary.
After the cost of generation decreases:
Testing Governance
Becomes increasingly important.
────────────────
47. Dakotah's biggest organizational challenge will ultimately be: Will the Founder become a Bottleneck?
This awareness has already appeared in the video.
In the early stages:
He personally:
Sales.
Ops.
Funnel.
Customer.
Hiring.
Very good.
But as the scale grows:
The CEO's most scarce time should gradually shift to:
Capital Allocation.
Key Hires.
Carrier Relationships.
Regulation.
Strategic Partnerships.
Category Strategy.
Board.
If he still personally:
Resolves orders,
The company will be capped by the CEO himself.
────────────────
48. Correct delegation is not "I don't care anymore"
It should be:
Define → Instrument → Inspect → Delegate.
First:
Define goals.
Second:
Establish metrics.
Third:
Continuous checks.
Fourth:
Confirm the judgment of the responsible person is correct.
Fifth:
Exit daily execution.
Sixth:
Only handle exceptions.
This is very similar to flying an airplane:
Autopilot
Is not:
No Pilot.
But rather:
Pilot transitions from:
Continuous operation
To:
Exception Manager.
────────────────
49. The Dakotah + Tushar dual Founder combination is also very worthy of study
Dakotah's background:
Goldman Sachs,
Carlyle,
Coatue,
Very focused on:
Finance + Strategy + GTM.
Tushar has long worked on ML/AI engineering at Goldman.
The two previously co-founded Poolit; Poolit once reached about $100M AUM but ultimately failed to establish a sustainable profit model and closed in 2023.
This background is very important.
Because Harper is not:
Two first-time entrepreneurs stumbling upon AI.
But rather:
Second-time Founder Reps.
────────────────
50. The failure of Poolit may instead be an asset for Harper
Dakotah later publicly admitted:
Poolit should have closed earlier,
His ego delayed his acknowledgment that the business model was not working.
This is actually consistent with what we discussed earlier about Ben Horowitz's:
Decision Debt
Excellent Founders ultimately discover:
Mistakes are not scary.
What is truly costly is:
Continuing to make mistakes after knowing they are wrong.
────────────────
51. Harper truly embodies an important path in AI entrepreneurship
The first idea:
"Sell AI to Brokers."
Failed or lacked appeal.
Then transformed into:
"Why not directly become a Broker?"
This is a business model-level Pivot.
Not:
Feature A
Changed to:
Feature B.
But rather:
Changing the Value Capture Layer.
This is very advanced entrepreneurial thinking.
────────────────
52. What industries are most suitable for AI-native services?
I would provide a very clear judgment framework.
Best to meet:
First, high labor costs.
Second, a lot of unstructured work.
Third, high repetition.
Fourth, outcomes can be quantified.
Fifth, the industry is highly fragmented.
Sixth, customers are already willing to pay a lot.
Seventh, AI can take on 50%-90% of the work.
Eighth, regulation/licensing instead forms entry barriers.
Insurance:
Fits very well.
────────────────
53. Therefore, the largest AI companies in the future may not look like software companies
They may superficially appear as:
Insurance brokers.
Accounting firms.
Law firms.
Recruitment companies.
Logistics companies.
Property companies.
Healthcare companies.
But the internal economic structure:
Is extremely Software-like.
This may create a new type of company:
Software-Defined Service Company.
────────────────
54. This may even overturn a classic rule from the SaaS era
In the past:
Software Eats the World.
The logic of Marc Andreessen's 2011 statement:
Software companies replace traditional businesses.
The next phase may become:
AI Eats the Service Provider.
Not:
"Insurance brokers use AI."
But rather:
"AI-native Brokerage replaces traditional Brokerage."
The distinction is very significant.
────────────────
55. But Harper is still far from true victory
This must be stated.
Financing:
Is not victory.
5,000 customers:
Is a good signal,
But still very early.
What truly needs to be observed are at least seven things.
First:
Retention.
Will customers renew in the second year?
Second:
Gross Margin.
Is it really close to Software-like?
Third:
Carrier Relationship.
Do Carriers like the business quality brought by Harper?
Fourth:
Loss Quality / Risk Selection.
Are the customers brought by Harper worse risks?
Fifth:
Automation Rate.
Can it really reduce headcount?
Sixth:
Compliance / E&O.
Can AI errors be controlled in the long term?
Seventh:
Revenue per Employee.
Does it continue to improve as scale expands?
────────────────
56. Especially Carrier Relationship is a moat that many AI entrepreneurs will underestimate
Insurance is not:
If customers are willing to buy,
They will definitely close the deal.
Carriers decide:
Whether to Quote.
What price.
What Coverage.
What Exclusion.
Harper currently publicly states it has over 165 underwriter relationships; TechCrunch reports its network as exceeding 160 insurance carriers.
Once matured:
Carrier Access
May even be more important than:
AI Model.
────────────────
57. Traditional giants are definitely not waiting to be disrupted
Marsh McLennan.
Aon.
Arthur J. Gallagher.
Brown & Brown.
These large insurance brokers have:
Decades of customer relationships.
Global Carrier Networks.
Professional Brokers.
Claims Expertise.
Data.
Acquisitions.
Corporate Relationships.
Harper's advantages:
Speed.
Software.
AI-native architecture.
The giants' advantages:
Distribution + Trust + Relationships + Scale.
Therefore, this war will not be easy.
────────────────
58. The moat that Harper may truly form should be a "four-layer overlay"
Not just AI.
But rather:
First layer: Customer Distribution
Increasing numbers of enterprise customers.
Second layer: Carrier Network
Increasing numbers of Underwriter Relationships.
Third layer: Workflow Data
Knowing which risks should go where.
Fourth layer: Automation Economics
Also Revenue, requiring fewer employees.
Only when all four layers are established:
Can it be a true:
Compounding Moat.
────────────────
59. If I were to write a core formula for Harper from a financial perspective
It would be:
Enterprise Value ≈ Customer Book × Retention × Commission Economics × Automation Leverage × Growth
And not:
"How powerful their model is."
Ultimately, the capital market cares about:
How much:
Durable Future Cash Flow
This company can generate.
────────────────
60. What this episode truly deserves for ordinary entrepreneurs to replicate is definitely not starting work at 5 AM
What should truly be replicated is:
First:
Do not just ask who AI can be sold to, but ask whether AI can allow me to directly become a new supplier in this industry.
Second:
Find an expensive, labor-intensive, repetitive problem that customers are already willing to pay for.
Third:
Founders must penetrate into specific transactions early on, rather than just looking at the Dashboard.
Fourth:
Every time a manual problem is solved, ask if it can be converted into software.
Fifth:
Do not automate actions, but automate complete Outcomes.
Sixth:
Do not sacrifice reliability for "100% AI."
Seventh:
Short-term rely on Founder Intensity, long-term must rely on Systems.
────────────────
61. The entire episode can actually be condensed into this evolutionary path for companies
Traditional Brokerage:
Human does work
↓
AI Assisted Brokerage:
Human + AI
↓
AI Native Brokerage:
AI does work, Human handles exceptions
↓
Autonomous Service Company:
Software owns most workflow
↓
Ultimately:
Revenue scales faster than Headcount.
This is what investors are truly betting on.
────────────────
62. Looking one level higher, Harper may represent a super entrepreneurial wave for the next decade
One of the best entrepreneurial questions of the past twenty years has been:
"Which industry has not yet been transformed by Software?"
The question for the next decade should be:
"Which huge service industry still pays most of its revenue to labor, while AI can take on a large part of the cognitive labor?"
Once this question is established,
The opportunity may not be:
$100/month SaaS.
But rather:
Directly reallocating the entire industry Revenue Pool.
Insurance is just one of them.
Accounting,
Law,
Recruitment,
Real estate,
Logistics,
Consulting,
Marketing,
Customer service,
Financial services
All have similar possibilities.
────────────────
63. Therefore, I believe the most valuable lesson from Harper is not "AI + insurance"
But rather a more universal formula:
AI Capability × Legacy Labor Cost × Ownership of Customer = New Company Category
If:
AI is strong,
But you do not control the Customer:
You are just a Vendor.
If:
You control the Customer,
But AI has not reduced costs:
You are still a traditional Service Company.
Only the combination of both:
Can lead to:
AI-Native Service Giant.
This is the true value of Harper.
────────────────
64. I recommend thoroughly upgrading the title and removing "unicorn"
Your original second direction is actually the most insightful.
My most recommended formal course title:
"From Selling AI Tools to Directly Becoming the Business: How Harper Restructures U.S. Commercial Insurance Brokerage with $47 Million"
If more trend-focused:
"AI No Longer Just Sells Software: How Harper Turns Traditional Insurance Brokerage into a 'Software Company'"
If more entrepreneurship-focused:
"From Failed Entrepreneurship to $47 Million Financing: Harper's AI Insurance, Extreme Execution, and Business Model Leap"
If more finance-focused:
"How AI Eats into Service Industry Profit Pools: Dissecting Harper's Insurance Brokerage, Commission Economics, and Automation Flywheel"
If you want the most shareable:
"Stop Selling AI to Traditional Companies: Directly Become Traditional Companies"
As your formal course, I most recommend the first one.
Because the biggest cognitive leap in this episode is this sentence:
The smartest entrepreneurs of the SaaS era sold software to industries; a more radical group of entrepreneurs in the AI era is starting to directly become the industry.
And whether Harper can ultimately become a super company does not depend on whether employees start work at 5 AM, nor whether Cursor can generate 100 tools in a day.
What truly determines its value is:
Whether it can transform the insurance brokerage model, which heavily relies on human experience, into a new capital machine driven by data, software, and intelligence, where Revenue growth outpaces Headcount growth in the long term.
If this holds true, Harper's significance far exceeds that of an InsurTech.
It will prove a completely new template for AI companies.
Harper is still in a period of rapid change, and I can continue to track its next round of financing, valuation, customer growth, and real automation rates.
Tracking Harper's subsequent progress.
D