Direct answer
Harper is the Main Street case of outcome-vertical integration. It is a licensed commercial insurance brokerage that rebuilt quoting, submission, and service around AI, and then kept the customer. It does not sell a seat to Marsh or a local independent. Dakotah Rice’s line, in the company’s own raise note: “Harper isn’t selling software to legacy brokerages and hoping they figure it out. We are the brokerage. We own the customer relationship.”
That is the same make-or-buy call as Crosby, pointed at a different profession. Crosby owns the signed contract and the malpractice. Harper owns the bound policy and the producer-of-record relationship. Neither owns a plant. Both refuse to rent the layer where the learning happens.
In February 2026 Harper announced $46.8 million (rounded to $47 million) in combined seed and Series A, led by Emergence Capital, with Y Combinator, Peak XV, Antler, and others. Company and press figures at that moment: 5,000-plus businesses served in about 13 months, later described as 6,000 to 7,000; 160-plus carriers; turnaround of one to two days versus five to seven at a typical shop; a claimed throughput of 1,000 customers a month versus 20 to 30 for a human sales team. About 65 to 70 percent of the book is excess-and-surplus, the harder risks admitted markets have been leaving. Headcount in mid-2026 sat in the low tens on PitchBook and founder interviews (Antler still said about 25 in August). Treat operating metrics as company-reported.
Harper scored 34 out of 40 on a non-factory rubric set beside Crosby, tied with AllDigital Specialty, behind Crosby’s 38. Crosby is a comparison, not a separate essay on this site.
xvertical does not work for Harper. Harper is not a client. The company is here only as a case a reader can check.
Where this sits next to the other cases
| Case | Owned layer | Industry |
|---|---|---|
| SpaceX | The vehicle, built in-house | Rockets |
| BYD | Cell to car, plus driving silicon | Auto and energy |
| Anduril | Lattice, plus enough hardware and factory | Defense |
| Harper | Bound policy and the customer | Insurance distribution |
Owning a plant and owning the customer outcome are both vertical integration. Do not flatten them. Harper does not take insurance risk. The carrier does. If you need the risk-stack version, that is AllDigital Specialty, a managing general agent that claims about 70 percent autonomous decisions on other people’s paper. That is a different case.
What vertical means in a brokerage
The commercial-insurance chain for a small manufacturer or a bar:
- Find the buyer (the daycare, the fleet, the shop)
- Collect applications and loss runs
- Route the risk to the right market (admitted versus excess-and-surplus, which managing general agent, which underwriter)
- Chase quotes
- Bind coverage
- Service endorsements, certificates, and renewals
- Get paid a commission
A broker-tech vendor owns a slice of steps 2 to 4 and invoices the broker. Harper owns steps 1 to 7. Harper Hub, the internal flagship, auto-completes forms, matches carriers and underwriters, and runs follow-ups. Licensed people still sit on the file. The product the customer buys is coverage in force, not a dashboard.
Emergence, the Series A lead, is explicit that the founders looked at selling tools to incumbents and abandoned it. That abandoned path is the control experiment. Software into a fragmented, email-and-spreadsheet industry dies in implementation. Owning the book lets every submission train the next one.
Rice’s stated ambition is the AI-native version of Brown & Brown, Marsh, or Gallagher for small business, then compliance, payroll, and personal lines. That is a land-and-expand speech. The integration that is real today is distribution: producer of record plus an internal operating system. Everything after “and then payroll” is a roadmap.
The numbers that matter
| Claim | Figure | Source | Date |
|---|---|---|---|
| Founded | 2024, kitchen table, then Y Combinator W25 | Company / The Insurer | 2024–25 |
| Founders | Dakotah Rice (CEO), Tushar Nair (CTO) | Company | — |
| Combined seed and Series A | $46.8 million / $47 million, Emergence lead | TechCrunch / company | 25 Feb 2026 |
| Other backers | Y Combinator, Peak XV, Antler, 10X Founders, Fellows Fund, Outset, Lobster | Company / press | 2026 |
| Customers at raise | 5,000-plus in about 13 months | Company / TechCrunch | Feb 2026 |
| Customers later | About 6,000 (The Insurer); 6,000 to 7,000 (Antler) | Press | 2026 |
| End-2026 target | 100,000 customers | CEO to The Insurer / Antler | 2026 |
| Carriers | 160-plus | Company / TechCrunch | 2026 |
| Early premium run-rate | More than $6 million annualized, 35 states (Y Combinator launch note) | Y Combinator profile | Early book |
| Lines | Workers’ compensation, general liability, professional liability; 3 to 4 lines per account typical | Company / The Insurer | 2026 |
| Mix | 65–70% excess and surplus | The Insurer | Feb 2026 |
| Ticket | Small and mid-sized firms up to about $500,000 premium | The Insurer | Feb 2026 |
| Cycle time | 1 to 2 days versus 5 to 7 typical (sometimes stated as 24 to 48 hours) | CEO / TechCrunch / Antler | 2026 |
| Throughput claim | 1,000 customers a month versus 20 to 30 for a human team | CEO | Feb 2026 |
| Team | About 25 (Antler, August 2026); about 32 (PitchBook) | Mixed | 2026 |
| Named markets | Manufacturers, healthcare, hospitality, transportation, construction, bars | Company raise note | 2026 |
How to read the table. The $47 million and the “we are the brokerage” sentence are solid. Customer counts moved from 5,000 to 6,000–7,000 across the year; do not stitch them into a precise growth chart. The $6 million premium figure is an early Y Combinator note, not 2026 gross written premium. 1,000 customers per month and 100,000 by year-end are founder targets. Cycle-time comparisons are against a stylized traditional shop, not a named competitor’s service level.
Why this is integration, not insurtech
Three tests.
-
They kept the profit and loss that software vendors try to rent. Commission sits on the bound policy. Whoever owns the customer owns the data that makes the next submission cheaper. Selling Hub to independents would have given Harper a software multiple and someone else’s book. They took the messier multiple and the book.
-
The operating system is internal on purpose. Harper Hub is not the product. The product is the placed coverage. That is the Crosby pattern, agents inside the firm, and the opposite of Applied, Fulcrum, or Comulate, which sell into existing agencies.
-
They picked the hard inventory. 65 to 70 percent excess-and-surplus is not personal-auto clickwrap. Surplus lines are messy applications, surplus-lines affidavits, wholesale relationships, and underwriters who still want a human. If the thesis only worked on admitted, vanilla general liability, it would be a form-filler. Harder risk is where a scaled operating system has something to eat.
What they do not pass: they do not own the capital that pays claims. Vertical integration stops at distribution. That is a smaller stack than an AI-native carrier or a full-stack managing general agent that keeps a share of risk. Say that out loud or the analogy to BYD overclaims.
What Harper still does not own
- The balance sheet. AIG, Hiscox, James River, USLI, and the rest take the risk. Harper takes commission. A hard market that tightens excess-and-surplus capacity can starve the operating system of product.
- The underwriting model. Matching and routing are not pricing. AllDigital’s claim is “the model declines or binds.” Harper’s claim is “the model finds the market faster.”
- Proof of 100,000 customers. That is a 2026 goal stated to the press, not a booked number.
- Independently audited throughput. 1,000 customers per month versus 20 to 30 is a founder ratio. Useful as a design target. Not a teardown.
- The back office of American small business. Payroll, compliance, life, and personal lines are a speech. Shipping them without becoming a distracted conglomerate is the 2027 problem.
- A published evaluation. Crosby released Redline Bench. Harper has not published an equivalent “did the submission land on the right desk” score. The loop may be real. It is not yet citable as a number.
Same four-part test
| Test | Factory version | Harper version |
|---|---|---|
| Expensive layer is scarce | Cell, wafer, seeker | Producer time, wholesale relationships, and messy risk data |
| Volume amortizes the stack | Cars off a line | Thousands of small firms on one Hub |
| Model co-designed with the work | Chip next to the motor | Agents trained on Harper’s own brokers and files |
| You are the first customer | Own fleet | Harper’s producers are Hub’s first users |
It fails when a startup sells “AI for agencies” and the agency still owns the insured. The learning then sits in a customer that can leave the vendor.
What operators should copy
- If you already know the tool should exist, run the service that tool would have enabled. Keep the customer.
- Automate the minutes: forms, routing, and chases. Keep a license on the bind.
- Start in the ugly inventory. Vanilla risk is where incumbents already have a process.
- Measure cycle time and placement, not model demos.
- Do not announce the conglomerate before the first line of business is a machine.
Do not copy the 100,000-customer slide. Copy the decision to stop selling software and own the book.
Key takeaways
- Harper is outcome-vertical integration in insurance distribution: own the insured, own the workflow, rent the capital.
- $47 million combined seed and Series A (February 2026), Emergence-led. 5,000 to 7,000 small and mid-sized firms, 160-plus carriers, majority excess and surplus.
- The founders explicitly rejected “sell software to brokers.”
- Speed and throughput are company-reported. There is no public evaluation bench, and no balance-sheet risk.
- Crosby still scores higher on liability and evidence. Harper is the one a shop owner would recognize.
Sources
- Harper, “We’re Announcing $47M,” 26 February 2026. harperinsure.com
- TechCrunch, “Y Combinator grad and AI insurance brokerage Harper raises $47M,” 25 February 2026.
- Emergence Capital, “Harper: Building the AI-Native Insurance Brokerage,” 25 February 2026.
- The Insurer, stealth exit, excess-and-surplus mix, and the 100,000 target, 25 February 2026.
- Antler founder story, August 2026 (team size, 6,000 to 7,000 customers).
- Y Combinator company page (early premium run-rate, Hub features).
- PitchBook, Harper (Insurance Brokers) profile (headcount, round stack).
Cycle times, monthly throughput, and year-end customer targets are founder or company statements. Customer counts are not a single audited time series.
Questions this essay answers
What is Harper?
An AI-native commercial insurance brokerage. It places workers' compensation, general liability, and professional liability for U.S. small and mid-sized businesses, with a heavy excess-and-surplus mix.
Is Harper an insurance company?
No. It is a retail broker and licensed agency. Carriers and managing general agents take the risk. Harper takes commission and owns the customer.
Why does that count as vertical integration?
Because Harper owns intake, submission, market-matching, binding coordination, and service in one firm, with an internal operating system called Harper Hub. A broker-tech vendor owns only the software layer.
How much has Harper raised?
$46.8 million, usually rounded to $47 million, in a combined seed and Series A announced 25 February 2026, led by Emergence Capital.
How many customers does Harper have?
More than 5,000 at the raise; later press said about 6,000 to 7,000. The 100,000 figure is a year-end 2026 target, not a current count.
How is Harper different from Crosby?
Same pattern, different owned outcome. Crosby is counsel of record and takes legal liability. Harper is producer of record and does not pay claims.
How is Harper different from AllDigital Specialty?
AllDigital underwrites on carrier paper and claims about 70% of decisions are autonomous. Harper distributes. The deeper risk stack is at AllDigital. The broader Main Street distribution story is at Harper.
Should every brokerage copy this?
Only if they will keep the book and rebuild the workflow. Bolting a chatbot onto email and spreadsheets is not this model.