Harper: Agents Read the Paperwork, People Close the Deal

Key points

• Agents handle the sorting of paperwork and the chasing between intake and the final decision.
• Unlicensed intake specialists verify the facts of each application first.
• Instead of accounts, veteran brokers and underwriters are hired to write the tests the agents must pass.

Entering an insurance brokerage market stuck in paperwork

Small businesses such as child care centers, construction contractors, freight fleet operators, restaurants, and manufacturers need several kinds of insurance to operate. Typical examples are workers’ compensation, which pays when an employee is hurt on the job, general liability, which covers accidents at the business, and professional liability, which covers mistakes in the work. But these businesses often go through a broker instead of buying directly from an insurer.

What an insurance broker does is more than introduce a product. The broker has to understand how the client’s business runs, fill out the application, and decide which insurer to ask. After sending the submission, the bundle of application documents sent to insurers, the broker has to keep contacting the underwriter, the insurer’s risk reviewer, and chase for a quote. When a quote comes in, the broker explains the options to the customer and closes the deal. Then comes a long stretch of after-sale work: issuing certificates of insurance, changing policies, managing payments, and renewals.

A large share of this long flow still runs by hand. According to a report that Ivans, an insurance connectivity company, released in August 2026 from a survey of more than 700 agents, agencies focused on commercial insurance re-key data more than 70% of the time. Only 7% use dedicated submission capture software, and 49% still use email. Among those agencies, 54% said they lose commercial deals every month because they cannot reach the right insurer fast enough.

Harper, founded in San Francisco in 2024, took aim at this bottleneck head-on. Dakotah Rice (CEO) and Tushar Nair (CTO), both formerly of Goldman Sachs, first planned AI software to sell to existing brokers. But they soon dropped the idea and decided to become a licensed brokerage themselves. Rice told TechCrunch that his family ran an insurance brokerage. Announcing its funding, Harper declared: “We are the brokerage.” Instead of building work tools for other companies, it started as a licensed agency that owns the customer relationship directly.

Harper went through the Y Combinator (YC) Winter 2025 batch. In February 2026, it announced a combined $47 million in seed and Series A funding led by Emergence Capital. According to TechCrunch, the company is connected to more than 160 insurers.

Placing agents between the phone call and the close

The full picture of how Harper works shows up in the details of the company’s many job postings. From them, agents handle reading, passing along, and chasing the paperwork, while people are placed at the two ends of the pipeline and at the point that controls quality.

The customer’s first entry point is a phone call. Harper is building a team of about 15 intake specialists in Atlanta, Georgia. Intake specialists do not need a separate insurance sales license. Their job is to take the customer’s call, fill out the application in real time, and pass the file to the next stage only when the facts, the open questions, and the next action are clear. The job posting stresses accuracy: “A fast call with a wrong answer in it is worse than no call.”

According to an intake specialist posting from February 2026, Harper’s voice AI systems “support and increasingly run intake calls.” Intake specialists use the voice AI on real calls and flag the spots where the agent loses context or asks the wrong question. In a voice engineering posting, Harper said its voice agent is “at 1% of where we need to be.” Still, over the long run it plans to hand the voice system outbound calls that qualify leads, round-the-clock answers to insurance questions and certificate requests, and the whole 20-to-30-minute phone application.

Once intake is done, the agents get to work in earnest. According to Harper, agents handle reading applications, routing submissions, and following up with underwriters. Rice told TechCrunch that AI takes care of submission routing, underwriter follow-ups, document collection, and pipeline management. According to a memo from Emergence Capital, the system sweeps across the whole network of wholesalers, managing general agents (MGAs), and insurers, compares each insurer’s appetite, pricing patterns, and underwriting preferences, and picks the best fit.

But chasing underwriters is not yet fully automated. According to one forward-deployed operations posting, staff spend one day contacting an underwriter directly to chase a stalled quote, and the next day shipping “the automation that makes sure no quote stalls again.”

Agents support service after the deal closes, too. The system gathers the customer account’s earlier records, the policies bound, and new requests in one place, and when a renewal date approaches, it shows the policies about to expire and the exposures that have changed. Agents also draft the routine follow-up messages in advance. Service representatives, who do not need a license, take inbound calls and contact customers from lists of payments, document collection, and renewals. Harper set rules of three attempts over three days for standard follow-ups and up to seven touches for payment issues. At the same time, it drew a clear line: “Quoting and coverage advice belong to licensed teammates.”

Making the final insurance offer to the customer and closing the contract is the job of the account executive. The account executive, the role responsible for the business relationship with customers at Harper, is split into four lanes: new business, renewals, partnerships, and service. New business, renewal, and service account executives must get a commercial insurance producer license within about 30 days of starting. A new business account executive takes over small-business owners whose interest and terms are already confirmed. They make more than 100 calls a day and are measured on the number of policies closed and revenue. In the posting’s own words, “AI now does most of the work of getting a business quoted,” “but a human still closes.” Renewal and service account executives read the policy and the contract requirements, talk with the placement experts, and settle the final plan.

According to an underwriting posting, the authority to choose the insurer, add the narrative to the submission, negotiate with the insurer, and deliver the final recommendation to the customer must rest with the licensed account owner. A brokerage posting also states that the placement team, licensed brokers, underwriters, and account owners keep the final decision on each account.

Giving veteran experts the exam questions instead of accounts

The most unusual part of Harper’s organization is a quality group that does not handle accounts. The company hires senior commercial insurance brokers and senior underwriters who do not own individual customer accounts and do not approve risks themselves.

These veterans do not advise customers. They write and grade the standards the AI agents must follow. They build test cases, expected results, grading criteria, and failure cases. They also review whether what the agent produced matches the real source documents.

They write routing rules that say when a submission with a given risk may be sent to a certain insurer, and when sending must be blocked or handed to a senior expert for review. Every rule has to carry its evidence and reason. The underwriting posting states the reason plainly: “A routing rule at Harper can move thousands of accounts.” If one rule is set wrong, the same mistake repeats on every customer account that passes through that route. Senior experts continuously audit the system’s decisions. When a correction comes up, they trace the wrong rule or the misread source document and turn it into the system’s permanent evaluations and product improvements.

Harper's quality improvement loop

Pulling the knowledge out of experienced practitioners’ heads and into writing is run as a separate role. Harper has an operating memory lead who collects operating knowledge from call recordings and Slack conversations and turns it into playbooks and decision logs. The principle is: “A workflow that isn’t clear enough for a new hire isn’t clear enough for an AI agent either.” If the same problem shows up three times, it becomes an official playbook, a quality check, training material, or a product requirement. Even how a salesperson overcame a customer’s objection is recorded as official data the company keeps.

In a recruiting post, Harper said its operations are “90%+ AI-led, pushing toward the higher 90s.” The YC profile lists 25 employees, and as of October 2026 there were 36 open job postings. In its 2026 job postings, the company says about 1,000 new customers come in each month, and as of February 2026 it had supported more than 5,000 businesses in its first 13 months.

Three points where it splits from traditional brokerages

Harper’s way of working differs from a traditional insurance brokerage in three major ways.

The first is where people’s time goes. At a typical independent brokerage, staff take an application and re-key it by hand into each system, then email several insurers one by one to chase for quotes. The time of key staff is tied up in the middle steps of passing along paperwork and repeating contact. Harper handed this whole middle stretch to agents. In its place, it gathered human labor at the two ends. At the front end, unlicensed intake specialists firmly pin down the accuracy of the data. At the back end, licensed sales professionals persuade customers and close the deal. Rice claimed to TechCrunch that a traditional brokerage’s sales team handles 20 to 30 deals a month, while Harper takes in more than 1,000 customers each month. This is not an independently verified statistic, but it is the company’s account of how its staffing differs.

The second is the appearance of entirely new roles. At a traditional brokerage, the value of a seasoned broker or underwriter comes from looking at individual contracts one by one and judging the risk. Harper does not give its veterans customer files. Instead, it has them write out their intuition and expertise as rules, test cases, and exception conditions that the system must follow. Quality control has moved from signing off on individual files to verifying the rules of the whole system.

The third is the speed customers experience. Harper claims that small-business customers can get insurance “in 24-48 hours,” not weeks. Rice said that work a traditional broker takes five to seven days to handle can be finished in one to two days at Harper. This figure has not been objectively verified by an outside body either, but it clearly shows the operating speed the company aims for.

The formula for professional services could change

If Harper’s way of operating is proven all the way in the market, it could be a serious shock to the existing brokerage industry.

A traditional brokerage had to add sales staff and account managers in proportion every time the customer accounts it managed grew. This is because there is a physical limit to how much paperwork one employee can handle. But if agents take over the sorting and chasing of paperwork, a firm can grow its customer count fast without adding much staff. Traditional brokerages, whose labor costs rose in direct proportion to the number of contracts, would come under heavy cost pressure.

Also, the speed of finding and connecting to the right insurer could itself become a core competitive edge. If many commercial insurance brokers are losing deals because they cannot find the right insurer in time, as the Ivans survey says, an agent system that checks more than 160 insurers at once to find the best product could be a differentiator. But how well Harper’s insurer matching actually works has not been independently confirmed.

Organizations in other regulated industries can take three important principles from Harper’s structure.

First, a division of labor that requires a license only for the final decision, not for the whole process. Harper strictly separated work that legally needs a license, such as advice, quote calculation, and closing the contract, from work that needs no license, such as intake, follow-up, and payment checks. It engraved this boundary clearly in every job description. That keeps expensive licensed professionals from spending their time on simple paperwork.

Second, giving top internal experts the job of grading the system instead of individual tasks. Instead of asking veteran staff to process more files, Harper made them the examiners who set the standards and catch errors to see whether the system judges correctly. The mistakes found there do not end as one-time fixes. They become new permanent test data that toughens the system.

Third, a culture of writing down how work is done before automating it. Harper treats undocumented personal know-how as a kind of system defect. Work that cannot be passed clearly to a new hire cannot be taught to an agent either.

This way of operating has not yet been fully proven at large scale. But a structure in which agents take over the paperwork in the middle, licensed people make the final decisions, and the best experts in the field write the system’s grading standards could well apply to other paperwork-heavy, tightly regulated professional services, such as accounting, mortgage brokerage, and corporate benefits administration.

What has not yet been fully proven

There are clear limits to keep in mind before being optimistic about Harper’s future.

First, most of what is known about how Harper works rests on the company’s own job postings, funding press releases, and an early investor’s memo. The several teams that appear in the postings are still being staffed, and the voice AI system, as the company itself admits, is still at an early stage.

Second, the true share of all work that agents handle without human involvement, and the error rates in passing along submissions and matching insurers, have not been published. The “90%+ AI-led” figure is also the company’s own phrase from a recruiting notice, not a result of any formal audit.

Finally, the figures on customer count, growth, and turnaround time are all Harper’s own claims. The comparison with traditional brokerages also rests on a single Ivans survey and CEO Rice’s personal claims. In particular, in the Ivans survey, agencies specializing in commercial insurance were only about 10% of the whole sample, around 70. Whether the speed and efficiency Harper promises can hold at large scale remains to be seen.

---

Sources

1. Emergence Capital — Harper: Building the AI-Native Insurance Brokerage Every Business Deserves — https://www.emcap.com/thoughts/harper-building-the-ai-native-insurance-brokerage-every-business-deserves — 2026-02-25
2. Harper — Series A announcement — https://www.harperinsure.com/news/series-a-announcement — 2026-02-26
3. Y Combinator — Harper company profile — https://www.ycombinator.com/companies/harper — accessed 2026-10-22
4. Harper — job posting, Account Executive, New Business — https://jobs.ashbyhq.com/harperinsure/52bdda0e-d294-485c-9657-8462d485908c — 2026-07-08
5. Harper — job posting, Account Executive, Service — https://jobs.ashbyhq.com/harperinsure/4919f9ef-73ce-4f9d-ae70-0c6856635d88 — 2026-09-24
6. Harper — job posting, Account Executive, Renewals — https://jobs.ashbyhq.com/harperinsure/9a109415-7062-403a-8c44-a535edd8489f — 2026-09-24
7. Harper — job posting, Customer Intake Specialist — https://jobs.ashbyhq.com/harperinsure/c90b946d-8dae-48a1-9914-64383c8e7b31 — 2026-02-23
8. Harper — job posting, Customer Service Representative — https://jobs.ashbyhq.com/harperinsure/81cdb8e5-7be0-4eae-ac26-46fcdc7fd235 — 2026-09-23
9. Harper — job posting, Senior Commercial Broker, Brokerage Intelligence — https://jobs.ashbyhq.com/harperinsure/141fc972-0309-43ee-975e-b3677b966bdc — 2026-09-24
10. Harper — job posting, Senior Commercial Underwriter, Underwriting Intelligence — https://jobs.ashbyhq.com/harperinsure/57521df4-7480-4ac4-90be-5a97650939bf — 2026-06-02
11. Harper — job posting, Operating Memory Lead — https://jobs.ashbyhq.com/harperinsure/5c1d20be-2113-4552-a4c3-b7546f2b274f — 2026-05-28
12. Harper — job posting, Forward Deployed Operations — https://jobs.ashbyhq.com/harperinsure/585d2c4a-06c4-4bb1-91cc-5f70c2a04b18 — 2026-06-19
13. Harper — job posting, Senior Operations Lead (Y Combinator job board) — https://www.ycombinator.com/companies/harper/jobs/KxUUfX3-senior-operations-lead — accessed 2026-10-22
14. Harper — job posting, Senior Member of Technical Staff, Voice (Y Combinator job board) — https://www.ycombinator.com/companies/harper/jobs/j3AYc4c-senior-member-of-technical-staff-voice — accessed 2026-10-22
15. TechCrunch — Y Combinator grad and AI insurance brokerage Harper raises $47M — https://techcrunch.com/2026/02/25/ai-insurance-brokerage-harper-raises-45m-series-a-and-seed/ — 2026-02-25
16. Risk & Insurance — Re-Keying And Workflow Friction Are Costing Carriers Placement, Ivans Survey Finds — https://riskandinsurance.com/re-keying-and-workflow-friction-are-costing-carriers-placement-ivans-survey-finds/ — 2026-08-28