Direct answer
AllDigital Specialty is the underwriting-stack case. Harper owns the customer and rents the capital. AllDigital owns the decision, to bind, decline, or send to a person, and still rents the balance sheet.
It is a Princeton, New Jersey managing general agent, founded in 2019, writing directors and officers, employment practices, and fiduciary liability for small private companies and nonprofits on admitted and non-admitted A-rated paper. It is not a licensed insurance company. AXIS provided early capacity in 2020. Core Specialty added excess in 2022. The carriers hold the claims. AllDigital holds the authority to select the risk.
Chief executive Athula Alwis told Insurance Business in May 2026 that roughly 70% of incoming business is handled without a person: 30 to 40% auto-approved, about 30% auto-declined, the rest to a human. That is stage three of his four-stage model: recommend, assist, execute, orchestrate. Bindable self-service quotes are claimed in two to three minutes. The firm says its U.S. admitted system has supported more than $100 million of premium over four-plus years with five underwriters. A U.S. patent on the liability-risk-selection method issued in 2022 (US 11,379,927).
On a non-factory rubric set beside Crosby, AllDigital tied Harper at 34 out of 40, behind Crosby at 38. Harper won distribution. AllDigital wins the layer Harper does not touch. Crosby is a comparison, not a separate essay on this site.
xvertical does not work for AllDigital Specialty. AllDigital Specialty 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 | Rented layer |
|---|---|---|
| SpaceX | The vehicle, built in-house | Some outside parts |
| BYD | Cell to car, plus driving silicon | Some foundry, some commodities |
| Anduril | Lattice, plus enough hardware | Some autonomy software, and the government budget |
| Harper | The insured, and producer of record | Carrier capital |
| AllDigital | Risk selection and the underwriting system | Carrier paper and surplus |
If you only read Harper, you will think outcome-vertical insurance means “be the broker.” That is half the chain. The expensive scarce layer in specialty insurance is not the certificate of insurance. It is the underwriter who will say yes to a $2 million directors-and-officers tower for a 40-person nonprofit at a price that still makes a loss ratio.
What vertical means at a managing general agent
The management-liability chain:
- Submission: application, financials, loss history, news
- Appetite test: does this risk even belong here
- Price and form
- Bind and issue
- Hold the capital that pays the claim
- Reinsurance
A carrier-tech vendor owns a slice of steps 2 and 3 and invoices the managing general agent. A retail broker owns step 1 and a relationship. AllDigital owns steps 2 to 4 on its own book, trains the same system for other managing general agents and carriers through AllDigital AI Services, and stops before step 5. That stop is the honest boundary.
Alwis’s line is the industrial analogue of “only solve in hardware what must be solved in hardware”: give the decision to the system. No legacy policy-administration system, and no twenty-year underwriting desk to retrain. The firm was founded as AI-first because there was nothing to migrate.
The dual motion matters. They underwrite, and they license the platform. Crosby refused that split. AllDigital sells the tool to peers. That is why the score is 34 and not 38: the spine is owned on their book, and the same spine is also a product. They design the decision engine, and they also sell the design.
The numbers that matter
| Claim | Figure | Source | Date |
|---|---|---|---|
| Founded | 2019, Princeton, New Jersey | Company / PitchBook | — |
| Live writing | About January 2021 | Insurance Business | 2025 |
| Form | Managing general agent, plus an AI-services vendor | Company | 2026 |
| Lines | Directors and officers, employment practices, fiduciary, follow-form excess | Company | 2026 |
| Paper | Admitted and non-admitted, A-rated | Company | 2026 |
| Early capacity | AXIS (2020); Core Specialty excess (2022) | Business Wire / Reinsurance News | 2020–22 |
| Autonomy mix | About 70% with no human: 30–40% auto-bind, about 30% auto-decline, about 30% review | Chief executive to Insurance Business | May 2026 |
| Quote time | 2 to 3 minutes self-service (the site now says 3) | Chief executive / company | 2025–26 |
| Premium, U.S. admitted | More than $100 million over 4-plus years, 5 underwriters | Company “Augmented” page | 2026 |
| Premium, earlier site copy | More than $80 million in four years | Company insurance page | Undated |
| Lloyd’s system | 80% of underwriting workload automated; “loop testing” | Company | 2026 |
| Training cycle | 6 to 9 months, then a person in the loop | Chief executive | 2025 |
| Patent | US 11,379,927, management of liability risk selection | USPTO / Justia | Granted 5 July 2022 |
| Headcount | About 17 | PitchBook | 2026 |
| Known capital | Seed about $2.2 million; later venture history is sparse | PitchBook | 2019–22 |
| AI maturity, own model | Stage 3 of 4: execution, not agent-to-agent orchestration | Chief executive | May 2026 |
How to read the table. The 70% split and the “five underwriters, $100 million” line are chief executive or company statements, not a statutory filing. Use the range of $80 million to more than $100 million of cumulative written premium, not a single-year gross written premium. PitchBook’s 17 people is the cleanest size signal: this is a specialist bench, not a 2026 mega-round. Capacity partners can change. AXIS and Core Specialty are the public ones.
Why this is deeper than broker software
The model says no. Harper’s system finds a market. AllDigital’s system is a market, within the authority the carrier granted. Auto-decline at 30% is the tell. A routing engine that never declines is a form-filler.
The unit of scale is the underwriter, not the producer. Five people in front of nine figures of premium is the industrial claim. If true, the scarce license is leveraged by the system, not multiplied by hiring.
They trained on decisions, not documents. Six to nine months with an underwriter in the loop, then continuous grading. That is closer to lawyers scoring redlines than to a chatbot on a form. The patent describes combining questionnaire answers with outside news and social data on a permissioned ledger. Treat the ledger line as 2019 architecture, not as the 2026 story. The story is the decision loop.
They started empty. Alwis said there was no legacy data mess and no staff revolt. That is the privilege of a firm founded in 2019. It is also why “just add AI to a 1928 mutual” is not this case.
What AllDigital still does not own
- The capital. When a directors-and-officers claim lands, AXIS, Core Specialty, or the current paper writes the cheque. A capacity partner who walks ends the product even if the model is perfect.
- The retail customer. Brokers still own the insured. AllDigital is wholesale. Harper’s advantage is the small-business relationship. AllDigital’s advantage is appetite and speed for the broker.
- A single product identity. There is the managing-general-agent book, and there is the “no AI team, no problem” vendor motion. The second motion is real revenue optionality and a diluted thesis.
- Stage four. Alwis puts them at execution, not agent-to-agent orchestration. Believe him.
- Public loss ratios. Autonomy without a combined ratio is an operations story, not an underwriting story.
- Broad evidence. One serious Insurance Business interview, company pages, a 2022 patent, and two capacity press releases. No published evaluation bench like Crosby’s Redline. The 70% figure will be repeated. It has not been audited in public.
Same four-part test
| Test | Factory version | AllDigital version |
|---|---|---|
| Expensive layer is scarce | Cell, seeker, edge node | Underwriting authority and labelled decisions |
| Volume amortizes the stack | Units off a line | Submissions through one appetite model |
| Model co-designed with the work | Chip next to the motor | Model trained on this bench’s binds and declines |
| You are the first customer | Own fleet | Their own book is the first deployment of their own AI services |
It fails when an incumbent licenses a model and leaves appetite, pricing, and decline authority in a 40-page underwriting guide no system can execute. It also fails if AllDigital becomes only a vendor and stops being the first customer of its own engine.
What operators should copy
- Own the decision, not the chatbot. Bind and decline in the same system.
- Measure autonomy as a mix: automatic yes, automatic no, and a person. Do not measure it as “AI-enabled.”
- Be the first book the model sees. Then, if you must, license it.
- Start where the file is thin and the judgment is patterned: small-firm management liability, not open-ended casualty.
- Publish a ratio you can defend, such as premium per underwriter or minutes to a bindable quote. Do not publish a vision of orchestration you have not reached.
Do not copy “we have a patent.” Copy the empty start and the 30% automatic decline.
Key takeaways
- AllDigital is outcome-vertical integration at risk selection, one layer deeper than Harper and one layer shallower than a balance-sheet carrier.
- Company and chief executive figures: about 70% decided without a person, quotes in 2 to 3 minutes, $80 million to more than $100 million of cumulative premium, five underwriters, and patent US 11,379,927.
- The form is a managing general agent on A-rated paper. AXIS and Core Specialty are the named capacity. Capital is rented.
- They also sell the platform. That is the score gap versus Crosby.
- Harper is on this site. Crosby is not.
Sources
- Insurance Business, “How AllDigital Specialty built an AI-first insurer from the ground up,” 27 May 2026.
- Insurance Business, “AI and cloud computing reshape small commercial insurance,” 1 July 2025.
- AllDigital Specialty, homepage, Insurance, and Augmented / AI Services pages, 2026.
- USPTO / Justia, US 11,379,927, granted 5 July 2022.
- Business Wire, AXIS partnership, 25 August 2020.
- Reinsurance News, Core Specialty arrangement, 23 September 2022.
- PitchBook, AllDigital Specialty Insurance profile (employees, funding stack).
The autonomy mix, premium totals, and underwriter counts are company or chief executive statements. They are not statutory financials. Capacity partners are those named in public releases and may not be the full current panel.
Questions this essay answers
What is AllDigital Specialty?
An AI-first managing general agent in Princeton, New Jersey, that underwrites directors and officers, employment practices, and fiduciary liability for small private companies and nonprofits, and licenses the same underwriting platform to other managing general agents and carriers.
Is AllDigital an insurance company?
No. It is a managing general agent. Carriers provide the paper and pay the claims. AllDigital selects the risk under delegated authority.
Why does that count as vertical integration?
Because AllDigital owns appetite, pricing support, bind or decline, and the issuance workflow in one system trained on its own decisions. A broker-tech tool owns none of the decline authority.
How is that different from Harper?
Harper is producer of record. It finds a market. AllDigital is a market, within carrier authority. Harper does not auto-decline 30% of submissions. It has no mandate to.
How is that different from Crosby?
Crosby takes professional liability on the work product. AllDigital does not take insurance-company risk. Both own the decision layer of a regulated service. Crosby does not resell the firm as software.
What does 70% autonomous mean?
Of submissions that arrive, the company says 30 to 40% are bound by the system, about 30% are declined by the system, and about 30% go to a human. That is a chief executive figure from May 2026.
Did they raise like Harper or Crosby?
No. Public venture history is small-seed scale. The proof offered is premium per underwriter, not a $47 million headline.