# Parsewise

Aiolum story: https://aiolum.com/story/parsewise/ · added 2026-11-05 · updated 2026-11-06

**In short.** Parsewise (AI that turns insurance and finance document packages into checked, structured records, United Kingdom): Three named insurance and mortgage customers, November 2026. Evidence grade E: no revenue figure was checked. Team: 2 co-founders in 2024; 6 people in November 2026 after Y Combinator funding. Launched 14 May 2025; still running.

AI that turns insurance and finance document packages into checked, structured records · United Kingdom · AI-native · team of 2 to 10 · still running · group: Finance, risk and compliance

- Outcome: Three named insurance and mortgage customers, November 2026 (evidence grade of the headline number: E; no revenue figure)
- Launch: 14 May 2025; months from launch to the outcome: 18
- Team: 2 co-founders in 2024; 6 people in November 2026 after Y Combinator funding
- Outside funding: Y Combinator, Spring 2025 batch; round size not verified
- Main factor: Domain experience
- Meters, 1 to 4: evidence strength 1, copyable pattern 1, capital needed 2
- Tags: AI-native, Small team, Funded, Running, B2B, Insurance

Links:

- Visit parsewise.ai: https://www.parsewise.ai/

**The product.** Software for commercial insurers, brokers and investment teams. It reads a whole package of documents (policies, claim reports, loss runs, bordereaux, emails, spreadsheets), fills a fixed set of fields, flags values that contradict each other, and links every value back to the page it came from. It is sold as yearly contracts with an assistant called Navi, and as a pay-per-page API for developers.

**Why it worked.** The two founders had seen the same problem from two sides: one at Palantir building document pipelines for insurance clients, one at Bain doing due diligence by hand. They built a layer on top of large language models that checks the whole document set and shows the source of each answer, which regulated buyers need before they trust the output.

Evidence grades: A = disclosed by a public company, a filing or a payment processor, or verified by one; B = two or more independent reports agree, or a journalist confirmed it; C = a founder claim that no one else verified; D = reported, not verified: one founder claim or a third-party estimate with no second source; E = no revenue figure found, so the headline is a non-money fact and the other facts carry their own grades; calc = computed on this page from graded facts.

## Summary

Parsewise is a London company that does multi-document extraction: a program reads many documents about one case at once, takes out the values a team needs (a reserve, a deductible, a flood limit), and compares them across the documents. Each value keeps a source trace, a link to the page and the box on that page where it was found, so an underwriter or claims handler can check it. Its customers are commercial insurers, MGAs (managing general agents, firms that underwrite on behalf of an insurer), brokers and investment teams.

Max Hofer and Greg Csegzi met studying computer science at Oxford in 2017, the company's Y Combinator page and launch post say. They founded Parsewise in 2024, joined Y Combinator's Spring 2025 batch, and posted the launch on 14 May 2025, then "live with pilot customers". In November 2025 the company was runner-up at the Slush 100 pitch competition, where a judge's write-up reported "live revenue" without a figure. In November 2026 the Y Combinator page lists 6 people, and the company's site names three customers: Compre, Thinksurance and HypoHaus.

No revenue figure was found in any source, so this story is grade E: the headline is a count of named customers, not money. The funding amount is also not verified: three startup directories give three different amounts.

## Pricing

Parsewise prices read on 5 November 2026 from the company's pricing page and its API page (grade C)

| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| API, Free / pay as you go | 0 | 0 | US$50 free credits, then US$0.005 a page to parse and US$0.005 a page per field to extract; up to 2 UI seats |
| API, Enterprise | custom | custom | volume discounts, as many agents and users as needed |
| Growth | quote, one-year contract |  | Workflows seats, up to 500 pages per run, three weeks of onboarding |
| Business | quote, one-year contract |  | adds Analytics credits and single sign-on |
| Enterprise | quote, one-year contract |  | adds dedicated engineering time and a private cloud option |

The three platform plans have no list price: each is a one-year contract scoped to the team and its expected usage, plus a one-time onboarding fee. Analytics usage is charged in credits by page count and extraction complexity, with extra usage billed automatically. The API is in US dollars and charged per page: parsing costs US$0.005 a page, and extraction costs US$0.005 for each page times each field. New API accounts get US$50 of free credits, which the company says is enough for about 1,600 pages with 5 agents.

## How much revenue does Parsewise make? The numbers, graded

A two-founder Y Combinator company with named enterprise customers and no published revenue figure

| Fact | Value | Grade | Note | Sources |
|---|---|---|---|---|
| Named customers, November 2026 | Compre, Thinksurance and HypoHaus | C | The company's customer-stories page describes how each one uses Parsewise. The Y Combinator launch post of May 2026 also names UBS; the company's own customer page does not. | [7, 5] |
| Revenue | no figure found | C | A Slush 100 judge's write-up of 20 November 2025 says: "Live revenue, daily usage inside financial institutions, and strong retention." No amount is given anywhere. | [6] |
| Founders | Max Hofer (CEO) and Greg Csegzi (CTO) | B | The UK company register lists the two as the only directors of Parsewise Ltd, both appointed 28 August 2025. Y Combinator lists them as the founders. | [10, 5] |
| Founded | 2024 | C | Y Combinator's company page. A first Parsewise Ltd was incorporated on 4 December 2024 and dissolved on 27 May 2025; the current Parsewise Ltd was incorporated on 28 August 2025. | [5, 10] |
| Team, November 2026 | 6 people | C | Y Combinator page, team size 6. The company's about page lists six people: the two co-founders, two founding engineers, a founding go-to-market lead and a forward-deployed engineer. The blog also has posts by one more author. | [5, 4, 9] |
| Funding | Y Combinator, Spring 2025 batch | C | Y Combinator page. The round size is not verified: VCBacked says US$1.6M pre-seed in April 2025, FundedIQ says US$500K pre-seed in April 2025, and StartupHub says US$2M seed. No company or investor announcement was found. | [5, 11, 12, 13] |
| Pitch result | Slush 100 runner-up, November 2025 | B | UpNext World's recap of the Slush 100 finals, 20 November 2025. | [6] |
| API price | US$0.005 a page to parse, US$0.005 a page per field to extract | C | API page, read 5 November 2026. US$50 of free credits for a new account. | [3] |
| Platform plans | Growth, Business, Enterprise; one-year contracts, no list price | C | Pricing page, read 5 November 2026. Onboarding takes three weeks on Growth and Business, two months on Enterprise. | [2] |
| AI models used | Google Gemini 3 Flash and Gemini 3.5 Flash | C | Company blog on the OfficeQA benchmark, 30 June 2026: Gemini 3 Flash for text extraction, Gemini 3.5 Flash for reasoning in Navi. Google Cloud's customer story on Parsewise describes the same setup. | [8, 15] |
| Benchmark claim | 58.65% correct on Databricks OfficeQA, against 57.90% for the best frontier model | C | Company blog, 30 June 2026. The company also reports a "revised" score that counts 15 of 133 questions where it says its answer is better than the official key. | [8] |
| Security and hosting | SOC 2 Type II, GDPR, hosting and model inference in Europe | C | Home page, 5 November 2026. | [1] |
| Time from launch post to the customer count | 18 months | calc | Calculated: May 2025 to November 2026. | [5, 7] |

How to read this table: a grade C value is the company's own statement (its site, its Y Combinator launch posts) or a third-party directory. A grade B value comes from an independent record: the UK company register or a judge's write-up. A "calc" value was computed on this page. No value is grade A.

## When did Parsewise launch, and how did it grow? Timeline

From a 2024 founding to a six-person team selling to insurers, in about two years. Each entry carries its grade.

- **2017** (grade C [5]): Max Hofer and Greg Csegzi meet studying computer science at Oxford. Hofer later works at Bain and does a PhD; Csegzi builds AI use cases at Palantir, including for insurance.
- **2024** (grade B [5, 10]): Parsewise is founded. A first Parsewise Ltd is registered in London on 4 December 2024.
- **14 May 2025** (grade C [5]): Launch post on Y Combinator, Spring 2025 batch: extraction from complex documents with a source for every value, "live with pilot customers".
- **28 August 2025** (grade B [10]): The current Parsewise Ltd is incorporated, with the two founders as directors.
- **20 November 2025** (grade B [6]): Runner-up at the Slush 100 finals. The judge's write-up reports live revenue and daily use inside financial institutions, with no figure.
- **23 March 2026** (grade C [9]): Navi, a chat assistant that sets up extraction agents from plain-language questions, is announced.
- **26 May 2026** (grade C [5, 3]): Second launch post: a pay-per-page API for multi-document processing.
- **November 2026** (grade C [5, 1, 7]): Six people. The site is aimed at commercial insurance and names Compre, Thinksurance and HypoHaus as customers.

## Why did Parsewise work?

### Domain experience

The CTO had built document pipelines for insurance clients at Palantir; the CEO had done more than 20 private-equity due diligence projects at Bain, the launch post says. They knew that the buyers' problem was trust in each value, not reading speed.

### A source for every value

Each value links to its page and position, and values that disagree across documents are flagged. A regulated insurer can then check and defend a decision, which a plain chat answer does not allow.

### Read everything, not a sample

The product reads the whole document set, so a bordereau is checked row by row. The company argues that sampling misses the errors that cost money.

### One narrow market

The May 2025 launch post named investment management, reinsurance and life sciences; the 2026 site speaks to commercial insurance. Pages per workflow (claims, underwriting, bordereaux) match the work of one kind of buyer.

## How it was built

These details come from the company's own site, blog and Y Combinator launch posts. They are grade C.

- Google Gemini models inside the product: Gemini 3 Flash for text extraction and Gemini 3.5 Flash for reasoning, per the June 2026 benchmark post.
- A data engine that the company says can parse thousands of documents per query and run millions of model calls, with agents that each extract one topic.
- Hosting and model inference in Europe, or inside a customer's own cloud on AWS, Azure or Google Cloud.
- SOC 2 Type II and GDPR compliance, which enterprise insurance buyers ask for.
- Each customer gets a configured setup: the team agrees fields, rules and checks, and tests on real cases during a three-week or two-month onboarding.
- An API with free credits and per-page prices, launched in May 2026 for developers who want to call the engine from their own product.

Sources: [8, 9, 1, 2, 3]

## Who tried this before Parsewise, and why did they fail?

Three companies in our database of failed startups promised to take the paperwork out of a document-heavy financial process. The values below are from that database and are not independently verified.

The failed startups come from our database of dead startups, retold here in our own words; Its entries are AI-assisted summaries of public sources, so they carry no evidence grade.

### Lendsnap (2016 to 2019; funding US$500K; cause: Competition)

- Promise: Automatic collection of borrowers' financial documents for mortgage lenders.
- Why it died: Broader mortgage platforms such as Blend and Roostify offered the same step inside an end-to-end product, and it could not win enterprise partners.
- Record: 

### Opkit (2021 to 2024; funding US$1.5M; cause: Ran Out of Cash)

- Promise: Automated health insurance verification and claims processing.
- Why it died: Hospitals were slow to adopt new software, integration was hard, and Epic offered bundled services. It failed to secure long-term contracts.
- Record: 

### Canvas (2011 to 2017; funding US$17M; cause: Competition)

- Promise: Turn paper forms and documents into digital workflows for businesses.
- Why it died: Microsoft and Adobe added form and document features to products customers already had.
- Record: 

## What changed

The three predecessors automated one step of a paper process and were overtaken when larger platforms put the same step inside a broader product. The same risk exists here: the model makers and the big document-AI services sell extraction too.

What differs is that large language models can now read documents of any layout without a template, so one small team can cover policies, claims and loss runs together. Parsewise sells the part the models do not provide: cross-document checks, a source for each value, and setup with each customer's own rules.

## Can one person copy Parsewise now?

**Verdict: Hard to copy alone.**

The product needs enterprise sales, security certification and onboarding work with each customer, and the founders brought years of insurance and due-diligence experience. A solo founder could copy the narrow version: a per-page API for one document type in one industry, with a source trace for each value.

1. Pick one document-heavy job in one industry that you know from your own work, for example bordereau checks or mortgage application review.
2. Return every value with the page it came from, and flag values that disagree across documents.
3. Offer a self-serve API with free credits and a per-page price, so developers can test it without a sales call.
4. Get the security certification your buyers ask for before you sell to large companies.

**What is still open.** Other regulated document work with the same shape: trade finance, construction claims, clinical trial files, public tenders.

## What could go wrong for Parsewise?

- **What is not verified.** No revenue figure was found. The only statement about money is a Slush 100 judge's write-up of 20 November 2025: "Live revenue, daily usage inside financial institutions, and strong retention", without an amount. The funding amount is also unclear: VCBacked says US$1.6M, FundedIQ says US$500K and StartupHub says US$2M, and no announcement by the company or an investor was found. The UK register shows no share-allotment filing for the current company. A reader could look for a funding announcement, a filed set of accounts (the first are due by 28 May 2027) or a founder interview with figures.
- **Team size and funding.** Two co-founders started it in 2024. Y Combinator lists 6 people in November 2026 and the about page shows the same six; the blog has posts by one more author. It has outside funding from Y Combinator.
- **Trust signals.** The May 2025 launch post quotes two customers by job title only ("Head of Family Office", "Lead Data Analyst at a Large Asset Manager"). The May 2026 post names UBS as a customer, which the company's own customer page does not list. The benchmark lead rests on the company's own test, and its higher "revised" score counts answers it judged better than the official key.
- **Depends on Google's models.** The June 2026 benchmark post says all its results use Gemini 3 Flash and Gemini 3.5 Flash. Google Cloud's own customer story on Parsewise confirms the setup. A price or terms change from Google would reach the product directly, although the post says the company wants to try larger models as well.
- **Competition.** Document extraction is a crowded market: the model makers, cloud document services and many startups sell it, and on this site PDF Vector sells a document parsing API to developers, while Parsewise sells mainly year-long contracts to insurers.

## Sources

1. Parsewise home page, read 5 November 2026 — https://www.parsewise.ai/ (grade C). Read in full. Source for the product, the target buyers, the security claims and the European hosting.
2. Parsewise pricing page, read 5 November 2026 — https://www.parsewise.ai/pricing (grade C). Read in full. Source for the three contract plans, the one-year term and the onboarding fee.
3. Parsewise API page, read 5 November 2026 — https://www.parsewise.ai/api (grade C). Read in full. Source for the per-page API prices, the free credits and the deployment options.
4. About Us, Parsewise, read 5 November 2026 — https://www.parsewise.ai/about-us (grade C). Read in full. Source for the six people listed on the team.
5. Y Combinator company page: Parsewise, Spring 2025 — https://www.ycombinator.com/companies/parsewise (grade C). Read in full, including the page data with the two launch posts of 14 May 2025 and 26 May 2026. Source for the founders, founding year, batch, team size and customers named by the founders.
6. UpNext World, 20 November 2025: Inside Slush 100: Diffraqtion Wins It All, Plus the Finals and Standout Founders — https://upnext.world/inside-slush-100-diffraqtion-wins-it-all-plus-the-finals-and-standout-founders/ (grade B). Read the section on Parsewise. Source for the runner-up place and the traction statement without a figure.
7. How teams put Parsewise to work, customer stories, read 5 November 2026 — https://www.parsewise.ai/resources/customer-stories (grade C). Read in full. Source for the three named customers and how each uses the product.
8. Parsewise blog, 30 June 2026: SOTA results on Databricks OfficeQA — https://www.parsewise.ai/blog/officeqa-sota (grade C). Read in full. Source for the benchmark claim and the Gemini models used.
9. Parsewise blog index and the post of 23 March 2026: Navi: A New Interface to Parsewise — https://www.parsewise.ai/blog (grade C). Read the index and the Navi post in full. Source for the post dates, the authors and the data engine description.
10. UK Companies House: PARSEWISE LTD, 16677569 (overview, officers, filing history) and 16117326 (dissolved) — https://find-and-update.company-information.service.gov.uk/company/16677569 (grade B). Read the overview, officers and filing history pages of both companies. Source for the directors, incorporation dates and the absence of share-allotment filings.
11. VCBacked: Parsewise Funding & Investors — https://www.vcbacked.co/company/parsewise (grade D). Read in full. A directory; its round size disagrees with the two other directories.
12. FundedIQ: Parsewise: Funding, Investors & Team (Nov 2026) — https://fundediq.co/parsewise-parsewise-ai-funding/ (grade D). Read in full. A directory; it also gives the wrong city (San Francisco).
13. StartupHub.ai: Parsewise — https://www.startuphub.ai/startups/parsewise (grade D). Read in full. A directory; it says Parsewise was founded in 2025, which disagrees with Y Combinator's page.
14. Google Cloud customer story: Parsewise, read 5 November 2026 — https://cloud.google.com/customers/parsewise (grade A). Read in full. The model vendor's own account of how Parsewise uses Gemini models and Google Cloud.
15. Database of failed startups, local export — /method/#dead (grade C). Local export. AI-written summaries, not independently verified. Source for the failed predecessors.

Contact: aiolum@livyden.com. Run Parsewise? Send corrections or a verified figure to aiolum@livyden.com (subject: Correction: Parsewise).