Industries / Finance and business services / Finance and banking / Private equity and venture capital
Finding deals
Finding deals is the work of spotting private companies worth backing, or of getting the chance to back them, before other investors do.
7
Milestones achieved, from 1957 to 2019.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1957
Forty-one refusals, one backer
Eight scientists who wanted to leave William Shockley's laboratory wrote to the New York investment firm Hayden, Stone, and the letter reached a young banker, Arthur Rock. Rock took their plan to 41 companies, among them Motorola, General Mills, Chrysler and Westinghouse, asking for $1.5 million, and none would invest until Sherman Fairchild of Fairchild Camera and Instrument agreed; the eight founded Fairchild Semiconductor in 1957, by Rock's account to NPR. Rock formed the venture firm Davis & Rock in San Francisco in 1961, and in 1972 Fairchild alumni started Kleiner Perkins and Sequoia Capital, by the Computer History Museum's account.
LessonLarge companies would not fund a breakaway team, in Rock's view for fear of unsettling their own staff, which left the opening to independent investors.
Credit
Arthur Rock; Sherman Fairchild; the eight founders
2005
Startups funded in batches
Paul Graham, Jessica Livingston, Robert Morris and Trevor Blackwell started Y Combinator in March 2005 with $200,000 of their own money and first funded a group of startups together in a summer program for students. Graham writes that they funded a whole group at once to learn angel investing quickly, then kept the model because it worked: "We'd fund two batches of startups a year." Programs of this kind, which invest in a cohort of startups for a fixed term and end in a public demo day, are now called accelerators; a 2014 research summary for the US Small Business Administration (SBA) named Y Combinator among the most successful and said accelerators had "sprung up all over the world". Y Combinator now runs four batches a year and invests $500,000 per company, by its own site.
Credit
Paul Graham, Jessica Livingston, Robert Morris, Trevor Blackwell
2013
One lead, many followers
On 28 March 2013 the staff of the US Securities and Exchange Commission (SEC) told AngelList it would not take enforcement action if AngelList let an experienced "lead angel" bring a company to its website and let other accredited investors, people who pass the SEC's income or wealth tests, invest alongside through a fund set up for that one company. Neither AngelList nor the lead could take a commission; both were to be paid with carried interest, a share of any gain on the investment. The letter let one investor's find reach many co-investors online without either party registering as a broker.
Credit
AngelList; the SEC Division of Trading and Markets
2013
Advertising allowed, rarely used
Regulation D is the set of SEC rules under which companies sell shares privately without registering them, filing a short notice called Form D. From 23 September 2013 the new Rule 506(c), required by the JOBS Act of 2012, let companies advertise such offerings openly as long as every buyer is checked as an accredited investor, ending "almost 80 years of regulatory practice", in the words of the SEC's economists. Through the end of 2017 there were 7,110 new offerings under the new rule against 87,890 under the old one, which bars advertising, and only 4% of the money raised under Regulation D, about $255 billion, came through Rule 506(c), by the count of the SEC's Division of Economic and Risk Analysis.
LessonLifting the ban on advertising left most private money moving through existing channels; the SEC's economists suggest that companies with investors already in reach saw no need to advertise.
Credit
the SEC, for the rule and the count
2016
Most deals come through contacts
In a survey of 885 venture capitalists at 681 firms, the average firm screened 200 companies to make four investments a year; over 30% of deals came through professional networks, 20% were referred by other investors and 8% by companies already owned, almost 30% were sought out by the investors themselves, and only 10% arrived unasked from founders. An earlier survey of 79 buyout firms managing over $750 billion found that they closed about 6 of every 100 deals they considered; 33% of closed deals came from investment banks and so were unlikely to be proprietary, meaning not shown to rival bidders, while the firms called almost 48% of their deals proprietary in some way, a claim the authors say they could not check.
Credit
Paul Gompers, William Gornall, Steven Kaplan and Ilya Strebulaev (venture, 2016); Paul Gompers, Steven Kaplan and Vladimir Mukharlyamov (buyouts, 2015)
2016
Software scans for startups
EQT Ventures launched Motherbrain in 2016 to sort through large numbers of technology startups. Its dealmakers see companies picked up from their digital traces and ranked by scoring models that show the reasoning behind each rank, with network data added, such as which EQT staff know a company and which advisers have met its team, by an account in the European technology outlet Tech.eu in November 2025.
Credit
EQT Ventures
2019
Less than 1p per pound
A report by the British Business Bank, the UK government's economic development bank, with Diversity VC and the British Private Equity & Venture Capital Association, commissioned by the Treasury in 2017, used PitchBook data on about 7,500 UK venture deals from January 2007 to April 2018. It found that for every £1 of UK venture capital, all-female founding teams got less than 1p, all-male teams 89p and mixed teams 10p. The bank's 2023 follow-up found all-female teams won 9% of equity deals in 2022 but 2% of the money, no better than in 2011, and that 65% of venture firms surveyed did not search any differently when trying to reach under-represented founders.
LessonWhen most deals reach investors through their own networks, the mix of who gets funded changes slowly.
british-business-bank.co.ukalternativecreditinvestor.combritish-business-bank.co.uk
Credit
the British Business Bank, with data from PitchBook
Watch for
Models score founder profiles
VCBench, a benchmark first posted in September 2025 by researchers at the University of Oxford and Vela Research, holds 9,000 anonymized founder profiles, 810 of them labeled successful: acquired or listed at a valuation over $500 million, or having raised more than $500 million. Precision here is the share of a model's picks that turned out successful. The authors report that several large language models beat the precision of top-tier venture firms on it, with GPT-4o at 29%. They caution that the benchmark's success rate of 9% is far above the real 1.9%, and that real investors choose only among deals they can reach, so the comparison may not hold in practice.
Achieved whenAn outside team reproduces the result, or models' calls made before outcomes were known are scored by an independent party once the outcomes are in.
Claimed by
Rick Chen, Yigit Ihlamur and co-authors, September 2025, revised May 2026
Open
01
A register of accelerator terms
A public, dated table of startup accelerators worldwide that records, for each program, what it gives and what it takes: cash invested, the equity or instrument taken (such as a SAFE, a simple agreement for future equity that turns into shares later), program length, cohort size, city, sector focus and demo-day date. Each field links to the program page it came from, and the table is re-read every quarter so that changes in terms show over time. It records published terms only and does not rate programs.
The briefCredit · unclaimed02
From grant to first round
A public link between US research grants to small firms and those firms' later private fundraising. Match each company in the award data of SBIR and STTR (Small Business Innovation Research and Small Business Technology Transfer, federal programs that pay small firms to do research) to its Form D filings, and publish, by agency, technology area and award year, the share of grant winners that later reported a private raise and the median time from award to first filing. It reports public records and rates, not a list of companies to approach.
The briefCredit · unclaimed
Open 01 · the brief
A register of accelerator terms
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, dated table of startup accelerators worldwide that records, for each program, what it gives and what it takes: cash invested, the equity or instrument taken (such as a SAFE, a simple agreement for future equity that turns into shares later), program length, cohort size, city, sector focus and demo-day date. Each field links to the program page it came from, and the table is re-read every quarter so that changes in terms show over time. It records published terms only and does not rate programs.
Why it is still open
Seed-DB, which a 2014 SBA research summary called "the best-known and most widely used database" on accelerators, was maintained only until about 2019 and has been shut down; its founder now points users to commercial databases. Even in 2014 almost 40% of its entries lacked portfolio size or value. The largest research database, the Global Accelerator Learning Initiative's, started in 2012 and covers more than 23,000 businesses that applied to more than 369 programs, but it follows the businesses' revenue, staff and money raised, not each program's terms, by the account of its funder Porticus. The public lists found on GitHub and Hugging Face in November 2026 give names, websites, stages and sectors, but not terms.
Built with what exists
A web agent such as Browser Use 1.0 to collect program pages (see the Note "Deleting 80% to start over"); Claude Fable 5.2 to pull terms into fixed fields (see the Note "Three working devices"); Python; a public Git repository, so each quarter's changes are a dated record.
First step
Fill the table by hand for 30 programs on three continents to settle the fields, then run the agent on the same 30 and score it against the hand version.
Done looks like
Terms for at least 500 programs in at least 40 countries, every field linked to its source page; a second person checking 50 random programs agrees on at least 95% of fields; four quarterly snapshots published.
Open 02 · the brief
From grant to first round
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public link between US research grants to small firms and those firms' later private fundraising. Match each company in the award data of SBIR and STTR (Small Business Innovation Research and Small Business Technology Transfer, federal programs that pay small firms to do research) to its Form D filings, and publish, by agency, technology area and award year, the share of grant winners that later reported a private raise and the median time from award to first filing. It reports public records and rates, not a list of companies to approach.
Why it is still open
Partial links exist but none is public. In August 2026 the Center for Strategic and International Studies (CSIS) used Form D filings to look for private raises among the 10,280 firms that have won Defense Department SBIR awards since 1983, without releasing the matched data. A 2026 National Academies review of the Defense Department's programs linked 5,919 award winners from 2012 to 2020 to private financing, but through Crunchbase, a commercial database of announced deals, and found 18% had reported a round. No join of all agencies' awards to Form D was found on GitHub or Hugging Face in November 2026. Form D also undercounts, because filing it "is not a condition" of using the exemption, by the SEC's own economists, so any match rate is a floor.
Built with what exists
SBIR.gov's bulk award file; the SEC's quarterly Form D data sets; Splink, the open-source record-linkage library from the UK Ministry of Justice's analysts, for matching names and addresses; Gemini Decision 1 to judge borderline matches (see the Note "A thousand questions, one pass"); Python.
First step
Take one agency's 2015 award winners, match them to Form D by hand, and measure how often names and addresses line up.
Done looks like
Every SBIR and STTR award winner from 2009 to 2024, across all agencies, checked against Form D, with the match table, rules and code public; a second person checking 200 random matches and non-matches agrees on at least 95%; rates published by agency and award year, with the Defense Department's set against the National Academies' 18%.