Upgrade to Pro to get implementation-ready specs for every company, the full report library, and 5 on-demand report requests per month.
If you only have a few minutes to spare, here’s what investors, operators, and founders should know about GazeHawk (S10).
GazeHawk turned ordinary webcams into remote eye trackers. Founded in 2009 by Brian Krausz and Joe Gershenson, the two-person YC company let customers submit a webpage or image, recruit from GazeHawk's tester panel or use their own participants, and receive gaze heatmaps and individual tracks without a laboratory or dedicated hardware.[1][2]
The hardware savings were real, but they did not disappear. They moved into calibration failures, over-recruitment, participant bonuses, manual quality review, video upload, and server processing. At $49 per usable participant, GazeHawk was as much a research service as a software product.[3]
In March 2012, Facebook hired both founders. It explicitly did not acquire GazeHawk's product or technology, and the founders expected to work on unrelated product and backend engineering.[4] This was an acqui-hire, not a conventional product acquisition.
Krausz and Gershenson started GazeHawk in December 2009 from a simple objection: eye tracking should not be confined to expensive consultants and large companies. Y Combinator and 500 Startups backed the company, and YC placed it in the Summer 2010 computer-vision batch.[1][5] The evidence does not establish the founders' prior employers, schools, or how they met.
The company launched publicly in July 2010. Instead of shipping a physical tracker, it used webcams participants already owned and recruited people willing to complete studies at home.[6] Customers supplied a webpage URL or image. GazeHawk distributed the test to its panel or a customer-provided group, processed the resulting gaze data, and returned analysis in a dashboard, sometimes within five days.[2]
That design widened access in two ways. It removed specialized hardware, and it made geographically distributed samples possible. A marketing team could study a landing page, checkout, advertisement, print image, or brand asset without bringing participants into a lab.[2] The service sold usable observations rather than equipment.
The tradeoff appeared immediately. Participants opted in, granted webcam access, and calibrated by following a dot around the screen. The webcam light stayed visible through a self-moderated study.[3] Contemporary laptops could not process gaze video in real time, so GazeHawk streamed webcam footage to its servers.[7] Cheap sensors did not create cheap, clean data by themselves.
The observed source material contains no verbatim founder quotations, so this report does not invent the two quotes requested by the canonical format. It does show the founders publishing the core tension themselves: webcam tracking was less accurate than dedicated hardware, environmental conditions mattered, and a material share of participants failed quality checks.[6]
GazeHawk sold remote visual-attention research. A customer uploaded an image or submitted a webpage, selected either GazeHawk's panel or its own webcam-equipped participants, and waited for processed results. The service supported conversion analysis, landing pages, checkout flows, ad placement, brand recognition, web advertisements, print advertisements, and other static images.[2]
Each participant opted in to webcam use and completed calibration by following an on-screen dot. During the test, the system collected a dense stream of gaze coordinates. Analysts could view aggregate heatmaps, inspect individual tracks or videos, and explore clusters such as behavior in the first ten seconds.[3][8]
The core technical compromise was remote video processing. A later academic paper, drawing on communication with the founders, recorded that GazeHawk streamed webcam footage to its servers because laptops could not perform the inference in real time.[7] This broadened device support but added bandwidth, processing expense, privacy sensitivity, and latency.
Accuracy remained below laboratory hardware. In one MacBook Pro test, GazeHawk reported typical error below 70 pixels, slightly more than half an inch, compared with roughly 35 pixels for a Tobii T60. Poor lighting and excessive head movement degraded results.[6] These were founder-reported results rather than an independent benchmark.
GazeHawk's operations compensated for unreliable inputs. The company acknowledged that more than 10% of users could fail even with custom hardware and that its at-home failure rate was higher. It recruited extra participants, manually reviewed each result, excluded weak tracks without billing the customer, and paid bonuses for usable data.[6] The customer saw a clean per-usable-participant product because GazeHawk absorbed the mess behind it.
GazeHawk targeted marketers, designers, conversion specialists, and researchers who valued eye-tracking evidence but could not justify a lab. Its public use cases centered on landing pages, checkout flows, advertisements, and brand recognition.[2] The site highlighted Timothy, CEO of Graffiti Tracker, and press mentions from AdAge, Huffington Post, Xconomy, Conversion Rate Experts, and TechCrunch, but it did not publish audited customer or revenue totals.[10]
No reliable market-size, study-count, customer-count, or revenue figure appears in the evidence. GazeHawk did publish an analysis using the first three fixations from 500 people selected across its study database, which shows a multi-study participant corpus but not its commercial scale.[11] Any broader estimate would be invented.
GazeHawk positioned itself against two categories. Dedicated trackers such as the Tobii T60 delivered better accuracy but imposed equipment and lab costs. Passive mouse-tracking products such as ClickTale could observe all website users silently, but GazeHawk argued that cursor movement could not reliably reveal which element first caught someone's eye.[3]
EyeTrackShop was the closest strategic competitor. The Tobii spinout offered webcam-based ad-effectiveness research, raised $3 million in 2012, and named Google, AOL, Microsoft, P&G, and JCDecaux as customers.[9] That evidence matters because it shows the category did not vanish when GazeHawk's team left.
Webcam tracking later became commercially and technically viable. Tobii now offers a standard-webcam extension and reports 1.64-degree median accuracy in a university lab study.[12] RealEye reports average desktop accuracy around 106 pixels, with persistent uncertainty for small areas of interest, and says it calculates gaze locally while sending only coordinates to its servers.[13][14] WebGazer.js runs entirely in the browser without sending video to a server, though maintainers say updates are no longer guaranteed after February 2026.[15]
The implication is double-edged. Better devices and local inference validate GazeHawk's premise, but hardware-free tracking is now a commodity feature. A rebuild must win through participant quality, study design, privacy, interpretation, or workflow integration.
GazeHawk charged $49 for each usable participant.[3] That unit concealed a bundle: participant recruitment, opt-in and calibration, bonuses for quality, excess recruitment to replace failures, video transmission, server processing, manual review, exclusion of weak data, and dashboard delivery.
The model protected customers from bad tracks because they paid only for usable results. It also concentrated risk inside GazeHawk. Every failed calibration consumed some recruitment and processing expense without generating revenue. Every manual review tied gross margin to human labor. The proprietary model architecture, tester payout, server cost, and gross margin were not disclosed, so exact unit economics cannot be calculated.
Y Combinator and 500 Startups invested, but no reliable funding total was found.[5] There is likewise no public acquisition price because Facebook did not buy the product or technology. The only defensible description is a team hire with undisclosed terms.[4]
The public record supports usage, not scale. GazeHawk had enough completed studies to sample 500 people across its database for a fixation analysis.[11] It named Graffiti Tracker's CEO as a customer and collected meaningful press coverage.[10] The product could return some studies within five days.[2]
No audited customer count, paid-study count, repeat rate, revenue, or panel size was published. Facebook's interest establishes confidence in the founders' ability to build the technology and platform, not product-market fit. The founders themselves said Facebook had not acquired GazeHawk's product or technology.[4]
GazeHawk's thesis was directionally right: webcams could make eye tracking available outside labs. The error was not technical impossibility. It was treating removal of dedicated hardware as removal of system cost. Commodity cameras produced noisier data under uncontrolled lighting, head movement, device placement, and participant behavior.[6]
The company compensated through calibration, excess recruitment, bonuses, manual quality checks, free replacement of weak results, video upload, and server-side processing.[6][7] Those were sensible remedies. Together, they transformed a nominal software advantage into a service operation. The $49 usable-participant price charged for the cleaned output, while GazeHawk carried the variance.
This is the structural mechanism: hardware savings moved down the value chain rather than disappearing. A lab controlled camera, lighting, head position, and supervision before data collection. GazeHawk removed the lab, then rebuilt parts of that control through software prompts and human operations after collection. Scale increased the number of noisy sessions requiring judgment unless automation improved faster than volume.
The strongest counterargument is that GazeHawk was simply early. Later products from Tobii and RealEye, open-source WebGazer.js, and the 2025 WebEyeTrack prototype show that webcam gaze inference can run locally and reach useful speed.[12][14][15][16] WebEyeTrack reports fewer than nine calibration samples, 2.4-millisecond inference on an iPhone 14, and a 2.32-centimeter error margin on GazeCapture, while still identifying head movement and real-world accuracy as open challenges.[16]
That evidence confirms both halves of the GazeHawk story. Modern hardware and browser inference reduce server processing and video-transfer costs. They do not eliminate calibration, participant compliance, or ambiguity at small areas of interest. RealEye's published accuracy distribution makes the remaining uncertainty explicit.[13]
On March 8, 2012, Krausz and Gershenson announced they would join Facebook after it was impressed by their ability to build the eye-tracking technology and platform.[4] They said their likely work would involve product and backend engineering unrelated to GazeHawk. The product and technology remained independent while they sought options for it.[4]
Calling this a conventional acquisition would erase the most informative fact. Facebook valued the two-person team's demonstrated execution, but did not integrate the product. Later academic literature identifies GazeHawk as shut down.[7] No public buyer, licensee, or maintained successor for the product was found.
The source material includes the founders' announcement but no preserved verbatim quote in the evidence index, so none is manufactured here. The outcome nevertheless draws a clear line between talent value and company value. A technically impressive platform can become a recruiting credential even when its service economics and independent distribution remain unproven.
Ready to rebuild GazeHawk?
Implementation-ready specs, every report, and 5 on-demand requests each month.