ATTEX
A head of data reading on a tablet in an office at dusk
CompareGround truth data next to attention scores

How ATTEX compares.

At scale, the attention vendors sell a score: a model’s prediction for impressions no camera saw. ATTEX licenses the data a score is checked against, the measured record of each ad on a real screen. A different kind of product, and a smaller, Nordic one.

§ 01A score, or the record

A modelled score, or the measured record.

A score · the attention vendors

A small gaze panel trains a model; the model scores every impression from what an ad server can log.

  • What you get: a number per impression or placement: attentive seconds (APM), a 0 to 100 score (AU), an attention index.
  • Typically knows the placement: format, size, position, time in view, domain, device.
  • For a scored impression, cannot say whether this person looked at this ad; no camera was present.
  • Inherits the reach of its training panel: a model trained in the UK and the US and applied to a Nordic in-app feed is an extrapolation.

Lumen LAMP, Adelaide AU, Amplified’s tag, GumGum, IAS; DV’s core product has no gaze at all. Lumen also sells direct studies on its panel. [1] [3] [6] [10] [12] [13]

The record · ATTEX

Paid panelists on their own phones and computers; Tobii’s eye tracking, through the camera, estimates gaze on the ad that was on screen.

  • What you get: the data itself. One row per ad: the audience (age, gender and region, a code and never a name), the app or site and the format, how long it was on screen, and whether and how long it was looked at.
  • What you do with it: plan and buy against it, or train, validate and audit your own models of advertising performance. The score is yours to build.
  • October 2026: 1,080,225 mobile impressions from 274 panelists, 48% with the eye tracker running; 38% of those looked at, for 1.24 s when they were.
  • An instrument with its own unpublished error: gaze calibration varies by device and light, and mobile and desktop use different fixation detectors. “Ground truth” means taken on the person, not “true”.

Two products built on the record also predict, and label it: PubMatic deal segments with fewer than 50 panel impressions behind them, and campaign-report rows with no panel data behind them. [19] [21] [22] [23]

§ 02Ten questions, six vendors

The same questions, put to each.

  • yes
  • partly
  • no
  • not found
ATTEXLumenAdelaideAmplifiedGumGumDV / IAS
Do you get the data itself, or a score?The record: one row per ad, licensed as data [20]A score at scale: attentive seconds per 1,000 viewable (APM); panel data in direct studies [1] [3]A score: 0 to 100 per placement (AU) [6]A score: seconds by kind; attentive reach [10]A score: seconds per impression [12]A score: attention indices [13] [14]
Is the number measured on the person?Eye tracking through the camera on every eye-tracked impression [19] [21]On its panel; LAMP scores impressions by model [1] [3]AU is a model score per placement [6]In sessions; the tag and AI predict [10] [11]A model scores live impressions [12]Tag signals (DV); Lumen-trained model (IAS) [13] [14]
Gaze, not only time in view?Whether and how long the eyes rested on the ad [20]Webcam on the panel, not at scale [1]Licensed gaze trains the model [6]Facial footage in sessions [10]Predicted seconds [12]No gaze (DV); Lumen’s (IAS) [13] [14]
Inside the apps, with the platform's own ads?Instagram, TikTok, Facebook, Snapchat, YouTube [19]Mobile method not confirmed [1] [5]In-app via licensed data [6]Sessions with inserted test ads [10]Own browser, mobile web [12]Snapchat and TikTok via Lumen [13]
The ad's context: app, format, hour?In every row; aggregates through the assistant [20] [26]Page context for the model onlyA placement score [6]Session metadata only [10]Browsing in the panel browser [12]Nothing beyond the impression
A Nordic panel?Sweden, Norway, Denmark and FinlandUK and US panels [1] [2]US, UK, Australia; no panel [6] [7]Australia, UK, US [11]Panel markets undisclosedGlobal, tag-based
Panel size, where statedAbout 600 active with gaze, Oct 2026 [21]About 2,000, UK and US [1] [2]No standing panel [6]Not statedUndisclosed [12]None (DV); Lumen’s (IAS) [13] [14]
Buyable at bid time?Deal IDs in PubMatic, Swedish display [25]DSPs, PMPs, IAS and DV, Netflix, Teads [5]Pre-bid in DV360, Amazon, The Trade Desk [7]API, tag, self-serve [11]Inside GumGum’s platform [12]Pre-bid in both [13] [14]
TV and CTV?NoneVia TVision [5]Via TVision [7]TV in sessions [10]None foundCTV via tag (DV) [13]
Independently audited?Not yet; instrument accuracy unpublishedPwC review 2023, commissioned [4] [9]MediaSense 2025; MRC in process [8] [9]None foundKantar check 2019, reportedDV MRC-accredited 2023; IAS none found [13]

Numbers in brackets are the sources in § 04. The vendor pages were read as search snippets on 9 November 2026, not fetched, and may have changed; where a point could not be confirmed the cell says so. TVision, the camera-based television panel behind Lumen’s and Adelaide’s TV figures, has no column because ATTEX does not measure television. [15]

  1. Lumen

    A UK and US eye-tracking panel; sold at scale as LAMP, a model that scores impressions, and as direct studies. [1] [3]

  2. Adelaide

    AU, a 0 to 100 placement score trained on licensed gaze and outcome data; native in the big DSPs; no panel of its own. [6] [7]

  3. Amplified Intelligence

    Session panels with inserted test ads in Australia, the UK and the US; a tag and an AI product predict from them. [10] [11]

  4. GumGum Attention Time

    An opt-in mobile browser panel of undisclosed size; a model turns it into seconds per impression inside GumGum’s platform. [12]

  5. DoubleVerify and IAS

    Tag-based indices at global scale; DV’s core product has no gaze, IAS and DV’s social product are built on Lumen’s panel. [13] [14]

§ 03The IAB/MRC guidelines

What the guidelines ask, and where ATTEX stands.

The IAB/MRC Attention Measurement Guidelines (version 1.0, November 2025, voluntary) ask that deterministic measurement be told apart from modelled, that training data be collected deterministically and validated every year, and that measurement reflect real use rather than simulated feeds. [18]

ATTEX measures deterministically, in real feeds, and is not a named contributor to the guidelines. They apply to its instrument as much as to anyone’s model: section 6.1 asks for the accuracy of the visual tracking itself, which ATTEX has not yet published. Lumen, Adelaide, Amplified, GumGum, DoubleVerify and IAS are among the contributors.

§ 04Sources

The vendor pages and most of the trade press were read as search snippets on 9 November 2026, not fetched, and may have changed since. Two documents were read in full: the Amplified Intelligence methodology deck and the IAB/MRC guidelines. Internal ATTEX material is marked as such and is available on request. The numbering follows the analysis of 9 November 2026; its [17] and [27] are not cited here.

  1. [1]Lumen Research, eye-tracking panel and media buying pages: lumen-research.com/eye-tracking-panel; lumen-research.com/media-buying (as quoted in search results, November 2026; the site was not fetched directly; page dates unknown).
  2. [2]Finnpanel, TV-vuosi 2019, Mike Follett deck: finnpanel.fi
  3. [3]IAB UK directory, LAMP: iabuk.com
  4. [4]PwC review of Lumen, 27 April 2023: advanced-television.com; lumen-research.com (secondary reports; the PwC document itself was not read).
  5. [5]Lumen partnerships. Netflix, 5 March 2026: exchangewire.com; Teads CTV, 6 May 2026: exchangewire.com and help.teads.com; Cint, 16 April 2026: cint.com; IAS Quality Attention: integralads.com; The Trade Desk billing: lumen-research.com
  6. [6]Adelaide: adelaidemetrics.com/blog/how-we-use-data-at-adelaide; adelaidemetrics.com/lp/why-adelaide-au; adelaidemetrics.com/blog/attention-metrics-vs-viewability-vs-duration; adelaidemetrics.com/privacy-au-ratings (wording from search snippets; the site was not fetched).
  7. [7]Adelaide distribution. Amazon DSP, 10 June 2026: adexchanger.com; DV360: ppc.land; Index Exchange, 7 January 2026: indexexchange.com; Teads AU floor: ppc.land
  8. [8]MediaSense review of Adelaide, 7 July 2025: exchangewire.com
  9. [9]MRC accreditation process: digiday.com; adexchanger.com; ppc.land
  10. [10]Amplified Intelligence, attentionTRACE and attentionPROVE methodology deck (2022 / February 2023), read in full: arf-research-dox.s3.amazonaws.com (current method not verified).
  11. [11]Amplified Intelligence, AttentionAI, 20 May 2026: advanced-television.com; attentionPLAN API: developers.amplified.co
  12. [12]GumGum and Playground xyz: gumgum.com; medium.com/gumgum-tech
  13. [13]DoubleVerify MRC accreditation, January 2023: doubleverify.com; no-camera statement: adexchanger.com
  14. [14]IAS Quality Attention explainer: ppc.land
  15. [15]TVision methodology: tvisioninsights.com
  16. [16]YOC on APM: yoc.com
  17. [18]IAB/MRC Attention Measurement Guidelines, final version 1.0, November 2025: iab.com
  18. [19]ATTEX methodology and panel: attexdata.ai/docs/methodology; panel rewards and terms on attexpanel.com
  19. [20]ATTEX definitions: attexdata.ai/docs; attexdata.ai/docs/data-dictionary
  20. [21]October 2026 figures from the ATTEX panel (ATTEX MCP server, query_impressions, Sweden, Norway, Denmark and Finland, 1 to 31 October 2026, run bd6c1c01), ATTEX, available on request; the active-panelist counts from the ATTEX board deck of 4 November 2026, ATTEX, internal.
  21. [22]PubMatic attention model, Sweden, October 2026. ATTEX, internal, available on request.
  22. [23]Campaign report method and rules. ATTEX, internal, available on request.
  23. [24]ATTEX Privacy Notice for panelists, last updated 12 November 2026: attexpanel.com/panel-privacy-policy
  24. [25]Delivery: attexdata.ai/docs/connect; attexdata.ai/docs/integration; the PubMatic deals: attexdata.ai, Products (the Swedish display scope: ATTEX, internal, available on request)
  25. [26]attexdata.ai/docs, “Inside the apps” and “One table”
  26. [28]Customer results and the independent study, as published on this site: Schibsted Attention Study 2023, supervised by Erik Modig, Stockholm School of Economics: PDF
§ 05Contact
Emelie Löfdahl

Questions on method, or the analysis behind this page, go to Emelie Löfdahl, Co-founder and CEO: emelie@attexdata.ai. She replies within one business day.

Talk to Emelie

Our own figures: the Docs.