In the guide I’ll show how to use eye-tracking for UX and CRO, how to link the results to KPIs, and how to build a research strategy useful for product teams, marketing, and management.
Visual attention metrics for business
For sites that use PrivatBank/Monobank payments for submissions, priority is a fast TFF to the CTA and high dwell on price/guarantees.
Practical thresholds depend on screen density and the task. In most BUSINESS SITE projects a +20–30% increase in dwell in the offer area and a 25–40% reduction in TFF to the CTA produced an uplift in clicks of 8–20% and improved overall conversion by 5–15%. I treat the micro-level (fixations and saccades) as the “building blocks of perception”, and the macro-level (scanpath) as the sequence of decision-making. Hence the KPIs: attention share at key AOIs, dwell increase when comparing variations, time-to-first-fix as an indicator of priorities, and the share of users whose scanpath goes through critical steps in the required order.
Segmentation by device is essential. In mobile interfaces scanpaths are shorter and fixations more compact, so TFF is especially sensitive to the placement of the CTA “above the fold”. For desktops the priority is the balance between the right column and the center, where gazeplot and scanpath similarity metrics are useful to understand how stable the path is across different screen sizes.
Methodologies: pros and cons
The laboratory protocol gives maximum precision: sampling rate 60–120 Hz (in the premium class: up to 1000 Hz), spatial accuracy around 0.3–0.6° and low latency. This is convenient for fine design optimization, complex forms and scenarios that verify micro-movements. Webcam-based solutions provide scale and speed; typically 30–60 Hz, accuracy 1.0–2.0°, higher latency, but this is enough for comparative A/B evaluations of landing pages and banners.
When comparing approaches I look at domain adaptation: on mobile front cameras and in different lighting conditions webcam algorithms require calibration and parameter adjustments for specific devices.
In remote tracking I always include protocols for environmental control and latency-alerts to ensure comparability of metrics between participants and to exclude artifacts.
Eye-tracking checklist for a website
At the planning stage I formulate the business question in funnel terms: where we lose attention and how this affects ROI. I describe hypotheses through attention metrics: “TFF for CTA ≤ 1.2 s”, “attention share of the offer ≥ 35%”, “dwell time on the trust block +25%”. In the brief for the contractor I include goals, success metrics, task scenarios, an AOI list, accuracy requirements, the format of deliverables and deadlines; I add a consent protocol (consent) and GDPR requirements.
Sampling: an area of particular responsibility. For CRO tasks, usually 24–36 participants per device segment is enough (power analysis with expected average effect size d≈0.5 and α=0.05), for advertising and creatives: 40–60, accounting for between-subject variability. I stratify by key characteristics: online shopping experience, age, interface language, device type. We ensure representativeness with screening and quotas.
Logistics and timing. Preparation of materials, scenarios and devices takes 5–7 working days, data collection, 3–10 days depending on the methodology, analysis and the report, another 5–10 days. I always plan quality checkpoints: calibration, track validity, lighting control, glare compensation, as well as a review of missing data. In total a typical project takes 3–4 weeks from the brief to implementing hypotheses into the design system.
Choosing equipment, software and vendors
I compare three classes of solutions: stationary laboratory trackers for maximum accuracy, SDKs for webcam tracking based on computer vision and deep learning, and wearable glasses for field scenarios. For UX on the web and mobile apps a webcam SDK with a good API is often sufficient; for complex forms and banking scenarios with signature verification and KYC a laboratory setup is more appropriate.
Criteria for choosing a vendor include accuracy and spatial precision, a stable sampling rate, measured latency, SDK/API support, TCO (total cost of ownership: license, upgrades, service), as well as vendor reviews and benchmarks on your real interface. At BUSINESS SITE we run a mini-validation before a project: calibration, drift correction, test AOIs, and comparison with a reference using control stimuli.
Data calibration and noise minimization
Calibration is the foundation of quality. I use 5–9-point procedures and add a quick recalibration when drift > 1° or after long pauses. In a session I monitor point validity, the percentage of lost samples, latency and illumination triggers, as well as head position (especially on laptops with a narrow camera field of view).
I compensate for gaps by interpolation only for short gaps and record the method in the protocol so downstream metrics remain interpretable. For privacy, we anonymize biometric data and isolate video stream storage.
Sessions to exclude: when track validity is below the threshold (for example, <70% valid samples), significant drift, high latency, and failure to follow the scenario. Clear exclusion criteria should be agreed in the brief to manage timelines and quality without surprises.
Preprocessing and event detection
For detecting fixations and saccades I use threshold and velocity algorithms (I‑DT, I‑VT) and, when necessary, probabilistic models for complex trajectories. Filter parameters are tuned to the sampling rate and device type; smoothing (for example, Savitzky–Golay) helps reduce noise while preserving the movement shape. Event detection affects attention metrics, so parameter transparency is part of the reporting.
Gaps up to 75–100 ms I fill with linear interpolation; long gaps I mark as unknown to avoid distorting dwell time. I define AOIs using a hybrid approach: manual annotation of key blocks plus gaze clustering (DBSCAN/Mean‑Shift) to reveal unexpectedly attention-grabbing areas. I aggregate heatmaps with normalization for inter-subject variability so that attention shares are correctly comparable across groups.
Attention analytics: heatmap and scanpath
Heatmap helps assess the distribution of attention and the visual hierarchy. Gazeplot shows the sequence of fixations and transitions, and a dwell map/attention map shows the depth of processing within an AOI. For a quick review by management I place attention heatmaps and click heatmaps side by side: discrepancies instantly indicate where attention is not converting into action.
I analyze sequences via scanpath: I compare design variants using MultiMatch and Levenshtein distance on AOI sequences. This is convenient for mobile UX, where scanpaths are compact and fixation order is critically important. AOI analysis includes TFF, time on AOI, transitions between zones, and comparisons by segments, for example new vs returning shoppers or carts with «Nova Poshta» versus pickup.
The key step is translating visualizations into hypotheses. If a banner captures 25% of the attention share from the price on the product card, I move the promo down or reduce its visual weight; if TFF for the CTA lags, we change contrast, copy, and placement. Then, an A/B test and verification of uplift.
Statistics, verification and significance assessment
For hypothesis testing I use bootstrap and permutation tests to avoid strict distributional assumptions. When the task requires a probabilistic formulation («the chance that TFF in variant B is ≥200 ms shorter»), Bayesian analysis with reports of posterior intervals helps. For multiple comparisons I apply FDR‑correction to control false positives.
Power analysis: a mandatory stage of planning: the expected effect, metric variance and desired power determine the sample size. Replication on a second panel reinforces confidence and reduces the risk of bias (bias), especially with remote collection. For business, not only statistical but also practical significance matters: for example, reducing TFF by 300 ms in the mobile cart often gives +3–5% to revenue per user, which already changes priorities in the roadmap.
Integration of eye-tracking with A/B and CRO
The idea is to combine data layers: gaze metrics, clicks, scroll, session replay and product analytics. I link AOIs to events in the analytics system, build combined KPIs (for example, «attention share of the offer ≥35% and CTR ≥9%») and design A/B tests that include eye-tracking in the early cohort. Then we scale the winner to all traffic.
Practice of optimizing landing page based on gaze data points to typical hypotheses: simplifying the headline, visual hierarchy of price and the button, reshuffling «trust signals» (delivery «Nova Poshta», payment with PrivatBank/Monobank cards, returns) into the area of early fixations. Predictive modeling of attention based on gaze patterns allows prioritizing interface elements and building attention‑driven personalization: changing the order of blocks for traffic segments without lengthy manual testing.
The combination of eye-tracking and click metrics makes hypotheses more accurate. If attention heatmaps and click heatmaps diverge, the focus shifts to micro-copy and control contrast, and if the paths coincide, the emphasis goes to the value of the offer and the proposition.
Calculation of ROI and TCO: cost and case
Costs consist of equipment/SDK licenses, moderator work, panel recruitment and rewards, data processing, analytics and integration. TCO includes infrastructure maintenance, QA, retraining of webcam models and support in subsequent sprints. When designing the budget I evaluate not only a one-off pilot but also the payback period taking CLV into account.
The ROI calculation method is simple: ROI = (Additional profit, Project cost) / Project cost. Additional profit = uplift in conversion × traffic × average revenue × observation period, plus contribution to CLV (for example, increased repeat orders). In BUSINESS SITE cases for an online store with payments through Ukrainian banks we reduced TFF on the CTA by 37%, increased dwell on the delivery block «Nova Poshta» and guarantees by 28%, which resulted in +14% to order conversion and +7% to average order value thanks to upsells. The project paid back in 7 weeks with a subsequent CLV growth of 4–6% over the quarter.
Implementing eye-tracking in a product team
I embed attention research into agile processes: short planning cycles → collection → analysis → implementation. In each sprint there is a slot for “attention hypotheses”, and the results are delivered in a storytelling format for designers, marketing and management. The BUSINESS SITE practice confirms: regularity is more important than a one-off “perfect” test.
A long-term program includes user panels (by device and segment), a standardized protocol, quality control when outsourcing and a library of typical AOIs by product. We use QA checklists, tracker benchmarks and automated reports for C-level with key KPIs: attention share, dwell increase, time-to-first-fix and contribution to business metrics.
GDPR, consent and biometric anonymization
Eye tracking is considered biometric data, so GDPR and consent practices are mandatory. I use informed consent with a clear description of purposes, retention periods, participant rights, and contact details of the responsible person. Anonymization and pseudonymization, segregated storage of video streams, and role-based access restrictions are standard in our projects.
In consent templates we record that data will be used only for research purposes and clarify that biometric signals are aggregated and are not used for individual targeting. Enhanced protective measures and separate procedures apply to vulnerable groups. The security infrastructure includes encryption at rest and in transit, access logs, and compliance documentation.
Machine learning and synthetic data
Deep learning increases the accuracy of webcam-based gaze estimation and robustness to varying conditions. Domain adaptation helps calibrate models for smartphone and laptop cameras and different lighting. For fine-tuning, synthetic datasets and open datasets are suitable, and the academic foundations (including Yarbus’s research on how gaze path depends on the task) help validate interpretations.
The combination of eye-tracking with EEG, GSR and facial coding strengthens behavioral analytics: we see not only gaze direction, but also the level of engagement and emotional reaction. In advertising and outdoor this enables attention-based creative optimization: creatives are ranked by their ability to quickly attract and hold gaze, and then go through A/B validation on clicks and conversions.
Research templates for C-level
The brief to the contractor includes: goals and hypotheses, KPIs (attention share, dwell, TFF), the target sample and stratification, scenarios, a list of AOIs, methodology (lab/webcam/field), requirements for accuracy and latency, data and visualization formats, timelines and risks. The protocol describes calibration, drift correction, interpolation rules and session exclusion criteria.
I structure the analytical report as an executive summary with key findings, visualizations (heatmap, gazeplot, scanpath), prioritized recommendations and an implementation roadmap. For management, financial arguments are important: the expected uplift, contribution to CLV, cost of changes and payback period. I build storytelling from the business objective to the visual insight and the A/B plan.
The acceptance checklist includes validation of source files, replicability of calculations, correctness of heatmap normalization and transparency of filter parameters. This procedure reduces misunderstandings and speeds up implementation.
Frequently Asked Questions
Our FAQ collects key explanations about which eye-tracking metrics are truly important when evaluating a landing page. Below you will find a list of metrics and recommendations on their interpretation and prioritization for improving conversion.
Eye-tracking metrics for evaluating a landing page
For a landing page I look at fixation duration and dwell time in the offer and CTA area, time to first fixation (TFF) for key blocks, and attention share between competing elements. If TFF is high, we strengthen visual hierarchy; if dwell on “trust signals” is low, we move them into earlier fixations; attention share indicates what “consumes” attention and how to place emphasis when optimizing the landing page based on gaze data.
Remote or laboratory tracking?
With a limited budget and UX/CRO tasks, remote (webcam) tracking provides sufficient accuracy and scalability; sampling rate 30–60 Hz and accuracy 1–2° cover variant comparisons. Laboratory eye-tracking is justified when micro-details, strict control of conditions and minimal latency are important. I tie the choice to the cost and ROI of an eye-tracking study: for quick decisions: webcam, for critical scenarios: lab.
How many participants are needed for eye-tracking?
I calculate sample size via power analysis, based on the expected effect and metric variance. For typical CRO tasks 24–36 participants per segment are enough, for creative testing: 40–60; with broad segmentation I build in stratification and control sample representativeness and bias via screening.
How to link gaze data to KPIs?
I build correlations between attention and sales KPIs by linking AOIs and events in analytics, calculate the contribution of attention to funnel conversions and attribute attention in marketing funnels. Then I measure uplift and calculate how to measure the return (ROI) of eye-tracking projects in monetary terms, including impact on CLV and the payback period.
Risks in remote data collection
Main risks are tracking quality, participant environment, and fraud. I use environment checks, latency alerts, validity checks, lighting and posture instructions, and data quality management when outsourcing: identity verification, CAPTCHAs and behavioral markers. To minimize artifacts and noise in the data: repeated calibration and session exclusion thresholds.
Conclusions
I view eye-tracking as a reliable tool for analyzing user attention that provides practical insights for UX, CRO, and advertising and directly impacts conversion and LTV. The workflow is simple: hypothesis planning → collection with quality control → preprocessing and event detection → analytics and statistics → implementation and ROI evaluation with A/B‑validation.
Call to action
Ready to move from intuition to measurable solutions? I suggest starting with a pilot brief and a checklist tailored to your product: goals and hypotheses, current KPIs (conversion, CTR, LTV), access to analytics, key pages/screens and an estimated budget. The BUSINESS SITE team will prepare an ROI estimate and an implementation plan – with clear timelines, risks and expected uplift, based on our experience in projects for e‑commerce, financial services and pharmaceutical brands.
Rate the article:
