Start with the research question
A price is difficult to interpret without knowing what the study must establish. Testing a single static advertisement is different from observing a multi-step task on a live site. The participant profile, number of stimuli and analysis plan all affect the work required.
Decide whether you need gaze measurements at all. If the question is whether a checkout is understandable, a task-based usability test may provide useful evidence without eye tracking. If the location and sequence of gaze are central, discuss that requirement with the research provider.
Ask for an itemised proposal
Compare scopes rather than a headline price. A quote that includes recruitment and interpretation is not equivalent to an equipment rental or a self-service software subscription.
- Define the audience and recruitment criteria.
- Specify stimuli, tasks, devices and session duration.
- Ask how calibration, data quality and unusable sessions are handled.
- List the deliverables: recordings, gaze data, findings and recommendations.
- Clarify participant incentives, preparation, analysis and follow-up costs.
What an AI prediction changes
An image heatmap can be generated without recruiting a sample for that analysis. That makes it convenient for early visual reviews and repeated design iterations. It does not supply the participant observations removed from that workflow.
Use the prediction to narrow a design question or choose variants worth researching. Do not convert the lower effort into a claim that the same research result has been obtained at a fraction of the price.
References: AI heatmaps versus eye tracking
What each approach delivers for the effort
Without quoting prices, the table sets out what changes between a study with participants and a prediction on an image. The right column is cheap because it answers a narrower question, not because it answers the same one faster.
| Item | Eye-tracking study | Predicted map |
|---|---|---|
| Participants | Recruited, screened, compensated | None |
| Equipment | Tracker, calibration, a lab or a remote setup | A screenshot |
| Time to a first result | Days to weeks, including analysis | Seconds |
| What it measures | Where these people looked, in what order, during a task | Where the model expects attention on this image |
| Repeat on a revision | A new session | Immediately, same settings |
| Evidence about comprehension | With a task and a debrief | None |
How well the cheap option agrees with the expensive one
Saliency models are evaluated against eye-tracking data, so the question has a partial answer. On the public MIT300 benchmark, UNISAL, the model Heatpoints runs, reaches an AUC of 0.877 and an NSS of 2.37 against 0.783 and 1.10 for a centre-bias baseline and 0.894 and 2.74 for the best listed model. Those images are photographs viewed freely, not interfaces with a task, which is precisely the case where a study earns its cost.
References: MIT/Tübingen Saliency Benchmark · How accurate are AI attention heatmaps?
Build a proportionate plan
For an early draft, begin with a clear design objective, a screenshot review and a prediction. For a consequential decision, add evidence from the intended users and a method that measures the outcome you care about.
This article deliberately avoids a universal dollar range. Published offers can bundle different services, and a precise-looking range without comparable scope would not help you estimate your project. Ask providers for current quotes using the same brief.