Most children aren’t diagnosed with autism until age 4 or 5. By then, they’ve already missed the most critical window for intervention—the 12 to 36 months when therapy is most effective and outcomes are best.
This delay isn’t because we can’t detect autism earlier. We can. The real problem is that traditional diagnostic assessment is slow, expensive, and depends on a small number of specialists. You need hours in a clinic. You need multiple assessments. You need specialists most families can’t access.
Over the last three years, something has shifted. Eye-tracking technology, combined with AI-powered screening tools, is beginning to compress this diagnostic timeline from hours to minutes. One FDA-cleared system now handles assessment in roughly 12 minutes. Another multi-clinic study showed 87.8% sensitivity across diverse populations.
This matters for three reasons. First, it’s clinical—kids get diagnosed earlier and access treatment sooner. Second, it’s economic—the autism diagnostics market is growing from $3.84 billion today to $9.50 billion by 2033. And third, it’s structural—this technology is starting to change how autism diagnosis works across the entire healthcare system.
I work with EarliPoint Health, one of the companies building these tools, so I have a front-row seat to this shift. But my goal here isn’t to sell you on any single company. It’s to walk you through what’s actually happening in autism diagnostics technology, what the data shows, what’s hype, and what it means for investors, operators, and treatment providers.
The Diagnosis Problem: Why We’re Missing the Window
Let me start with the clinical reality.
Autism spectrum disorder shows clear signs in the first 18 months of life. We’ve known this for over a decade. Parents notice it. Pediatricians see it. But diagnosis—actual, clinical diagnosis—typically comes 2 to 3 years later.
Why the delay?
The current standard for autism diagnosis requires behavioral observation, parent interviews, and sometimes direct testing. The Autism Diagnostic Observation Schedule (ADOS) is the gold standard, but it takes 1 to 2 hours and requires a trained administrator. You need a developmental pediatrician, psychologist, or developmental specialist. These specialists are bottlenecked—there’s often a 6 to 12 month wait for an appointment.
Geography matters too. Rural areas have almost no specialists. Many insurance plans don’t reimburse early screening, so it’s out-of-pocket. Families with lower income or less healthcare literacy get delayed even further.
Meanwhile, early intervention therapy—ABA (Applied Behavior Analysis) and other behavioral interventions—works best in the 12 to 36 month window. After age 3, outcomes are measurably worse. This means that the delay in diagnosis is directly causing worse outcomes for kids. It’s not just inconvenient. It’s clinically significant.
The market sees this too. Autism prevalence has climbed from 1 in 150 kids (2000) to roughly 1 in 36 today (CDC, 2023). More children means more need for diagnosis and treatment. The bottleneck hasn’t expanded to match demand. That’s why we’re seeing technology investment here.
The Technology Landscape: What’s Actually in Development
The autism diagnostics space isn’t monolithic. There are several technology approaches running in parallel:
Eye-tracking and gaze analysis. This is the most clinically advanced. Eye-tracking measures where a child looks, how long they look, how quickly they disengage. The assumption is that children with autism have distinct gaze patterns—differences in social attention, in object persistence, in visual processing. The system compares a child’s gaze data against normative databases and flags risk.
Digital screening questionnaires. Apps like the M-CHAT (Modified Checklist for Autism in Toddlers) ask parents behavioral questions. Low-cost, scalable, but entirely dependent on parental reporting. Useful as a first-pass filter but limited in objectivity.
AI behavioral analysis from video. Some companies are training models on video of child behavior—movement patterns, play behavior, engagement with objects. The idea is to flag atypical behavior automatically. This is early-stage; the clinical evidence is still building.
Voice and speech analysis. Atypical prosody and speech patterns are autism markers. Some startups are using machine learning to detect these patterns from audio. Again, still in earlier research phases.
Multi-modal approaches. The most promising tools combine several signals—eye-tracking plus behavioral video, for instance—to increase accuracy.
Of these, eye-tracking is the most mature. It has FDA clearance, multi-clinic validation, and published accuracy data. That’s where I’ll focus here, because that’s where the real clinical and commercial traction is.
Eye-Tracking: The Most Advanced Clinical Technology
Here’s what eye-tracking does: It removes subjectivity from the diagnostic process.
Traditional diagnosis relies on clinical judgment. A psychologist watches a child play, asks questions, interprets behavior. Judgment matters. Experience matters. But two clinicians watching the same child might reach different conclusions.
Eye-tracking is objective. A child looks at a screen. The system measures gaze. The data either matches known autism patterns or it doesn’t.
How it works in practice:
The child sits in front of a screen. Calibration takes a minute or two. Then the child watches visual stimuli—shapes, faces, objects, social scenarios. Eye-tracking hardware measures pupil position and gaze direction, typically at 60-120 Hz (60-120 data points per second). The system logs where the child is looking and when.
Software analyzes this gaze data against metrics like:
- Dwell time: How long the child looks at faces vs. objects.
- Latency to orient: How quickly the child looks at a face when it appears.
- Visual engagement: Patterns of fixation and saccade (eye movement).
These metrics are compared against normative data from large healthy populations and against known autism patterns. A machine learning model integrates these features and produces a risk score.
The clinical data:
EarliPoint Health received FDA clearance in June 2023 for their eye-tracking system. Here are their published metrics:
- Sensitivity: 78% (detects true autism cases 78% of the time)
- Specificity: 85.4% (correctly identifies non-autism cases 85.4% of the time)
- Overall accuracy: 82.1%
- AUROC: 0.90 (a measure of discriminative ability; 0.90 is very strong)
- Assessment time: ~12 minutes
- Target age: 16 to 30 months
For context: The ADOS, the current gold standard, has similar sensitivity (83%) and specificity (83%) but takes 60-90 minutes.
SenseToKnow, another eye-tracking system, published multi-clinic data on 475 toddlers with some impressive results:
- Sensitivity: 87.8%
- Specificity: 80.8%
- Results were consistent across sex and racial demographics—no performance degradation for girls or children of color, which is important (autism is historically underdiagnosed in these groups).
These aren’t perfect tools. A 78-88% sensitivity means you’re missing some true cases. A 80-85% specificity means some false positives. But they’re fast, scalable, objective, and they’re catching real cases.
The key insight: This technology is designed as a screener, not a diagnostic replacement. It’s meant to flag risk quickly so clinicians can prioritize full assessment. In a bottlenecked system, that’s actually powerful. You’re moving from “diagnostic wait-list is 6-12 months” to “risk assessment in 2 weeks, then prioritize clinical evaluation.”
AI-Powered Screening: What’s Working and What’s Hype
Beyond eye-tracking, there’s a lot of energy in AI-based screening. Some of it is solid. Some is still speculative.
Face and facial expression analysis is one area. The theory: Children with autism show atypical facial expressions and emotional recognition patterns. Some AI models are trained on images or video to detect these patterns. The research is encouraging but early. You’re seeing papers that show 75-85% accuracy in controlled lab settings, but real-world validation is limited. This will likely work as an adjunct tool (combined with other data) before it works as a standalone screener.
Movement and posture analysis is another track. Video-based models that track body position, motor patterns, engagement with objects. Again, promising research, but clinical validation is still building. Some early papers show 70-80% accuracy, but the sample sizes are small and often in clinical settings.
Voice and prosody analysis has shown promise in research. Atypical speech patterns (monotone speech, unusual rhythm, echolalia) are autism markers. Machine learning models trained on speech data can pick up these patterns. But, same story: research accuracy is decent, clinical deployment is limited, and reimbursement is unclear.
The reality gap: There’s often a gap between research accuracy and clinical applicability. A model that shows 82% accuracy in a controlled study with 50 kids might show 65% accuracy when you deploy it in a busy pediatric clinic with diverse populations. The real world is messier. Generalization is harder.
The tools that are actually moving into clinical practice—eye-tracking, primarily—have two things in common:
- Multiple independent validations (not just one company’s data)
- FDA clearance or equivalent regulatory pathway (which forces external validation)
Everything else is still in “promising research” territory. It might mature into clinical tools. But it’s not there yet.
The Market Opportunity: Why Money Is Flowing Into Autism Diagnostics
The market for autism diagnostics is growing fast, and for good reasons.
Market size:
- 2024: $3.84 billion
- 2033 (projected): $9.50 billion
- CAGR: 11.9%
For comparison, the broader diagnostics market is growing around 5-6% annually. Autism diagnostics is growing nearly 2x faster.
Growth drivers:
- Increased awareness and prevalence. Autism diagnoses have climbed from 1 in 150 (2000) to 1 in 36 (2023). Some of this is real prevalence increase; some is better detection. Either way, it means more families seeking diagnosis.
- Earlier screening adoption. More pediatricians are doing M-CHAT screening at well-child visits. More school districts are screening for developmental delays. More insurance is covering early screening. This expands the population getting assessed.
- Reimbursement expansion. Many states now mandate autism screening at age 18-24 months through Medicaid. Medicare covers diagnostic assessment. This means the addressable market isn’t just private clinics—it’s primary care, school districts, early intervention systems.
- Technology scaling. Eye-tracking and AI-based tools remove the specialist bottleneck. One system can screen hundreds of kids per clinic per year. That’s addressable market expansion.
Investment landscape:
The Autism Impact Fund raised $60 million and has invested in 16 companies focused on autism. Major investors include CVS, Deerfield Partners, and Optum (which invested in Cortica, an AI behavioral analysis company).
This signals institutional capital betting on the vertical. When CVS invests in autism diagnostics, it’s because they see it as core to their healthcare strategy. When PE firms like Deerfield allocate capital here, it’s because they see defensible unit economics and scale potential.
The picks-and-shovels play is interesting too. Not all the money goes to diagnostic tools. Some goes to electronic health record (EHR) integrations, telehealth platforms, referral networks, and outcome tracking systems. The infrastructure layer matters as much as the tool itself.
What This Means for ABA and Treatment Providers
If earlier diagnosis becomes standard, the demand side for ABA and behavioral therapy will shift structurally.
Earlier referrals. Right now, many kids don’t enter ABA until age 3.5-4. With earlier diagnosis, you could see referrals at 2-2.5 years. That’s a 12-18 month earlier start, which changes outcomes (for the child) and revenue (for the provider).
Wider referral funnel. Today, the referral funnel is: parents notice delay → pediatrician referral → diagnostic wait-list → diagnosis → ABA referral. Diagnostic technology compresses steps 2-4. It also means pediatricians have a fast, objective screening tool, which might increase the number of referrals coming from primary care (currently underutilized as a referral source).
Data integration opportunities. If a child gets screened with eye-tracking data showing specific patterns (e.g., high object persistence, low social attention), an ABA provider could use that data to personalize therapy. “This child’s core challenge is social attention. Here’s the therapy we’d recommend.” That’s a differentiation angle for forward-thinking providers.
Scaling challenges. Earlier diagnosis means higher volume, but it also means you need to scale capacity. If you’re a 10-therapist ABA practice and suddenly you have 30% more referrals coming in, you need to hire, train, and manage more staff. That’s operationally hard.
What This Means for Investors and Operators
This is where the real strategic clarity helps.
Platform plays vs. point solutions. The most defensible companies here aren’t just building the diagnostic tool. They’re building the platform around it—EHR integration, referral workflows, outcome tracking, insurance coding, reimbursement management. A point solution (just the eye-tracking tool) gets commoditized. A platform (diagnosis + referral network + outcome tracking) has moat.
Full-stack vs. specialized. Some companies are building full-stack solutions (diagnosis + telehealth + therapy), while others are focusing on the diagnostic piece and partnering with existing providers. Full-stack is harder to execute but higher value if you get it right. Specialized is faster to market but more competitive.
Regulatory moat. FDA clearance is real competitive advantage. It’s expensive ($500K-$2M, 18-24 month timeline), but once you have it, you have a differentiation story that’s defensible. Every competitor needs to go through the same process.
Reimbursement is the real gate. Technology accuracy doesn’t matter if insurance won’t pay for it. Right now, reimbursement for autism screening is expanding but still fragmented (state-by-state, payer-by-payer). Companies that navigate reimbursement early (getting CPT codes, building payer relationships) win. Companies that ignore it die.
Risk factors:
- Clinical displacement. If the technology doesn’t perform in real-world settings, adoption will stall.
- Reimbursement uncertainty. If payers don’t cover it, the addressable market shrinks dramatically.
- Specialist resistance. Developmental pediatricians and psychologists have incentives to maintain the current system. Some will adopt these tools; others will position them as inadequate.
- Demographic bias. Early AI systems sometimes perform worse on underrepresented groups. This is improving, but it’s a real risk.
What to Watch in 2026 and Beyond
A few things will determine how this sector evolves:
FDA pipeline clarity. How many more eye-tracking or AI-based systems get FDA clearance in the next 18-24 months? More clearances mean more competition and faster adoption. Fewer means the existing players (EarliPoint, SenseToKnow) have longer runway.
Reimbursement coding. The American Medical Association and CMS will need to create CPT codes for AI-based autism screening. Right now, reimbursement is happening under broad codes or bundled into clinic visits. Specific codes would expand the addressable market significantly.
Real-world performance data. Lab accuracy is one thing. Real-world accuracy—how these tools perform across diverse clinics, patient populations, and demographic groups—is another. Companies that publish real-world performance data will build trust faster.
Telehealth and home-based screening. Some companies are exploring at-home eye-tracking (using webcams and tablets). If you can do screening in the home rather than requiring a clinic visit, you remove another barrier. This is technically feasible but needs validation.
International expansion. Autism prevalence and diagnostic needs are global. European and Asian markets are starting to invest in diagnostic technology. This is a growth frontier that will drive investment and consolidation.
Integration with broader health platforms. The real value might come when autism diagnostics integrates with broader pediatric health platforms, school health systems, and insurance workflows. The technology is necessary but not sufficient. The platform is where the value accrues.
Bottom Line
Autism diagnostics technology is at an inflection point. Eye-tracking and AI-based screening have moved from research to clinical deployment. The market is growing nearly 12% annually. Capital is flowing in. The clinical case for early diagnosis is ironclad.
But this isn’t “technology solves everything” territory. Real adoption requires regulatory clearance, reimbursement coverage, clinical workflow integration, and clinician buy-in. The companies that navigate these pieces—not just the ones that build the best technology—will win.
For parents: This means your kid might get diagnosed earlier and access therapy sooner. That’s real progress.
For ABA and treatment providers: Prepare for earlier, higher-volume referrals. The diagnostic funnel is compressing.
For investors and operators: The defensible positions are in platforms (not point solutions), reimbursement navigation, and real-world performance validation. Regulatory moat matters. Picks-and-shovels plays (infrastructure) are underrated.
The diagnostic bottleneck that’s kept kids in the 2-5 age range untreated for 2-3 years is finally starting to crack. The technology is here. Now it’s about deployment.
