IRONWOODADVISORY GROUP
AI Accessibility Assurance

Your website is accessible.
Is your AI?

Customer-facing AI is moving beyond answering questions. It now recommends, decides, books, guides and acts — often connecting a digital conversation to a real-world outcome.

Ironwood independently tests whether AI can reliably serve people with disabilities from request to outcome.

We evaluate whether AI understands accessibility requirements, preserves them through complex tasks, uses reliable information about the real world, avoids unsupported assumptions, and knows when human intervention is required.

The Gap

Accessibility doesn't stop at the interface.

Traditional accessibility testing asks
Can a person with a disability use the interface?
AI creates another question
Can the AI actually serve that person correctly?

Consider a traveller telling an AI agent:

"I use a power wheelchair, cannot walk independently and need assistance at both airports. Help me book my trip."

The interface may be perfectly accessible.
The AI can still fail.

  • It may misunderstand the person's requirement.
  • Forget it later in the journey.
  • Use incomplete information about an aircraft, airport or hotel.
  • Make an unsupported accessibility claim.
  • Fail to escalate when the answer cannot be verified.
  • Or complete a booking that does not actually work for the traveller.

Ironwood tests that gap.

The Ironwood Assurance Chain

We test the journey from human need to real-world outcome.

Understanding is not enough. The outcome has to work.

Sample Scenario

From a simple prompt to a complex accessibility journey.

Customer request
"I use a power wheelchair. Book me a flight and make sure the assistance I need is arranged at both airports."

Passing this test requires much more than recognizing the words power wheelchair.

The agent may need to preserve mobility requirements throughout the booking, use authoritative airline and airport information, identify information that cannot safely be assumed, request additional details when necessary, escalate unanswered questions, and ultimately produce an itinerary that works for the traveller.

Ironwood records and scores every material decision along the journey.

RecognitionRetentionEvidenceReasoningEscalationOutcome
What Failure Looks Like

Not every AI failure is a model problem.

An accessibility failure can happen at several different layers.

Behaviour failure
The AI had the information but misunderstood, forgot or mishandled the customer's requirement.
Knowledge failure
The AI did not have sufficiently reliable accessibility information to answer correctly.
Integration failure
The information existed within the organization but was not available to the AI when it needed it.
Policy failure
The organization had no defined answer or process for the situation.
Escalation failure
The AI lacked sufficient certainty but continued instead of involving a human.

Ironwood identifies not only that the system failed, but why.

Services

Evidence, not opinions.

Every Ironwood engagement produces reproducible scenarios, scored outcomes and evidence behind material failures. Testing combines structured evaluation with input from people with lived disability experience and domain specialists.

Executive Briefing
30 MINUTES · FREE

A focused conversation about your customer-facing AI, the journeys it supports and where accessibility exposure may exist.

If independent testing is not warranted, we'll tell you.

Book a Briefing
Re-Test & Continuous Assurance
ONGOING

AI systems do not stand still. Models change. Prompts change. Tools change. Data changes. Policies change.

Ironwood re-tests critical journeys after changes and tracks whether previously successful accessibility outcomes continue to hold.

How an Assessment Works

Start with humans. Build what can be trusted.

01
Scope
Identify the AI systems, customer journeys and accessibility situations where failure could materially affect the customer.
02
Design
Create realistic scenarios using accessibility expertise, domain knowledge and lived experience.
03
Establish Ground Truth
Determine what a correct outcome requires and identify the authoritative information the AI should use.
04
Run
Execute scenarios against the AI system and capture its decisions, responses, tool use and outcomes.
05
Validate
Human reviewers assess reproducibility, severity, real-world impact and root cause.
06
Report & Re-Test
Deliver scored results, evidence and remediation priorities — then verify the fixes.
The Product Direction

From assessment to continuous assurance.

Ironwood begins with human-led testing for a deliberate reason:

You cannot automate a methodology you have not yet proven.

Every assessment helps strengthen the scenario library, scoring framework, evidence requirements and understanding of how AI fails people with disabilities.

As those systems mature, repeatable elements of testing can increasingly be automated.

The long-term direction is a continuous assurance layer that allows organizations to test their AI systems repeatedly as models, prompts, tools and underlying data change.

Today
Human-led assessment
Tomorrow
Repeatable assurance
Over time
Continuous testing
Where This Matters Most

When digital decisions meet the physical world.

The risk becomes particularly important when an AI interaction determines whether someone can successfully complete a real-world task.

Travel & Transportation
Flights, airports, mobility assistance, ground transportation and accommodation.
Hospitality
Rooms, facilities, physical access and guest services.
Healthcare
Appointments, facilities, procedures and patient journeys.
Financial Services & Insurance
Complex customer-service journeys and individualized support.
Retail & Commerce
Product suitability, adaptive requirements and physical-product recommendations.
Government & Public Services
Essential services where accessibility failures can prevent participation entirely.
Why Ironwood

Our accessibility experience began long before AI.

Ironwood is built on more than a decade of operating experience in adaptive commerce through Inclusia Brands — serving people with disabilities, families, clinicians and organizations across North America.

That experience exposed us to something technology teams do not always see:

Accessibility is not an abstract compliance requirement. It determines whether someone can actually participate in the real world.

AI is now becoming part of that journey. Ironwood exists to independently test whether it works.

10+ years
Operating customer-facing businesses serving people with disabilities
$11M+
Adaptive footwear sold through Inclusia Brands
112K+
Customer orders served across North America since 2014
Why Now

AI is moving faster than the testing layer around it.

AI systems increasingly plan, recommend, use tools and complete multi-step tasks. The more responsibility they assume, the more consequential their mistakes become.

  • Traditional accessibility testing remains essential.
  • AI safety testing remains essential.
  • Neither alone answers the question Ironwood is built around.
Can this AI reliably serve a person with a disability all the way to a successful outcome?
Get Started

Know before your customers do.

If your organization is deploying customer-facing AI, we can help determine where accessibility risk exists and whether independent testing is warranted. The first conversation is free and specific.

Book a 30-Minute Briefing
anthony@ironwoodadvisory.ai
Independent AI Accessibility Assurance
Testing the journey from human need to real-world outcome.