Co-authored by Assaf Elovic and Harrison Chase. You can also find a version of this post published on Assaf's Medium.

Why do some AI products explode in adoption while others struggle to gain traction? After a decade of building AI products and watching hundreds of launches across the industry, we’ve noticed a pattern that has almost nothing to do with model accuracy or technical sophistication.

The difference comes down to what we call “CAIR” — Confidence in AI Results. This psychological factor can actually be measured, predicted, and optimized for. It’s not just another vanity metric. It’s the hidden variable that determines whether your AI product will succeed or fail, regardless of how impressive your underlying technology is.

Understanding user adoption barriers

Adoption is fundamentally blocked by fear. For maximum adoption, you want low fear and high confidence. We need a metric for this confidence that we can optimize for. That’s where CAIR comes in.

CAIR measures user confidence through a simple relationship that balances the value users get against the psychological barriers they face:

The equation makes intuitive sense:

• Value: The benefit users get when AI succeeds
• Risk: The consequence if the AI makes an error
• Correction: The effort required to fix AI mistakes

When CAIR is high, users embrace AI features enthusiastically. When CAIR is low, adoption stalls no matter how technically impressive your AI is.

CAIR is primarily determined by product design decisions, not just the underlying AI capabilities.

Of course, probability of success matters — if your AI fails most of the time, no amount of clever product design will save you. Success probability is already baked into the Value component of our equation. When AI fails, users get zero value, which tanks CAIR naturally. More importantly, we’re focusing on the variables you can actually control as a product team. Model accuracy improves over time as AI vendors ship better models, but the product design decisions that determine Risk and Correction effort? Those are entirely in your hands, and they’re what separate successful AI products from failed ones using identical underlying technology.

Why Cursor conquered coding

Consider Cursor, the AI-powered code editor that’s taken the developer world by storm. In theory, code generation should create low confidence -incorrect code could crash systems serving millions of users. Yet Cursor has seen explosive growth because it’s engineered an experience where confidence approaches maximum.

Let’s apply our equation to Cursor using a simple High/Medium/Low scale:

• Risk: Low — Code is generated locally in a safe environment, never touching production systems
• Correction: Low — You simply delete the suggestion and write it yourself
• Value: High — Hours of saved coding time and mental effort

CAIR = High ÷ (Low × Low) = Very High

This is the product design genius of Cursor. Imagine if they had designed the product to automatically commit generated code to production systems. The Risk would skyrocket to High, but Correction could still remain Low/Mid if they enabled easy rollbacks through version control. Even in this high-stakes scenario, the low correction effort would still yield reasonably high CAIR because while the stakes are high, recovery is easy.

However, if the system lacked easy rollbacks, both Risk and Correction would become High, completely changing the equation:

Alternative Design CAIR = High ÷ (High × High) = Low

That dramatic decrease in confidence would likely kill adoption, even with identical AI model quality. This illustrates why CAIR is fundamentally about product experience, not just technology capability, and why the separation of Risk and Correction effort allows for more nuanced product design decisions.

Creative writing tools like Jasper follow the same pattern. They position AI as a collaborative partner rather than an autonomous creator, keeping CAIR high by ensuring users retain editorial control.

The mid-CAIR opportunity: Monday AI

Monday.com offers a fascinating case study in moderate confidence. Their AI Blocks can create automations and deploy changes, but these modifications take effect immediately on users’ Monday boards, boards that often contain business critical workflow data connected to real operations.

Let’s break down each variable:

Risk: Medium — Monday boards are “production data” that teams rely on for daily operations, often connected to other workflows throughout an organization. An incorrect automation could cascade across departments, send wrong information to clients, or disrupt project timelines.

Correction: Medium — Because boards are interconnected with other systems, fixing AI mistakes requires detective work to identify changes, understand downstream effects, and manually reverse automation changes across multiple workflows.

Value: High — The AI can automate tedious manual workflow management, potentially saving teams dozens of hours per week on routine operations like updating project statuses and triggering follow-up actions.

CAIR = High ÷ (Medium × Medium) = Moderate

This moderate CAIR creates adoption hesitation, especially for users managing critical workflows. The psychological barrier isn’t the AI’s capability, it’s that users must commit to changes without being confident about them first.

The framework suggests a clear product improvement: adding a preview interface where users can evaluate AI changes before they go live. This single design change would dramatically reduce Risk from Medium to Low by separating the “testing” from the “deploying” mental models.

This simple UX improvement would boost CAIR significantly and likely translate to much higher adoption rates. It’s a perfect example of how CAIR analysis can identify high-impact product improvements that require zero changes to the underlying AI technology, just smarter product design around the same AI capabilities. It’s a perfect example of how CAIR analysis can identify high-impact product improvements that require zero changes to the underlying AI technology.

While workflow optimization tools like Monday AI represent moderate risk scenarios, some domains face fundamentally different challenges due to both high stakes and AI’s inherent limitations with numerical reasoning.

High stakes domains: Where design matters most

Financial services and healthcare illustrate how AI’s fundamental limitations, particularly with math and numerical reasoning, create inherently low CAIR that requires careful product design to overcome.

The challenge isn’t just about consequences, it’s about capability gaps. LLMs are fundamentally unreliable with mathematical calculations, numerical analysis, and precise reasoning. This makes domains like accounting, tax preparation, and investment advice extremely challenging regardless of how well you design the user experience.

Consider AI tax preparation software. The core issue isn’t just that mistakes have consequences (IRS audits, penalties), but that LLMs can’t reliably perform the mathematical calculations and rule applications that tax preparation requires. If such a system automatically filed returns, the CAIR would be devastating:

Auto-Filing Tax AI CAIR = High ÷ (High × High) = Very Low

TurboTax’s success comes from recognizing this fundamental limitation. Rather than just offering final expert review, they’ve built human oversight throughout the entire process, guided workflows where humans input data, AI suggests optimizations, and humans approve each step. The AI augments human capability rather than replacing human judgment on numerical calculations.

Similarly, auto investment traders face the dual challenge of mathematical complexity and severe consequences. Autonomous systems that execute trades create paralyzing low confidence:

Autonomous Trading CAIR = High ÷ (High × High) = Very Low

Successful platforms like Wealthfront achieve moderate CAIR by limiting AI to pattern recognition and trend analysis while keeping humans responsible for all numerical decisions and trade execution.

Healthcare faces identical patterns. AI diagnostic tools struggle with the numerical precision required for dosage calculations, risk assessments, and measurement interpretations. Successful ECG analysis tools achieve moderate CAIR by positioning AI as a pattern detection tool that highlights anomalies for human interpretation, rather than attempting numerical diagnosis.

The pattern is clear: high-stakes domains requiring numerical precision aren’t waiting for more accurate AI, they’re designing around AI’s fundamental limitations while leveraging its pattern recognition strengths. This creates sustainable competitive advantages that are harder to replicate than pure technical improvements.

The five principles of CAIR optimization

After analyzing successful AI products, we’ve identified five reliable strategies for optimizing CAIR. These aren’t theoretical, the data consistently validates their effectiveness across diverse industries and use cases.

1. Strategic human-in-the-loop (Optimizes all three variables) — Teams often think “autonomous” equals “better,” but this dramatically increases Risk in most scenarios. You must add human oversight, but strategically. Adding it everywhere kills Value, while placing it at key decision points maximizes CAIR. For instance, requiring approval for every suggestion kills productivity, but requiring approval before irreversible actions maintains both safety and utility. The art lies in identifying where human oversight adds the most CAIR optimization for the least value dilution.
2. Reversibility (Reduces Correction) — When users know they can easily undo an AI action, the effort to correct mistakes plummets. The psychological safety of a clear “escape hatch” transforms anxiety into confidence. We’ve consistently seen adoption rates double simply by adding prominent undo capabilities.
3. Consequence isolation (Reduces Risk) — Create safe spaces for AI experimentation through sandboxes, previews, and draft modes. This separates the “testing” from the “deploying” mental models, effectively eliminating consequence fears during exploration. Sandbox environments consistently show 3–4x higher adoption rates.
4. Transparency (Reduces Risk and Correction) — When users understand why an AI made a decision, they can better assess its reliability (reducing perceived Risk) and identify specific issues to fix (reducing Correction effort). Explanation features dramatically increase repeated use because users can correct specific flawed assumptions rather than throwing out entire AI outputs.
5. Control gradients (Increases Value while managing Risk) — Allow users to calibrate CAIR to their personal comfort level. Start with low risk features and progressively offer higher value capabilities as confidence builds. This acknowledges that everyone has different risk tolerance and creates a natural progression path.

Reframing AI product development

CAIR fundamentally changes how we evaluate AI readiness. Instead of only asking “Is the AI accurate enough?”, we should also ask “Is the CAIR high enough for adoption?”

This shift moves the conversation from purely technical metrics to a balanced technical product perspective. The most important questions become:

• How easily can users correct AI mistakes?
• How severe are the consequences of AI errors?
• How much value does successful AI completion provide?
• How much control do humans retain at critical moments?
• How clearly are limitations communicated?

For product leaders, this insight is liberating. You don’t need to wait for perfect AI to create successful products. An 85% accurate AI in a high-CAIR design will consistently outperform a 95% accurate AI in a low-CAIR design in terms of user adoption and satisfaction.

For organizations, this means AI initiatives should be jointly led by product and AI teams, with product design decisions considered as important as model tuning in determining success. AI readiness assessments must include CAIR calculations, not just technical evaluations.

The companies that will win in AI aren’t necessarily those with the most advanced models, they’re the ones that understand how to engineer confidence and maximize CAIR through thoughtful product design.

Start here: Calculate the CAIR score for your current AI features. Identify your lowest CAIR touch points, pick one, and implement a single CAIR optimization principle. Measure adoption before and after. You’ll likely see the difference immediately, and more importantly, you’ll have a repeatable framework for every AI product decision that follows.

The question isn’t whether your AI is good enough. The question is whether your users are confident enough in it to find out.