ExpertsFMWhy AI Could Finally Fix Healthcare Access | Zocdoc CEO Oliver Kharraz
The a16z Show · October 17, 2026
5 Selected Lessons:
case · 2 heuristic · 2 priority · 1
Patient experience over every other metric in network management
Source
The a16z Show — Why AI Could Finally Fix Healthcare Access | Zocdoc CEO Oliver Kharraz (October 17, 2026) · at 27:05 ▸ · at 29:37 ▸
When this applies
when a platform must choose which stakeholder's metrics to optimize for first among several (supply partners, revenue, end customer)
The takeaway
In a multi-sided marketplace, rank the end customer's experience above all other operating metrics, and trust that other stakeholders benefit as a consequence.
Why it works
Optimizing first for the end customer's experience creates conditions (trust, usage, word of mouth) that downstream benefit the other stakeholders (providers, payers), whereas optimizing for supplier convenience first can degrade the customer experience that drives the whole system.
One level up
Align incentives around whichever party's satisfaction compounds trust and volume for the whole system.
Where it stops holding
This stops working when the end customer's short-term preference conflicts with system-wide safety/quality (e.g., consumer demand for speed over credentialing rigor).
Detail
hierarchy:
- Patient experience / patient metrics
- Provider utilization and satisfaction
- Payer/partner relationship metrics
- Other operating metrics
rationale: We're really much more driven by the patient experience and patient metrics than we are by any other metric. And that's because we believe that when the patient wins, everyone wins — paired with maintaining stricter-than-regulatory qualification standards and removing underperforming doctors.
In their words
And we're really much more driven by the patient experience and patient metrics than we are by any other metric. And that's because we believe that when the patient wins, everyone wins.
Loyalty is earned by quality, not engineered by lock-in
Source
The a16z Show — Why AI Could Finally Fix Healthcare Access | Zocdoc CEO Oliver Kharraz (October 17, 2026) · at 9:37 ▸
When this applies
when deciding whether to use structural lock-in versus service quality to retain customers
The takeaway
Measure whether your retention is already high organically before investing in mechanisms to force retention.
Why it works
Once a customer makes a high-consideration choice (like picking a doctor), they tend to stick with a good experience on their own, so artificial retention mechanisms add friction without adding loyalty.
Where it stops holding
This stops working when customers are indifferent between options or switching costs are otherwise so low that no amount of quality retains them without deliberate stickiness mechanisms.
Detail
rule: Don't lock patients in structurally; let demonstrated quality of care drive repeat bookings
threshold: 75-80% rebooking rate with the same provider when booking in the same specialty
rationale: High organic repeat rate shows lock-in mechanisms are unnecessary and that openness doesn't cannibalize retention.
In their words
we see actually extremely high loyalty to the providers that they go to, a roughly 75 to 80% rebooking rate with the same provider when a patient books in the same specialty. So it speaks to the fact that it's not really necessary to lock them in through the systems you build.
Full agency stays with the end user, even when AI agents mediate
Source
The a16z Show — Why AI Could Finally Fix Healthcare Access | Zocdoc CEO Oliver Kharraz (October 17, 2026) · at 14:33 ▸
When this applies
when a marketplace must decide whether to apply its own judgment to rank options for users versus passing all options through
The takeaway
Don't change your fairness/agency model just because a new intermediary (an AI agent) sits between you and the end customer — extend the same rules to the new intermediary.
Why it works
Treating the AI agent as an extension of the user (not a separate customer to satisfy) preserves the original design principle of user agency and avoids introducing new, potentially conflicting ranking logic.
Where it stops holding
This stops working when AI agents need curated or ranked results because raw option lists overwhelm agent decision-making, requiring the platform to build new ranking logic.
Detail
rule: Present all options to the patient (or their agent) and let them choose; don't apply platform-level preference in routing
threshold: N/A — stated as an unconditional policy
rationale: We really look at that as one the patient and the agent. We don't differentiate between these two things.
In their words
We have always sort of given all the agency to the end user, to the patient, right? Like we present the options the patient chooses. And in our view, these AI agents are just going to be extensions of the patients.
The post-it note discovery
Source
The a16z Show — Why AI Could Finally Fix Healthcare Access | Zocdoc CEO Oliver Kharraz (October 17, 2026) · at 4:39 ▸
When this applies
when a system repeatedly fails in ways that smart analysis of the formal data can't explain
The takeaway
When a data/system problem resists analysis by experts examining the formal records, go observe the actual human workflow that produces the records.
Why it works
The discrepancy wasn't in the software at all — it was a human workaround (office manager using a post-it instead of updating the EHR) invisible to any data audit; only direct observation of the actual workflow revealed it.
One level up
The ground truth of a process often lives in informal human workarounds, not in the system of record.
Where it stops holding
This stops working when the root cause truly is in the formal system/code rather than in human workflow deviations, making direct observation less useful than code review.
Detail
situation: A practice kept producing cancellations because the doctor's reported availability in the EHR didn't match reality, and the smartest engineers at Zocdoc couldn't figure out why despite deep analysis.
decision: Kharraz personally sat next to the office manager to observe the actual workflow instead of analyzing data remotely.
outcome: Discovered the doctor communicated unavailability via a post-it note stuck to the monitor instead of updating the EHR — one of roughly a thousand similar 0.1% edge cases the team eventually solved.
lesson: Direct observation of real-world workflow uncovered what pure data analysis could not, and this class of problem had to be solved one edge case at a time over years.
In their words
I sat next to the office manager. And what I realized is that sometimes the doctor would walk out and say, Hey, I'm out next Friday. And the office manager would actually take a post-it note and attach that to their monitor instead of going into the HR.
From flat subscription to usage-based discovery-booking fee, with free facilitation
Source
The a16z Show — Why AI Could Finally Fix Healthcare Access | Zocdoc CEO Oliver Kharraz (October 17, 2026) · at 23:51 ▸
When this applies
when a flat-fee pricing model under- or over-charges different segments of a heterogeneous customer base
The takeaway
Price according to the specific unit of value you create for each customer rather than a flat fee that misprices heterogeneous usage.
Why it works
A flat subscription fee misprices value for both high-volume and low-volume customers, causing churn at both ends; charging only for the specific value delivered (a 'discovery' booking where the platform created the match) aligns price with value and removes friction for both those who only need basic facilitation and the regulator's scrutiny.
One level up
Pricing structure should mirror the actual locus of value creation, not administrative convenience.
Where it stops holding
This stops working when usage-based pricing introduces too much complexity or unpredictability for customers to budget around, or when regulatory constraints (like anti-kickback rules in healthcare) prohibit value-based fees.
Detail
situation: Zocdoc originally charged every doctor an identical flat subscription fee regardless of how many patients they saw.
decision: Zocdoc worked with regulators to get permission to charge only for 'discovery' bookings (patient found the doctor via general search) and keep bookings for a specifically sought doctor free.
outcome: 97% of providers signed up in the last 12 months came via inbound word-of-mouth referral, and the low financial bar to join expanded the provider network.
lesson: Aligning price with the specific value created (discovery) rather than flat access fees improved provider satisfaction and growth.
In their words
we charge a provider for a discovery booking, meaning a booking where the patient was doing a general search and we connected them to that doctor. Notably, we do not charge for bookings where a patient is looking for a specific doctor to begin with, and all we do is facilitate the booking.