Clean, governed data will make AI useful and value-based care work in 2026
Good data is the lifeblood of care. As 2026 approaches, the healthcare leaders who win will treat data quality as a clinical safety issue, not a back-office task. That's the throughline from Joanna Engelhardt, vice president of product management at Health Gorilla, who brings deep experience in EHRs, ambient documentation, coding and value-based care.
Her take is straightforward: AI and accountable care both stall when data is incomplete, inconsistent or stuck in the wrong system. Fix that, and everything else gets easier.
AI needs completeness, context and checks - not just "clean" data
Most teams sanitize data, then stop. That's not enough. AI needs a full picture of the patient, or the model will quietly underperform no matter how advanced it is.
Think about how often patients are asked to recall their own medical histories - past surgeries, anesthesia reactions, medication dosages. Those gaps become model blind spots. To make AI reliable at the point of care, health systems need to:
- Deduplicate and reconcile records across sources with strong identity resolution.
- Run plausibility checks (e.g., dose ranges, age/diagnosis mismatches, time-order conflicts).
- Attach context (who recorded it, when, care setting, certainty) so models don't treat every data point as equal.
- Keep data fresh - addresses, coverage, meds and encounters change constantly, so real-time or near-real-time exchange matters.
Until responsiveness, completeness and contextual accuracy are baked in, AI's ceiling will be set by the gaps in the chart.
Value-based care: data must show up where decisions happen
Value-based care lives or dies on timely, relevant data. You need to find high-risk, high-cost patients early, then make proactive interventions easy to execute during real visits - not weeks later in a dashboard no one opens.
Health systems need insights that prioritize care, surface gaps, and measure intervention impact. But insights aren't useful if they're a chore to use. The experience has to be intuitive, fast, and integrated into clinical workflows so clinicians act confidently, not reluctantly.
Outcomes beat outputs: a story we've all lived
Engelhardt shared a familiar scenario: a child's urgent care X-ray couldn't be accessed by the orthopedic clinic. A CD, a manual pickup, an unreadable file, and finally an emailed image viewed on a phone. Technically "interoperable." Practically useless.
That's the difference between outputs (boxes checked) and outcomes (care improved). Teams that win obsess over the handoff in real life - the nurse, the MA, the physician in a 20-minute slot - and design for their reality. Ship less, but make it count.
2026: drop the brakes on data flow - with consent and privacy intact
The industry can't keep hoarding data under the banner of caution. Patients should own their data and share it securely with any provider or app they trust - with granular consent and strong guardrails. That's how outcomes improve at scale.
Domestically, frameworks like TEFCA are moving the ball. In 2026, the focus should be clearing remaining bottlenecks at home. In 2027, we'll need to confront the harder problem: global records that don't follow the patient across borders.
A practical playbook you can start now
- Stand up a data quality service: deduplication, identity resolution, plausibility and temporal checks as a shared utility.
- Standardize on FHIR APIs for ingestion and exchange; treat event notifications as table stakes for care coordination.
- Attach provenance and context to every data element (source system, timestamp, setting, confidence).
- Measure freshness, completeness and match rates; publish a monthly scorecard that product and clinical leaders review together.
- Push risk, gap and next-best-action insights into the EHR workflow (not a separate portal).
- Build granular consent and audit-by-default into your platform; make it easy for patients to share and revoke.
- Add SDOH and pharmacy data where appropriate; they often move the needle more than another lab feed.
- Test in real clinics with "at-the-elbow" feedback; prioritize fixes that cut clicks and reduce rework.
- Vet vendors for TEFCA participation and QHIN connectivity; avoid one-off integrations that age poorly.
- Tie incentives to outcomes: fewer avoidable ED visits, closed gaps, lower total cost of care - not feature counts.
For leaders working on VBC models
If you need a quick refresher on federal programs and direction, start with CMS' overview of value-based programs: CMS Value-Based Programs. It's a useful reference point for aligning your data strategy with reimbursement.
The bottom line
AI will help. Value-based care will progress. But only if the data is complete, current and contextual - and moves where it needs to, when it needs to. Build for outcomes, not checkboxes, and 2026 will look a lot better than 2025.
If your teams are upskilling on applied AI for healthcare roles, this curated list can speed the process: AI courses by job.
Your membership also unlocks:
