Here’s a pattern I’ve seen repeatedly: a health service decides to implement clinical AI, creates a project plan with an aggressive timeline, then watches everything take twice as long as expected.

This isn’t unique to healthcare—technology projects routinely exceed timelines. But clinical AI has specific factors that contribute to delays, and understanding them helps create realistic plans.

Typical Timeline Expectations vs Reality

When organisations first consider clinical AI, they often imagine something like:

• Month 1-2: Vendor selection and contracting
• Month 3-4: Technical integration
• Month 5-6: Testing and validation
• Month 7: Go-live

This adds up to about six months from decision to deployment.

Here’s what I more commonly observe:

• Month 1-4: Governance approval and vendor selection
• Month 5-8: Contract negotiation
• Month 9-14: Technical integration
• Month 15-18: Testing, validation, and local approval
• Month 19-21: Staged rollout and stabilisation

That’s 18-24 months from decision to full deployment. For complex implementations, longer still.

Why the gap?

Governance Takes Time

Before any clinical AI implementation, governance bodies need to approve it. This typically means:

Executive approval. Someone needs to commit budget and organisational resources.

Clinical governance review. The clinical risks and safeguards need assessment. This often involves clinical quality committees that meet monthly and require advance submission of documentation.

IT governance review. Technical architecture, integration, and security need approval. Another committee cycle.

Privacy review. Data flows and privacy implications need assessment, often involving health information managers or privacy officers.

Ethics review (sometimes). Depending on the application and organisation, ethics committee review may be required.

Each of these processes has its own timeline, documentation requirements, and decision cycles. Getting through all of them can take three to six months even when nothing goes wrong.

Organisations that plan for six-month total timelines are often still in governance when they expected to be going live.

Contract Negotiation Is Slower Than Expected

Vendor contracts for clinical AI involve complex issues:

Liability allocation. If the AI contributes to a patient harm, who is responsible? Negotiations on indemnification and liability can be protracted.

Data rights. Does the vendor have rights to use your data for AI improvement? What data leaves your environment? What happens to data when the contract ends?

Performance guarantees. What performance levels does the vendor commit to? What happens if performance degrades?

Exit provisions. How can you terminate if the AI doesn’t work? What’s the wind-down process?

Integration obligations. Who is responsible for technical integration? What support does the vendor provide?

Large health services have procurement and legal teams that review contracts carefully. This takes time. Vendors often have standard contract terms that don’t match health service requirements, requiring negotiation.

I’ve seen contract negotiations alone take six months. Budget three to four months minimum.

Technical Integration Is Complex

Connecting AI to clinical systems is harder than it appears.

Interface development. Even with standard interfaces like FHIR, connecting AI to electronic medical records, PACS, pathology systems, and other clinical infrastructure requires development work.

Data quality issues. AI systems expect data in specific formats. Source systems often have data quality issues that need remediation before AI can work effectively.

Infrastructure requirements. AI systems may require computing resources, network configurations, or security controls that need to be provisioned.

Testing environment setup. Before testing with real data, test environments need to be configured to match production conditions.

Integration timelines depend heavily on how well your IT environment is prepared. Organisations with modern, well-architected systems integrate faster than those with legacy infrastructure.

Budget at least four to six months for integration. For complex implementations with challenging legacy systems, longer.

Validation and Approval Have Their Own Timeline

Before clinical AI goes live, you need to validate that it works in your environment:

Technical testing. Does the system function correctly? Do integrations work? Are response times acceptable?

Clinical validation. Does the AI perform as expected on your patient population? This often requires running AI on historical data or prospective parallel testing.

User acceptance testing. Do clinicians find the system usable and useful? Workflow integration testing matters.

Approval processes. After testing, clinical governance needs to review results and approve production deployment.

This phase often reveals issues that require resolution before proceeding. A validation study that shows unexpected performance on certain patient subgroups might trigger additional investigation. User acceptance testing might reveal workflow problems that need resolution.

Budget three to six months for validation and approval. If significant issues emerge, add more.

Change Management Is Underestimated

Clinical AI implementation isn’t just technology—it’s workflow change. That means:

Training. Clinical staff need to understand how to use the AI, what it does and doesn’t do, and how it fits their workflows. Training takes time to develop and deliver.

Communication. Affected staff need to know what’s coming and why. Clear communication takes planning and effort.

Workflow integration. Clinical processes may need modification to incorporate AI. This requires clinical input, documentation, and sometimes rostering or resourcing changes.

Champion development. Successful implementations have clinical champions who support adoption. Identifying and supporting champions takes time.

Organisations that treat AI implementation as purely technical underestimate change management needs and face adoption problems at go-live.

What Accelerates Timelines

Some factors that can help:

Governance readiness. Having AI governance structures already in place, rather than building them for each project.

Clear clinical ownership. Strong clinical sponsors who can drive decisions and resolve issues quickly.

IT maturity. Modern, well-documented clinical systems with standard interfaces.

Experienced partners. Working with AI consultants Melbourne or other experienced implementation partners who’ve done this before.

Realistic expectations. Leadership that understands implementation takes time and doesn’t create artificial pressure for faster timelines.

Phased rollout. Starting with a limited pilot rather than organisation-wide deployment.

What Slows Timelines

Factors that extend timelines:

Governance gaps. No existing AI governance framework, requiring policy development before project can proceed.

Contested clinical ownership. Multiple clinical areas with unclear accountability for the AI implementation.

Legacy technical environments. Old systems with poor integration capabilities, requiring extensive technical work.

Vendor immaturity. Vendors who overcommit and underdeliver, creating delays and rework.

Scope creep. Projects that keep expanding as additional requirements are identified.

Resource constraints. IT, clinical informatics, or clinical staff time not actually available for the project.

Realistic Planning Advice

Based on experience, here’s how I recommend planning clinical AI implementations:

Don’t promise specific go-live dates early. Until you’re through governance and contracting, the range of possible timelines is too wide for firm commitments.

Plan for 18-24 months minimum. For significant clinical AI (not just small pilots), plan for at least 18 months from initiation to full deployment.

Build contingency. Whatever timeline you estimate, add 30-50% contingency. Things will go wrong that you can’t predict.

Phase the rollout. Plan for pilot deployment before broader rollout. This allows learning and adjustment.

Track time carefully. Understand where time goes. This helps identify bottlenecks and improve future projects.

Celebrate incremental progress. Long implementations lose momentum. Recognise milestones along the way.

The Organisational Patience Problem

Many organisations struggle with implementation timelines because of misaligned expectations.

Executives may expect rapid returns on AI investment. Vendors often promise faster implementation than realistic. Technology press coverage suggests AI adoption is faster than it actually is.

Setting realistic expectations early helps. Educating leadership about typical healthcare AI timelines prevents unrealistic pressure. Comparing to other clinical system implementations (which also take years) provides context. I’ve heard similar feedback from AI consultants Brisbane—the organisations that succeed are those where leadership understands what realistic looks like.

Patience isn’t about accepting delays—it’s about planning accurately and executing well. Organisations that understand realistic timelines can plan accordingly and achieve better outcomes than those that set unrealistic expectations and constantly feel behind.

---

Dr. Rebecca Liu is a health informatics specialist and former Chief Clinical Information Officer. She advises healthcare organisations on clinical AI strategy and implementation.