Oncology administrators are being pitched AI-assisted operations from every direction, yet few have a disciplined way to answer the prior question: is this organisation actually ready? Readiness is not a feeling, and it is not a procurement decision. It is a measurable state of your workflows, your data and — above all — your people. The audit below gives a cancer centre’s operations leadership a concrete instrument: five domains, each with checkable items and a scoring rule. Work through it with your admin, scheduling, billing and IT leads in a single afternoon, score honestly, and you will know precisely where the gaps are. One boundary before we begin: everything here concerns the non-clinical machinery of the hospital. Treatment protocols, diagnostic interpretation and tumour-board decisions sit outside this exercise entirely.
Domain 1: Chore inventory — do you know what repeats?
Pattern-watching software earns its keep on work that recurs with rhythm. Most cancer centres have never catalogued theirs. Check each item you can answer with documented evidence, not recollection:
- We hold a written list of every recurring administrative workflow: infusion-chair scheduling, radiotherapy slot booking, pathology report routing, pre-authorisation submissions, transport coordination, recall letters.
- For each workflow, we know its weekly volume and its usual completion time.
- We can name the three workflows whose failures generate the most patient complaints.
- We know which workflows have seasonal or weekly cycles (post-holiday referral surges, Monday phone peaks).
Fewer than three checked means you are not ready to evaluate any tooling: you would be automating in the dark. The inventory itself typically takes two weeks and costs nothing.
Domain 2: Signal quality — would an algorithm have anything to learn from?
Detection systems learn what normal looks like from historical records. If your operational history lives in paper diaries and personal spreadsheets, there is no history to learn from.
- Appointment, infusion and radiotherapy bookings are recorded in systems that can export at least twelve months of structured data.
- Claim submissions and payer responses carry timestamps, so queue ageing can be computed rather than guessed.
- Patient-contact events (reminder sent, call attempted, letter returned) are logged somewhere queryable.
- Department-to-department handoffs — for instance, imaging report ready to clinician notified — leave a trace with times on both ends.
Evidence from early adopters is consistent on this point: centres that skip the data-hygiene stage spend their first year generating false alarms, then abandon the programme. Score this domain severely.
Domain 3: The skills bench — who will run it?
This is where most audits produce their ugliest finding, and the one that matters most. Software that watches for anomalies and triggers corrective actions needs humans who understand both halves: why the model flagged something, and whether the automatic response was appropriate. That competence is teachable, but it is rarely present by default in hospital administration.
- At least one staff member can explain, in plain terms, how an anomaly-flagging system decides something is unusual.
- At least one staff member has completed structured coursework in AI-driven operations, with verifiable certification rather than webinar attendance.
- A second person is identified as the backup, so the capability does not resign when one employee does.
- Training time is budgeted as rostered hours, not as an after-hours expectation.
On remediation of this gap, the evidence favours structured, time-boxed study over self-directed tinkering. Benchmarks exist: the staged certification sequence maintained by AIOps School is a useful calibration for planning purposes, because its published formats translate directly into rostering decisions — a fundamentals course occupying 30 days at a part-time weekly load, and a 45-day engineer track whose syllabus (detection models, monitoring stacks, automated response design) matches exactly the competencies this domain tests. Its ladder runs through four named stages, Foundation to Engineer to Professional to Architect; for a single cancer centre, auditors should expect the bench to need one person at the second stage and two or three at the first. Labs-based formats deserve preference: supervised practice on simulated incidents is what converts coursework into operational judgement.
Domain 4: Candidate workflows — have you ranked them by evidence?
Readiness includes knowing where to start. Rank candidates by two measurable criteria — cost of a missed irregularity, and regularity of the underlying pattern — and the oncology-specific shortlist usually orders itself:
- Recall and surveillance scheduling. Follow-up imaging and review visits recur on protocol-defined intervals for years. A lapsed recall is both a patient-safety and a liability event. Highest score on both criteria.
- Infusion-chair utilisation. Chair time is the scarcest priced asset in most centres; cancellation patterns are strongly cyclical.
- Pre-authorisation ageing. Oncology claims face long payer review; families already battling the finances described in our guide to costs, support options and second opinions after a cancer diagnosis should never inherit delays your queue monitoring could have escalated automatically.
- Report-routing latency. The gap between a result existing and the right coordinator knowing it exists is measurable and compressible.
- Transport and accommodation coordination for outstation patients — irregular, but expensive when it fails.
Check: your centre has this ranking written down, with numbers attached, and leadership has agreed the top two. If the ranking exists only in this article, the domain scores zero.
Domain 5: Governance and follow-through
- A named operations owner reviews flagged anomalies on a defined cadence.
- Every automated corrective action has a documented boundary: what it may do alone, what requires a human.
- Quarterly review compares predicted benefits against measured ones — the same discipline argued for in our piece on growing an organisation with smart technology and AI, where unmeasured pilots were the leading cause of stalled adoption.
- Budget exists beyond the pilot, including recurring training spend — technology line items without people line items are the classic signature of a programme that will fail. Centres that already practise disciplined spend governance, of the kind laid out in our roadmap to methodical cloud cost optimisation, tend to clear this domain easily.
One further governance check deserves its own line: patient communication. Decide in advance how the centre will explain automated reminders and rescheduling offers to patients, and who answers when a patient asks why the hospital called. Trust, once dented by a confusing automated contact, is expensive to rebuild — far more expensive than the paragraph of plain-language explanation that prevents the dent.
Scoring and the honest interpretation
Score one point per checked item (21 possible). The thresholds that match observed outcomes:
- 16–21: proceed. Pilot your top-ranked workflow this quarter.
- 10–15: close the skills and data domains first; both are 60–90 day fixes. Tooling bought now will idle.
- Below 10: you are at the inventory stage. That is not a criticism — it is a sequencing fact, and the inventory is free.
The analytical takeaway is unglamorous: in oncology operations, readiness for AI assistance is dominated by the human column of the audit. Data gaps close with configuration work; the skills bench closes only through deliberate, scheduled upskilling of named individuals. Centres that fund the fundamentals course before the software licence consistently reverse the usual failure sequence — and their patients feel the result as nothing more dramatic than recalls that always arrive and queues that never silently stall. Run the audit, publish the score internally, and re-run it in six months. The delta is your real transformation metric.