Innovation Through Deletion (ITD)™
The step the healthtech playbook skips. Before anything is built or bought, ITD defines the outcome, derives what the outcome actually requires, and sorts every existing step into one of four kinds of work. Then it deletes, and only then automates what is left.
Four kinds of work. Four different interventions.
Two questions place any step: does it need the judgment of the person doing it, and does it serve the defined outcome?
The classification only runs against a defined outcome with a measurable indicator. Without one, there is no reference point for what any step is for.
The sort takes judgment. The part of a note that carries care forward is load-bearing. The parts that exist for billing and legal defense are compensation. That is why ITD classifies specific steps against a specific outcome, not product categories.
Load-bearing
Serves the outcome directly. This is the work that should exist.
Diagnosis, treatment decisions, paying a valid claim correctly
AUGMENT WITH AIUnassigned
Real work nobody was given. It lands on whoever is closest.
Care coordination across settings, follow-up nobody owns
ASSIGN IT ON PURPOSECompensation
Covers for a failure upstream. Faster is not fixed.
Prior auth paperwork, appeals, re-keying data between systems
FIX UPSTREAM; KEEP THE CORESediment
Exists because it exists. The reason is gone.
Reports no one reads, duplicate attestations
DELETE ITFive steps, in order.
The healthtech playbook usually starts at the last one.
Define the outcome
Name what the system is supposed to produce, in measurable terms, without reference to how it is produced today. Define it with what the patient experiences, not only what the clinician and the payer need.
Derive the requirements
Work out what that outcome actually requires, then compare it with what is required today. The gap between the two is the first finding. It is usually large, and it is where the money is.
Classify and map
Sort every current step into load-bearing, compensation, sediment or unassigned. Record who introduced it and whether that person carries accountability for the result. EHR access logs show which fields and templates nobody downstream ever opens.
Delete by category
Fix the upstream failure behind each compensation step. Delete sediment. Assign unassigned work on purpose. Fix what you own; escalate what you don't. Delete in reversible pilots, with a clinical owner's sign-off.
Redesign and automate the core
Apply Lean, Six Sigma, Agile and AI to the load-bearing steps that survived. ITD does not replace these methods. It tells them where to apply.
Only assignment moves work into load-bearing.
Unassigned becomes load-bearing
It already serves the outcome. It fails because nobody owns it. Give it to the role whose judgment it needs, with matching accountability.
Compensation is separated
Fix the upstream failure and the workaround goes. Where a load-bearing core sits inside, such as the safety judgment in medication reconciliation, keep and strengthen it.
Sediment is deleted
It does not move. Watch for the counterfeit move: relabeling compensation as load-bearing. A prior authorization tool described as clinical decision support is still compensation work.
No accountability without authority.
The buyer signs the contract. The user works with the tool. The clinician carries the risk. The patient lives with the result. In healthcare they are rarely the same person. For any AI output a licensed professional is asked to sign, the rule becomes a four-part test.
Can they see what was recorded, which sources and model produced the output, and what the AI added or changed?
Can they correct, reject or override it without penalty or extra work?
Did the people who carry the license help choose the tool, set its limits and decide when to turn it off?
If it's wrong, who answers? If the vendor controls how the model behaves and changes, the vendor should carry a matching share.
Authority over an output you cannot trace is not authority.
Automating returns a share. Deleting returns all of it.
For any step: C is its full annual cost, s the share a tool saves, R the new review work the tool creates, V the vendor cost, D the downstream rework the step causes and F the cost of fixing the upstream failure.
Has a ceiling below C, and shrinks as review work grows.
Removes the whole cost, plus the rework it caused downstream.
| Automating the step as it is | Deleting first (ITD) | |
|---|---|---|
| What is saved | A share of the step's time or cost | The whole step, plus the rework it caused |
| New work created | Review, correction and sign-off of AI output | None from the deleted step |
| Who carries the risk | The licensed professional who signs | Realigned to whoever holds the authority |
| When the rule or payment changes | Value can disappear overnight | Value holds, because the outcome still has to be produced |
ITD is not anti-automation. Step 5 is automation. The point is sequence.
The method is published with a permanent DOI: 10.5281/zenodo.24122071.
Read the full paper.
The methodology, the prior authorization worked example, future-proofing and what ITD asks of each player.