Caliper Lab
Economic impact · Medical coding

Which parts of medical coding could AI take on?

The evidence-gathering fifth of the job. The steps that decide payment still need a coder.

PreviewEdition 1, Nov 2026Sectors: Healthcare
$2.0bn

a year in US hospital wages for documentation and coding work.

20%

of a coder's time sits in the one step where AI is ready today: finding the evidence.

7 of 8

steps where AI is not ready or not yet tested.

AI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidenceAI ready: finding the evidencePartialPartialPartialPartialPartialPartialPartialPartialPartialPartialPartialPartialPartialPartialPartialNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot readyNot testedNot testedNot testedNot testedNot testedNot testedNot testedNot testedNot testedNot testedNot testedNot testedNot testedNot testedNot testedNot testedNot testedNot tested
Each dot is 1% of a coder's time
  • 20 AI ready: finding the evidence
  • 15 Partial
  • 47 Not ready
  • 18 Not tested

Why it matters. AI coding vendors price against the coder's salary. Most of that salary sits in steps AI cannot yet do, which caps the saving any buyer should model.

AI is ready for the evidence step; the payment judgments still need a coderShare of a coder's time in each step, by AI readiness
0%10%20%30%40%Find the evidenceFind the evidence: 20% (Ready)20% · ReadyQuery the physicianQuery the physician: 10% (Not tested)10% · Not testedCode the diagnosesCode the diagnoses: 25% (Not ready)25% · Not readyChoose the main diagnosisChoose the main diagnosis: 8% (Partial)8% · PartialPresent on admissionPresent on admission: 7% (Partial)7% · PartialCode the proceduresCode the procedures: 12% (Not ready)12% · Not readyGroup to the payment categoryGroup to the payment category: 8% (Not tested)8% · Not testedReview and sign offReview and sign off: 10% (Not ready)10% · Not ready

Source: Caliper Lab revenue cycle wage model; Caliper Lab medical coding benchmark, Edition 1.

What we did

We split inpatient coding into eight steps, allocated a coder's time across them, priced each from US wage data, and read AI's readiness at each step from our medical coding benchmark.

Full results

StepShare of coder timeAI readinessWhat the benchmark shows
Find the evidence20%ReadyEvery code cites a real line in the chart
Query the physician10%Not testedDrafts 3 to 5 queries per chart; usefulness not scored
Code the diagnoses25%Not readyFinds 38 to 44% of the coder's diagnoses
Choose the main diagnosis8%PartialRight on 69 to 84% of charts
Present on admission7%PartialRight where it answers; silent on half
Code the procedures12%Not readyAdds 7 to 10 times the procedures a coder records
Group to the payment category8%Not testedPlanned for Edition 2
Review and sign off10%Not ready0 to 1 of 55 charts passed untouched

How we measured it

  • Wage pool from US occupation wage data (BLS, May 2025) and task shares (O*NET) for hospital documentation and coding roles.
  • Readiness at each step read from our medical coding benchmark, Edition 1.

Limits

  • Step shares are our allocation of coder time; a measured time study will replace them.
  • The pool is labour cost. It is not a saving, and AI being ready does not mean the cost goes away.
Cite this study. Caliper Lab, "Hospital coding: cost and AI readiness by step", The Bench, Nov 2026. https://thecaliperlab.com/study-hospital-coding-costs.html

See everything we cover in healthcare.

AI in healthcare