Cigna's $200M AI savings projection runs to about 2.4 basis points of a year's revenue

Take the $200 million apart before deciding what it proves. Spread over three years it is about $66.7 million a year ^[1], and Cigna books $275 billion of revenue annually ^[3], so the annual saving is roughly 0.024 percent of the top line, call it two and a half basis points ^[2]. Against national health spending above $5 trillion ^[10], of which Cigna's own revenue is about 5.5 percent ^[5], the figure disappears entirely. That does not make the program a bad idea. It does mean the number's job is reputational rather than financial: it exists to be quoted, in roughly the way I am quoting it.

The harder gap is on the other side of the ledger. Cigna's disclosed AI money, $100 million through 2028, is aimed at clinician documentation time and prescription speed ^[2], which is a different program from the condition-detection work carrying the savings projection ^[1]. So the headline benefit arrives with no cost beside it, and a benefit without a cost basis is a savings claim rather than a return. Worth noting where the committed dollars actually point: administrative throughput, not the outcome model that produced the press number.

Projected avoided cost is the softest currency in this industry, because the counterfactual is a modeled group of members whose cancer or kidney disease was caught early ^[1] and whose later claims therefore never existed to be counted. Actuaries do this work honestly and routinely. It still cannot be tied to a line in the cash statement, which is why "lead with measurement" ^[7] deserves to be tested against the company's own second example.

That example is the biosimilar campaign, or rather the more interesting version of the same question, because it is checkable. Cigna mined thousands of past member conversations about biologics and biosimilars, rewrote its digital messaging, and reports that more than 80 percent of the people it reached chose the biosimilar ^[5]. Humira can cost a patient $7,000 a month ^[4], which is $84,000 a year ^[3]. Andresen puts the result at a couple hundred million dollars of savings for patients, plus a lot more margin for Cigna ^[6]. Take the low end of that as $200 million and it equals the entire annual branded cost of about 2,381 patients ^[4], so unless the horizon is longer than a year, the real switcher count has to be considerably larger, because a biosimilar is not free ^[11]. Every input in that calculation sits on a price list and a prescription count. Someone outside the company could run it.

There are three ways this goes. Cigna publishes the cohort method and the spend behind the $200 million, and enterprise health AI gets a benchmark other payers can be measured against. Or it stays a press figure, and buyers keep discounting every such figure to zero, which is what they do now. Or the pharmacy mix does the persuading instead, because drug spend moves visibly and within a quarter. My view, and it is probably wrong on timing rather than direction, is the third: the number that changes how anyone models this is the one attached to a switched prescription, not the one attached to a prevented disease. What would change my mind is the $200 million appearing inside medical cost guidance with a stated method, or the 80 percent turning out to describe a few hundred contacted members ^[5], which would shrink the auditable case to an anecdote.