# Approval is what lets you hand over more work | BYOM blog

URL: https://byom.co/blog/why-approval-is-the-product  
Markdown: https://byom.co/blog/why-approval-is-the-product.md  
Last updated: 2026-10-02

> Surgery, aviation, banking and software teams found that a visible check before an irreversible step is what made delegation safe. What the evidence says, and where it is thin.

By Kina (Checked by the BYOM team). Published 2026-09-17. 9 minute read. Series: In control.

## Key takeaways

- The WHO surgical checklist study in eight hospitals found major complications fell from 11% to 7% and inpatient deaths from 1.5% to 0.8%, though it was a before and after design.
- Code review research from Microsoft suggests the main payoff of a second pair of eyes is shared knowledge and awareness as well as caught defects.
- Banks use maker checker rules so that one person creates a transaction and another confirms it, and a retailer pricing error shows what happens without one.

A common worry about handing work to software is that approval slows everything down. The record from other fields points the other way. Checks placed at the right moment are what let busy, skilled people delegate more, because the cost of a mistake is capped before it happens. This post goes through the evidence from medicine, aviation, banking and software, notes where it is weaker than its reputation, and ends with a retail pricing error that shows the alternative.

## Surgery: a checklist at three moments

In January 2009 the Harvard Gazette reported a study in the New England Journal of Medicine, led by researchers at the Harvard School of Public Health in collaboration with the World Health Organization. It analysed 7,688 patients in eight hospitals, in Seattle, Toronto, London, Auckland, Amman, New Delhi, Manila and Ifakara. Of those, 3,733 were treated before the checklist was introduced and 3,955 after, between October 2007 and September 2008.

The checklist is a single page that takes minutes to complete. It is used at three points: before anaesthesia is given, before the skin incision, and before the patient leaves the operating room. It covers safe anaesthesia, preventive measures against infection, effective teamwork and other core operative practices. Major complications fell from 11 per cent to 7 per cent. Inpatient deaths after major surgery fell from 1.5 per cent to 0.8 per cent.

- **11% to 7%** major complications, before and after the checklist (Harvard Gazette, 2009)
- **1.5% to 0.8%** inpatient deaths after major surgery (Harvard Gazette, 2009)
- **19** items on the WHO checklist, at three pause points (World Health Organization, 2026)

The study's senior author, the surgeon Atul Gawande, was quoted saying the results were startling, and that they indicated gaps in teamwork and safety practices in surgery that are substantial in countries both rich and poor. The Gazette reported that the improvement was of similar size in high income and lower income settings. Donald Berwick, then president of the Institute for Healthcare Improvement, said he could not recall a clinical care innovation in the previous 30 years that had shown results of that magnitude. The lead author, Alex Haynes, said even many clinicians who had been sceptical became advocates once they saw the benefits.

Read the design before the headline. This was a before and after comparison in the same hospitals, not a trial with a separate control group, so some of the change may come from other things happening at the same time, including the attention that comes with being studied. The World Health Organization's page, on the other hand, says the checklist has gone on to show significant reduction in both morbidity and mortality, that it is used by a majority of surgical providers around the world, and that more than 500 professional organisations and institutions have endorsed it. The evidence is encouraging and consistent in direction, and the size of the effect in any one place will vary.

What matters for delegation is the shape of the check, and the Gazette's own description makes it clear. The tool is one page, takes minutes, and is used by the whole team rather than by one person reading silently. It does not ask the surgeon to be less skilled or less trusted. It stops the team at defined moments, before steps that cannot be undone, and asks a short set of questions. The Gazette also reported that sceptics came round after seeing it work, which is a useful reminder that people accept a check more readily when it is brief and when its value shows up in their own results.

## Aviation: the check that came after a crash

Flight Safety Australia traces the pilot's checklist to a crash on 30 October 1935. The Boeing Model 299 prototype bomber took off at Wright Field, pitched up, climbed to about 200 feet, stalled, crashed and caught fire. The cause was gust locks that had been left engaged. The chief test pilot, Major Ployer Hill, and the Boeing test pilot Leslie Tower died of their burns. The article says Hill or his co pilot overlooked the locks, and that Tower, in an observer's seat, tried to release them during takeoff but could not reach them.

Skilled pilots missed a critical item under operational pressure. The Army Air Corps went on to adopt the bomber, and the article says it trained inexperienced civilian pilots to fly it using a series of checklists. More than 12,500 of the aircraft were built during the Second World War. The source does not credit the checklist alone for that record, and neither should we. The reasonable reading is narrower: when a task is too complex to rely on memory, a written check before the irreversible step lets more people operate it safely.

The same article is candid about limits. It cites a study of 60 flights that found 899 deviations, 22 per cent of them related to checklist use. It notes that Spanair Flight 5022 crashed with 154 of 172 on board killed when the crew failed to notice that the flaps were not set, despite several checklists. A checklist that is run by rote is not a control. It works when someone actually looks at the thing being checked.

The article also quotes Asaf Degani on why the check sits where it does. Takeoff, approach and landing make up only 27 per cent of flight time, he is quoted as saying, yet account for 76.3 per cent of accidents. The checks cluster around the few moments where an error is hardest to recover from, which is the same logic as the surgical pause points. The article gives a taste of how far this went in practice: the Convair B 36 needed six hours of ground crew preparation and then another hour for a preflight check of 600 items. That is far heavier than the surgical checklist's 19 items, and a reminder that the right length depends on the stakes and the complexity of the thing being checked.

The Gazette's account of adoption shows how quickly a well placed check can spread. The Institute for Healthcare Improvement announced a rapid implementation sprint across the 4,000 hospitals in its 5 Million Lives Campaign, which it described as two thirds of American hospitals. The United Kingdom, Ireland, Jordan and the Philippines had already set up national programmes for every operating room. The researchers estimated that using the checklist in all American operating rooms would save 15 billion to 25 billion dollars a year through preventing major complications. These are projections and not measured outcomes, and the article should be read that way. What they show is that a check which costs minutes was judged worth a national effort.

## Banking: maker and checker

Banks have a name for the same idea. In a maker checker arrangement, also called four eyes, at least two individuals take part in each transaction. In the words of the Wikipedia entry on the principle, one individual may create a transaction, and the other should be involved in confirming or authorising it. It is used by financial institutions and in the information systems that process their transactions, and it implements segregation of duties.

The entry lists the benefits as fraud prevention through dual authorisation, error detection through independent review, an audit trail for compliance, stronger internal control and shared accountability. It also lists the costs honestly: slower processing, the expense of needing two people, bottlenecks if an approver is unavailable and administrative complexity. Every approval design has to live with that last set. The way to live with it is to apply the second person to the steps that matter, not to every keystroke.

What these arrangements have in common is that the second person sees the specific item before it takes effect. A policy document that says staff should be careful does not do that.

## Software: what code review is really for

Software teams adopted a version of this, the code review, long before AI wrote much code. Researchers have asked what it delivers. A 2013 study by Christian Bird and Alberto Bacchelli, published at the International Conference on Software Engineering and hosted by Microsoft Research, used observations, interviews and surveys of developers and managers at Microsoft, and analysed hundreds of review comments. Defect finding was the main motivation for review, but the study found that reviews are less about defects than expected. They delivered more in knowledge dissemination among team members, better visibility into what the project was doing, and alternative approaches to problems.

The same paper concluded that understanding the code and the change is the key aspect of reviewing, and that developers use many strategies to get that understanding, most of which current tools do not support. Applied to approvals, the point is plain: a reviewer who cannot see what a change does cannot meaningfully approve it, however many boxes there are to tick.

Google's case study, by Caitlin Sadowski and colleagues, drew on 12 interviews, a survey with 44 respondents and the analysis of review logs for 9 million reviewed changes. Its abstract describes lightweight, tool based review that Google adopted widely, and examines why it is used and how well it works.

There are two lessons for anyone approving AI work. First, a review catches fewer defects than people assume, so keep your other checks in place. Second, review does other jobs that matter: the approver learns what is changing in the store, and the team knows who changed what. Those effects accumulate, and they are part of why delegation gets easier with time.

## What it looks like when nobody checks

In late November 2025 the Italian retailer MediaWorld, as reported by SlashGear, listed iPad Air tablets at 17 euros instead of the regular 1,012 euros. Customers bought them online and collected them in store. The report says the error went unnoticed for 11 days, after which the retailer emailed customers with three options: return the iPad for a 17 euro refund plus a 23 euro discount on a future purchase, pay about 821 euros to keep it, or ignore the message. The retailer's position was that Italian regulations allow it to void sales where the price was obviously an error. The article notes that those regulations may have required contact by post and not email, and says it was unclear how the retailer would proceed. This is a news report, and the legal position is the company's own claim.

The article adds earlier examples of the same kind: Zappos losing 1.6 million dollars in 2010, and Anthropologie cancelling orders for an 8,000 dollar couch that had been listed at zero in 2017. In each, a price was wrong in public before the retailer noticed. The report does not say how the price was entered, so we cannot say which step failed. What it does show is that the wrong number reached customers and stayed there for 11 days. A check that takes seconds would have shown a tablet priced at under two per cent of its normal figure.

## What this means for delegating to software

Put the evidence side by side and the pattern is consistent. In each field the check comes before an irreversible step, it is short, it is performed by a different person or at a defined moment, and it is visible afterwards. None of them depends on the doer promising to be careful. They also pay off most where volume is high, since a routine step costs seconds and one error can cost far more.

For a store, the practical test of an approval design is whether it makes you willing to delegate more. If the approval is so heavy that you start to dread it, you will delegate less. If it is so light that you stop reading, you are exposed to the pricing error. The aim is a step that shows what will change, where and why, so that a decision takes seconds and means something.

## Where BYOM fits

In BYOM, a change Kina wants to make to a connected system is an Action. It waits in your approvals with what will change, where, and why. You stay the approver. Kina stays accountable.

The product page: [See approvals](https://byom.co/approvals), [See trust](https://byom.co/trust).

## Sources

- [Harvard Gazette, surgical safety checklist drops deaths and complications, 2009](https://news.harvard.edu/gazette/story/2009/01/surgical-safety-checklist-drops-deaths-and-complications-by-more-than-one-third/)
- [World Health Organization, safe surgery tools and resources, 2026](https://www.who.int/teams/integrated-health-services/patient-safety/research/safe-surgery/tool-and-resources)
- [Flight Safety Australia, a brief history of the checklist, 2018](https://flightsafetyaustralia.com/2018/11/one-thing-at-a-time-a-brief-history-of-the-checklist/)
- [Microsoft Research, expectations, outcomes and challenges of modern code review, 2013](https://www.microsoft.com/en-us/research/publication/expectations-outcomes-and-challenges-of-modern-code-review/)
- [Google Research, modern code review, a case study at Google, 2018](https://research.google/pubs/modern-code-review-a-case-study-at-google/)
- [Wikipedia, maker checker, 2026](https://en.wikipedia.org/wiki/Maker-checker)
- [SlashGear, MediaWorld iPad pricing error statement, 2025](https://www.slashgear.com/2036711/mediaworld-17-dollar-ipad-accident-statement)