Ellen Kim's Profile

Interesting idea. It sounds like they’re viewing this almost like a niche of consulting, as a third party service that you can pay for. But given the critical necessity of understanding the company’s business, culture, and data, I wonder how feasible this would be. My inclination is that it needs to be a team within a company, stable over the years, and with people who have a deep understanding of the business.

Nothing more dangerous than an AI algorithm without understanding the source data and what is driving the output.

Addendum:

Could the scores be “weaponized” and be used for other applications, like health insurance agreements or CMS reimbursements?

Interesting and (to be honest) slightly intimidating proposal. Challenging to quantify a worker’s quality (physician, nurse, phlebotomist, janitor, everyone)! Hard to get data and the necessary surrounding data. A classic example is patient outcomes without accounting for their comorbidities. Patient survey scores, without patient demographics, SES, education level, health literacy? Phlebotomy success rates without accounting for patient access, age, comorbidities (central line? dialysis patient?). Restroom cleanliness scores without time of day, nearby patient volume, hospital/clinic location, accessibility?). Which, if any, of these scores would be significantly associated with the patient experience and patient outcomes?

Anecdotally, some surgeons track their volume, outcomes, and complication rates (on their own or as part of their department’s quality metrics) and tell patients who are “doctor shopping.” But is that the only metric that matters? How accurate is it?

I worry about “garbage in, garbage out” and the potential to limit access to care or damage the patient-healthcare system relationship and trust. If patients in a certain region don’t have access to a 5 out of 5-star provider/hospital, but don’t have the resources to go to a different provider/hospital, then will they lack trust and start to question their quality of care? Confirmation bias for people with higher scores who may not actually be good or become complacent? Will people with lower scores be fired or required to remediate certain skills? What if there is a lack of healthcare workers (or at least, a geographic distribution issue with supply) — is it better to hire someone with a low score or to be understaffed?

Clearly I have a gut response to this, and in part because you seem so enthusiastic about it without acknowledging some of the challenges. Would be very interested in learning more about [potential] applications in the US and internationally.

Nice summary of the article first followed by your thoughts.

Like we’ve discussed in some of our classes, I think especially initially it’s best to work with some of the stakeholders and people closest to data production (top, bottom, and everyone in the middle) at every step of the process (ideation, data collection, analysis, interpretation), so they see with transparency what you’re doing and are invested in the result. As different departments gain experience with the people analytics team, you start to gain their trust and interest, and you gain confidence in the team as well, then you can start to a little more independently.

You made a great point in the last paragraph. It can often be harder to point to specific gains or usefulness. If you have a few good experiences with different departments, they may voluntarily champion your value to the higher ups.