@lhochstein "It's complicated" Having been in Amazon, Goog…

@lhochstein 1) "Fail closed" -- jargon for "optimised to say no by default".

Many processes in Google were like this, probably the largest other one being the hiring process. (Many Googlers hated the hiring process since they felt it said no to people who would 100% have been great hires...

@lhochstein ...sometimes being burned to the extent of refusing to do any more interviewing.)

But the promotion process is (was) the same way - if the committee couldn't convince itself the case was watertight, it wouldn't promote.

@lhochstein This obviously led to a situation where a lot of cases were refused and meant a lot of heartache and unnecessary departures, but also a situation where by god if you'd MADE that bar, it was close to undeniable.

There was a very strong shared sense of what each level meant.

@lhochstein Some more so than others: I think L5/Senior is probably the one most folks have the largest degree of precision on, and above that it's harder to tell. But one benefit was that title/grade inflation was very strongly counteracted by "the system".

@lhochstein Of course, because this is the real world, the benefit was also a drawback.

For example, minorities often struggled in this system (as elsewhere) for all obvious reasons. I was peripherally aware of a case where a minority achieved a fantastic business outcome, but did it ...

@lhochstein ... by being smart about what they persuaded other people to write, software-wise, and by actually writing much less code than was anticipated, meaning it all got done faster.

Great! Except the cttee was split since this was seen as relationship management, not tech execution.

@lhochstein There were a number of other cases where the promo was awarded but the tech bit was stripped in the promotion: so TPM level X -> PM level X+1, recommended transition to manager from IC, etc etc.

Funnily enough, those things had a habit of happening to women, and not to men...

@lhochstein ...but since women were such a small proportion and their aggregate statistics were fine (as far as we were told), there wasn't much appetite for fighting these micro-aggressions.

Particularly since they generally came with more money and respect, though not quite as much.

@lhochstein But, important point I'll come back to: since the process was (apparently) so thoroughly engineered to be objective, faceless, and unapproachable, it also served to launder terrible decisions.

@lhochstein I exaggerate a little, but not much. It was a manager-run process -- note, not manager-determined!

There were two main mechanisms for feedback: in the 'determination' phase, the cttee could ask for clarification. In the 'appeal' phase, the manager could supply new evidence.

@lhochstein Importantly and characteristically, they could not say 'your decision sux lolsob', they could only say 'the learned cttee has probably understandably overlooked this particular trifling piece of evidence'

*supplies huge info dump*

@lhochstein I partook in all of these as individual and manager and cttee member.

Overall the process was very much respected, in that there was a genuine perception title inflation was kept to a minimum and false positives were very low, but the sheer COST of the thing was spectacular.

@lhochstein I don't just mean flying people around to conduct the committees, though that was a part of it. The incentives were structured such that NOT having a piece of information going into a process was much more of a problem than having too much information (from the applicants PoV).

@lhochstein So there were strong motivations to make a hugely detailed case, taking ages to even parse, let alone understand.

(Poor cttee members, who were forced to read these huge documents, often without any expertise in the business area in question.)

@lhochstein But this comes to my second point - detail over big picture.

The system as a whole rewarded supplying mind-numbing amounts of detail, and hyper-focused on the stated attributes of each level. Yet often the detail actually hindered understanding whether or not G was benefitting.

@lhochstein Arguably that's good - G could benefit from an outcome distinct from whether or not the person's "operating at level X" - but it's also bad, particularly when the applicant is demonstrating flexibility of approach to make the right thing happen, aside from what the ladder says.

@lhochstein Eventually they introduced character (not word, IIRC!) limits for promo packs, so the packs started linking offsite to documents with more detail, the links were link-shortened, the English became more telegraphic and compressed, etc.

@lhochstein "Detail over big picture" was also part of a transformation in the performance process, where people's performance used to be assessed by a 39 point (IIRC?) scale. Eventually they threw that out and went for something more conventional.

@lhochstein The stated reason was false precision: the organisation spent a LOT of time tracking in huge detail the performance curves of people within these buckets, but that extra detail didn't actually amount to much in the way of changed outcomes.

@lhochstein It was uncharacteristic for Google (then) to forswear extra information that could theoretically result in a finer, more granulated decision, so I felt that was a genuinely pragmatic (and correct) change.

Normal Google culture wouldn't have done it, and there were complaints.

@lhochstein I kept waiting for the corresponding change in the promotion process, but it hadn't by the time I left.

@lhochstein However, and this brings me to my third point, the system as a whole opted to favour written, mostly-kinda-sorta-objective measurements of output over other things.

This works well for (say) software engineers and SREs up to (say) Senior, or thereabouts.

@lhochstein But there are a whole bunch of "edge cases" (some of them majoritarian, actually) it didn't handle well.

In particular, higher-level promotions, because of the "have to find someone of level X+1 to comment" requirement, meant a radically shrinking pool to fish from for support.

@lhochstein It's crucially important -- I've had this lesson repeatedly over the past, say, five years -- to manage relationships at that level. Obviously not just for promo support. For getting your actual job done!

@lhochstein Yes, sure, "what do you, leader of dept N, want from the teams supporting you in dept M". That's basic. But also, more profoundly, "Who are you? How do you see the world?"

Others use that to frame things as they think the other wants to see it. But it doesn't have to spin.

@lhochstein It can be a genuine recognition that businesses are run by people, and people are put in place by other people just like APIs are put in place by a programmer: to manage complexity. To hide information. To allow details to be forgotten to make the bigger picture easier to grok.

@lhochstein And that, ultimately, was where I felt the Google promo process fell down the hardest (though it fell down hard elsewhere, and also succeeded elsewhere) -- the behaviours it rewarded were not the ultimate behaviours which the business needed.

@lhochstein The business needed folks who could write detail, but also understand when to discard it.
It needed folks who could manage relationships just as much as written records.
Folks who felt supported in taking risks in a process designed to say no.

@lhochstein Ultimately, though I definitely appreciated the output of the process from a company-wide point of view, I felt it was slowly-but-surely weeding out people with very different views & skills that the company would find itself regretting - eventually.

@lhochstein I don't know the changes which have been made since I've left - I wonder if they line up against the problems I remember, and outline here.

Perhaps in the future, someone will design the perfect process; this clearly ain't it, but is anything else?

THREAD ENDS (thankfully)