Also, it doesn’t take into account skilled/hard-to-hire roles that are effectively cost centers, like security. How much revenue am I responsible for as a security engineer? Probably none, especially since the consequences for a breach are usually negligible.

It is most definitely not impossible. How do you think companies decide on budgets? They're not just guessing; they're predicting how different projects will impact revenue. Alexa isn't important to Amazon because it's "cool" — it's important because they've put an actual dollar amount to its return on investment.

> How do you think companies decide on budgets?

With great effort and generally mediocre results. Figuring out the contribution of an individual project is hard. An individual person even more so. And this isn't because we're not smart enough, it's because the idea of an individual contribution when the whole is greater than the sum of its parts is not coherent.

So how do you split the returns? Well, the entire field of cooperative game theory has something to say about that, and the answer is complicated but the gist is that it depends less on contribution and more on negotiating power.

Are you willing and able to just leave if you don't get what you want? Are you hard to replace? Can you do your stuff with a different company if you leave? All that is much more important than how much your presence benefits the common enterprise compared to the counterfactual where nobody did your job.

> How do you think companies decide on budgets?

From what I've seen, by horse-trading, and fighting highly political, bitter, zero-sum turf wars that depend way too much on how charming department heads/directors are, and how well they relate to the CxO's and/or board. Its not uncommon for the CEO's pet-project to get a budget completely disconnected from the its value (current or future).

Companies - like anything collective that relies on human judgement - are terrible at digging deeper than first-order effects; which is why sales people tend to get paid much more than the engineers who developed the product: "I closed a $X million deal, so I get Y% of it". The engineers get paid a flat rate, because it's impossible to quantify how much value they added, and when. Baseline engineer salaries are broadly determined by how much other companies are paying, and more specifically by how good the individual engineers are able to negotiate.

I have never seen these things come from anything other than guesswork. It's educated guesses, but guesses nonetheless. It's why products get killed too early or too late. Measuring these things properly and causally is hard and sometimes infeasible.

I make planning software at a quantitatively sophisticated tech company, and it’s quite ironic to me that we don’t have a precise understanding of the expected impact of our software on the company. But that’s normal. We don’t build software that will barely break even. We build software that we expect can be massively successful, and the resource allocation is more about opportunity cost and uncorrelated bets than predicted outcomes.

Companies that make low margin highly competitive products, like paper mills, are the ones with a rigorous understanding of the costs and benefits of every activity.

> it's important because they've put an actual dollar amount to its return on investment.

The issue isn't whether you can put a dollar amount, but on the confidence in accuracy. I've seen how this is done in my company: Construct a narrative, assign numbers to that narrative (with some justification that doesn't always involve real world data), and come up with a total. They are incentivized to inflate the numbers as their goal is to increase their budget - not be accurate. If all departments did this, and you add the numbers they come up with, we'd easily end up with a number 10x our annual revenue, which is ridiculous.

Some folks who are not vested in this will aim for accuracy. And even then, they will have a large confidence interval - easily off by 2x.

How are you going to go from there and estimate an employee's value? Even if the project's value was 100% accurate, you still need to break that down to each employee. And since most employees don't work independently, you need to model the interaction effects (my work depends on your work - if you perform poorly, my contribution to the bottom line is reduced).

If the estimate of the value of the project is off by 2x, then you've already got a pathetic lower bound on the interval length of your confidence interval.

Alexa is actually an easier case to model. My job is to improve the internal communications infrastructure so that important messages get delivered (IT announcements, CEO communications, etc). How will you model my contribution to the bottom line?

I can assure everyone: At any decent sized company, no one is trying to come up with your value to the company.

They don't need to put a dollar amount on ROI of Alexa to develop it. Alexa is a loyalty-generating machine. Even if you never use it to make a purchase, it helps the brand, and it keeps you using Amazon products.

On top of that, it does also provide a sales channel that nobody else has. Practically nobody else has a voice-activated robot sitting in people's kitchens that buys paper towels and plays Spotify. It's the difference between all your competitors making sales via telegraph, and you introducing a home telephone just to make orders with your company. Funding it is a no-brainer.

Hate to break it to you, but any accountant with any data savvy can you give a more or less accurate ballpark figure with the right details. (This is also why you should be chummy with Tim in Accounting; you'll be amazed the things you can pick up shooting the shit with the finance department).

BI, in fact, is almost entirely the artform of synthesizing the answer to that very question for the purposes of executive decisionmaking.