Eric J Ma's Website

A while back, I was part of a team that was going in circles. We had three options in front of us, and every discussion round ended the same way: people advocating for their preferred option in words, others pushing back in words, and no convergence in sight. There was no slam dunk benefit to any one option, hence the loop. As a team member (not the team lead), I started to feel that we needed a different way to decide than continuing to talk in qualitative terms. The Bayesian in me piped up: let's express our beliefs as probability distributions, and decide from there.

I'm pretty sure I'm not the inventor of this trick; someone out there has probably formalized it under a fancier name. But I did arrive at it that day while watching my team talk past each other, and it worked well enough that I've wanted to write it up ever since. I call it funny money allocation.

The mechanics

Here is what I did. I gave every person 1,000 points of funny money, effectively \$1,000 each, and laid out the ground rules:

• There are three options on the table.
• Allocate your 1,000 points across them however you like: all of it on one outcome, or any split you want.
• You also don't have to spend all 1,000 points. Express that you're unsure by spending fewer points and keeping the funny money for yourself.

That last rule matters more than it looks. Keeping points is really an allocation to a fourth outcome: "I don't know." It's a legitimate epistemic state, and the design makes room for it instead of forcing it to hide inside a pick. In a vote, uncertainty gets collapsed into a choice. Here, it gets its own allocation.

One ground rule, baked into the setup, does all the work: your total allocation can't exceed 1,000 points. Words are unbounded; declaring "I strongly prefer option two!" costs nothing. Points are capped. The moment you have to take points away from one option to fund another, you have to rank your own convictions. And once you divide each allocation by 1,000, you have exactly what I was after: a probability distribution over the outcomes, one per person.

What the bets revealed

When the points came in, the exercise turned out to be very revealing.

I bet small: 70 points out of my 1,000, keeping the other 930. I genuinely wasn't knowledgeable about the underlying biology separating the outcomes, and my allocation said so out loud. Others bet big: 900 points on one option, leaving just 100 split between the remaining two. Where I was shrugging, they were leaning hard.

That 70-versus-900 spread was exactly the information we needed. Pool everyone's allocations and you get the team's collective distribution: which option had the most support, how concentrated that support was, and who held it. The people with the deepest expertise had put their (funny) money where their mouths were, and their bets helped us see very clearly where the team's preferences lay.

This is analogous to the kind of thing my former colleague Clayton Springer would have prescribed. His version: state your hypothesis in the form of a molecule, because that's concrete. Ours: state your beliefs in the form of a funny money distribution, because we were also trying to be very concrete. Either way, the point is to make your belief inspectable.

From bets to a decision

Beliefs expressed as probability distributions made real discussion possible. Once they were on the table, the argument shifted from "my adjectives vs. your adjectives" to "why did you allocate that way?", and the second question is much more productive. The exercise also surfaced something we hadn't articulated: there was more than one legitimate way to converge.

Consensus was the path we took. We talked through the distributions and rounded off some of the numbers to make the final call easier to act on and to fit our actual sample numbers. Summing up the numbers like this actually allows those with strong opinions to sway and influence the decision. And that's the point: with beliefs expressed as distributions, the decision maker can get creative. Another path is to have the person who needs to make the call treat the funny money allocation as a quantitative measure of belief, marry it with the qualitative beliefs that people have (the articulated reasons for their allocations and such), do some behind-the-scenes work, and come to the final conclusion. Done that way, everybody feels as if they've had their opinion heard, even if the outcome may not be exactly what they would have prescribed.

Why this beats voting

I think this trick has real advantages over an outright vote.

A vote collapses your belief into a single choice: one person, one bit. Funny money lets me express the strength of my conviction. When I want to say "I lean this way, but only slightly", I bet 200 points instead of 900. Degree of opinion survives the aggregation instead of getting flattened into a binary.

Funny money also separates confidence from preference, which voting mashes together. "Low confidence, mild preference" looks like 150 points on my preferred option and the rest in my pocket. "High confidence" looks like a big, concentrated bet. Those are genuinely different epistemic states, and they deserve different representations.

Abstention, or rather deferring to experts, is built into the design of funny money allocation. Choosing not to bet on one of the three outcomes, or choosing not to bet heavily on one of the three outcomes, is a visible and legitimate stance that says, "I'm willing to defer to the experts." You can participate honestly while being genuinely uncertain, and your honesty stays visible in the pooled distribution.

If you run one

A few practical notes, based on four or five successful runs and some reflection since:

• Collect allocations independently before sharing them. Anchoring is real: once the first big bet lands, everyone else's distribution shifts toward it.
• Have whoever spent the fewest points kick off the reveal, then build upwards to the folks with the most conviction, because anchoring will happen. Power dynamics matter too: anyone in a position of authority should go absolutely last.
• Sum the allocations publicly, then talk. The pooled distribution tells you where the room stands; the conversation after the reveal is where the decision actually gets made.

The second point gets to the heart of the problem. A qualitative shouting match becomes a set of quantitative statements of belief, and the team can begin listening to why someone has a certain strength of belief. It's the same instinct as going Bayesian with your data analysis: qualitative hand-flagging of opinions, like qualitative hand-flagging of outliers, gives way to something more principled. Next time your team is going in circles, hand out a thousand points of funny money and see what distributions show up. I suspect you'll be surprised!