Reward the idea, not the rank

A junior analyst might have a better idea about how your organization should use AI than your CEO does. Most reward systems can't handle that. Where innovation rewards exist at all, they tend to follow the org chart: if bonuses scale with seniority, the same idea might earn a partner $50,000 and a junior analyst $1,000 and a thank-you.

People notice. I think it costs organizations more than it seems to, because good AI ideas increasingly come from people close to the work, and those people are rarely senior.

What rank-tied rewards do

Four things tend to follow when rewards track rank. Junior people stop offering ideas because they can see it won't matter. Managers take credit for their teams' ideas, since that's where the reward goes. People propose only safe ideas that won't upset anyone. And the people generating the most value eventually leave for somewhere that notices.

The alternative is simple to state: the value of the idea decides the reward, whoever had it. I'd rather the best idea win the argument than the most senior person in the room.

Same idea, same reward

Suppose the organization pays $10,000 for a good idea. For a CEO that's a symbolic amount. For a junior analyst earning $50,000 a year it's a fifth of their salary. Paying both the same for the same quality of idea sends a clear message about what's valued: the thinking, whoever did it. If the analyst's idea is better than the CEO's, the analyst gets more.

The CEO shouldn't be exempt either. A small, public reward for a senior leader's good idea shows that everyone plays by the same rules.

How I'd set up the fund

I'd keep it separate from base pay and promotion, so it can move faster and doesn't get tangled in performance reviews. Start with a fixed pool, an amount you can afford to get wrong, and adjust it after the first year based on what it produced. If you want the fund to grow with the value it creates, tie later pools to profit.

Judge ideas on four things: impact (did it make money, save cost, improve productivity or reduce risk), novelty, adoption (is it actually being used) and how easily others can build on it. Leave out title, tenure and department.

Not every idea's value can be measured the same way, so I'd use tiers:

1. Measurable direct impact. If an idea saved a verifiable amount, the contributor gets a share of it, say 2 to 5%.
2. Adopted, but hard to quantify. A cross-functional committee (not just executives) estimates the value and rewards accordingly.
3. Promising, still in development. A smaller reward for reaching this stage, with more if it moves up a tier.
4. Valuable but not implemented. A nominal reward for thinking that helped, such as a risk spotted or a better idea sparked.

An illustration makes the first tier concrete. Say someone finds a workflow change that saves ten hours a week for each of 50 people. At an assumed $100 an hour fully loaded, over a 50-week year that comes to about $2.5 million. Two percent would be $50,000, which lands very differently from a mention in a team meeting.

Blind review, public credit

Anonymous submissions make judging fairer, but you can't reward someone you can't identify. So I'd keep ideas anonymous while they're evaluated, reveal the contributor once an idea is selected, and then celebrate them publicly. The celebration matters as much as the money, because it shows everyone else what gets rewarded.

Where this can go wrong

Calling a system merit-based doesn't make it fair. If anything, I'd expect a system that advertises itself as meritocratic to get less careful about bias, because people assume the label has taken care of it. The fix is to publish the criteria, keep the committee cross-functional, and check every cycle who is winning, by level, department and gender.

Money can also crowd out curiosity. Pay people for ideas and some will start chasing the payout rather than the problem. I'd keep the amounts meaningful without letting them dominate, and reward sharing and reuse as well as winning.

There are precedents for tying pay to peer judgement rather than hierarchy, though none tests this exact design. Valve's 2012 employee handbook describes a company with no managers where peers rank each other every year to adjust each person's compensation to their value, and it's candid about the costs, including that a poor hire can "go unchecked for too long" (Valve handbook). W. L. Gore describes working in a "lattice" without the constraints of traditional chains of command. Both are whole-company designs rather than an innovation fund, so I'd treat them as evidence that peer-based recognition can work, and no more than that.

Run it for a year

Define the pool and the criteria, pick a cross-functional committee, and explain the rules plainly, with examples of what qualifies. Then run it for a year: track submissions, pay the rewards, collect feedback and adjust. Celebrate the winners in public, especially the junior ones, because that's where the message lands.

This fits with the rest of how I think adoption should work. An AI budget funds the experiments, a shared register of solutions with stage gates feeds ideas into these tiers, and pushing AI decisions closer to the work only pays off if the rewards reach the same people.

The piece I haven't solved is valuing prevention. An idea that stops a security incident or a bad decision from happening can be worth the most and be the hardest to prove, and so far I don't have a better answer than a committee's honest estimate.