Optimizing Incentives: An Update from Gauntlet

Optimizing Incentives: An Update from Gauntlet

Hello all! Gauntlet has been optimizing Sushi’s incentives over the past year. Let’s take a look at the progress made to date.

Sushi Before Gauntlet

In May 2021, the Sushi team was juggling the management of a large incentives budget with other critical development workstreams. At the time, the Onsen program allocated incentives across pools using a fixed tiered allocation structure based on market cap.

The Sushi team observed that the return on liquidity incentives to a given pool was highly volatile, changing rapidly with market conditions. This, coupled with early Onsen program success and expanding demand, led Sushi to reach out to Gauntlet to develop a continuous, data-driven allocation approach.

How Gauntlet Built Sushi’s Optimization Models

Gauntlet’s data science and engineering team worked with the Sushi team to deploy the optimization engine. The objective function was to reduce incentives for inelastic pools and increase incentives for elastic pools to maximize the impact of incentives.

  • Data integration from a range of public and private sources to form a complete picture of the asset exchange.
  • Modeling using ML and proven optimization methods to identify optimal rewards allocation. Models measure trading volume elasticity with respect to incentives.
  • Optimization by delivering recommended parameter changes directly to mainnet. The impact of incentive changes are fed back into the models, forming a PID controller.

Results: Increased ROI and Rewards Efficiency

Gauntlet shared weekly recommendations with the Sushi team. Over the past year, Sushi implemented 612 recommendations from Gauntlet across multiple pools.

When analyzing token emissions, a common assumption is that there is a strong correlation between incentives, liquidity, and volume. Across numerous pools, Gauntlet’s models take into account dynamic conditions to provide visibility into the drivers of returns on incentives when compared to this naive approach.

As an example, the charts below show heuristics protocols would typically monitor to assess productivity. Based on this, this particular pool, MANA-WETH, looks in line with average productivity and shows a correlation between incentives, liquidity, and volume:

Average in the above chart is the 7 day rolling average metric for all supported pools

However, if we drill in further, we see that this pool is inelastic, and incentives don’t have much impact on volume or liquidity:

As a result of Gauntlet’s recommendations, along with an overall reduction in incentive spending and market dynamics, Sushiswap’s ROI (revenue/incentive spend) increased from 10% ($0.10 earned for every dollar spent) to 133% ($1.33 earned for every dollar spent) and projected annual revenue increased by $18.3M.

What’s Next

To date, optimization efforts have been targeted to the Ethereum mainnet. Over the past year, Sushi has expanded to numerous additional chains. Adding to the multi-chain complexity is the proposed Sushi 2.0 restructure and important considerations regarding the end of DAO funded emissions.

Gauntlet’s Incentive Optimization team has been conducting research to expand the engagement scope to serve Sushi more holistically. For the near future, incentives will continue to be Sushi’s largest expense. The value of optimizing those incentives remains critical for Sushi.

Looking forward, Gauntlet stands ready to collaborate with Sushi core and community on the road ahead. We are happy to discuss and answer any questions! :sushi:


Thank you for this. Quite interesting. Who do you liaise with at Sushi currently on this?

Look forward to continued optimization for the platform.

Me :grin:

Not gonna lie onsen was a nightmare to mange and even more of of a nightmare trying to figure out what the right number of sushi per day was regularly for all the pairs we’ve added to onsen over the last year.

Gauntlet helped alleviate all of that right away, and their models have improved a ton of over time as well. Was really fun and I’ve really enjoyed working with them through this process of building out and improving the models as well.


Glad you enjoyed it. We mainly work with Jiro along with a few others on the core team.

  1. What is the scope of the revenue you are referring to? Total revenue or revenue generated by incentivised pools? Am I right to assume that it would be more logic to assess the quality / impact of Gauntlet’s solution within the scope of incentivised pools?

  2. Do you have the breakdown for every pool the model has been working on? Would a Normal distribution (Gaussian) representation be relevant to help the community understand the impact of the model on the whole sample?

  3. Imo, it would be interesting to ‘understand’ the ROI, all things being equal (i.e. net of volatility). My point is that, would it not be relevant to have a volatility function represented in the last graph? To what extent the ROI has been fueled / improved by volatility vs. Gauntlet’s solution?

Do you have a Time series with the incentive budget (emission) before vs. after Gauntlet’s incentive takeover?


Revenue refers to revenue generated by incentivized pools we made recommendations for.

You can visualize our estimated elasticities for all pools as a histogram as of 5/2. The overall takeaway here is that many pools even today remain inelastic and can continue to be targeted for incentive reductions.

Great callout. Multiple factors influenced the ROI increase. One large contributing factor was an overall decrease in budget, which is something we’ve been working on internally and have begun to scope out with the core Sushi team.

Yes: https://i.imgur.com/MPvi0tk.png

1 Like

What else ya got in there ?