The studies behind the levers

Seventeen studies and reports, each with a link and what it found.

The rules of the lending board are not new ideas. Each one stands on research that anyone can read.

Every link opens free of charge, with no sign-up. Where the journal charges for an article, the link opens a free version, the authors' own copy or a working paper, and the line under the title says so; where only slides or a summary are free, it says that too.

Did the team cause the volume?

On the board: volume from a partner, a curator or an issuer counts at the share the team caused, measured with a holdout where one is possible, and a channel measured only by attribution gets a wide range.

1
Close Enough? A Large-Scale Exploration of Non-Experimental Approaches to Advertising Measurement
Brett Gordon, Robert Moakler and Florian Zettelmeyer · Marketing Science, 2023 · the link opens the free version on arXiv

In 663 advertising experiments at Facebook, ads lifted the early steps toward a purchase by a median 29 percent, but two common methods without a control group put the lift at 83 and 173 percent. Even with more than 5 000 data points for each user, neither method found the true effect reliably.

2
The Unfavorable Economics of Measuring the Returns to Advertising
Randall Lewis and Justin Rao · Quarterly Journal of Economics, 2015 · the link opens the free 2013 version, which has an earlier title

In 25 large online advertising experiments, the typical range around the measured return on spend was about 100 percentage points wide, because purchases differ so much from one person to the next. A clear answer usually needed more than 10 million person-weeks of data.

3
Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field Experiment
Thomas Blake, Chris Nosko and Steven Tadelis · Econometrica, 2015 · the link opens the authors' free copy

When eBay paused its ads on searches for its own name, the free search results won back about 99.5 percent of the lost clicks. For other keywords, simple regressions put the return on ad spend above 1 600 percent, while the experiment found a loss of 63 percent.

Does more selling pay, and when?

On the board: each extra unit of a programme brings fewer meetings than the one before, a new hire reaches full speed after 6 months, and deals that go live after the twelve months are shown in flight, not counted. From the plan to twice the plan, each 10 percent more budget brings about 7 percent more loans, against about 3 percent more sales in the meta-analysis: each win here is a large account, and a young protocol is still far from meeting every counterparty worth a meeting. Where that is not so, the next dollar of budget pays less than the board shows.

4
Personal Selling Elasticities: A Meta-Analysis
Sönke Albers, Murali Mantrala and Shrihari Sridhar · Journal of Marketing Research, 2010 · the link opens the free 2008 working paper at the Marketing Science Institute, which has an earlier title

Across four decades of studies, 10 percent more selling effort brought about 3 percent more sales, so doubling the effort lifts sales by only about a quarter. The effect was larger for young products and in Europe than in the United States, and has weakened over time.

5
A Field Experiment on Search Costs and the Formation of Scientific Collaborations
Kevin Boudreau, Tom Brady, Ina Ganguli, Patrick Gaulé, Eva Guinan, Anthony Hollenberg and Karim Lakhani · Review of Economics and Statistics, 2017 · the link opens the free author manuscript at PubMed Central

At a medical research symposium, scientists placed at random in the same 90-minute session were about 75 percent more likely to apply for a grant together than pairs kept apart. The odds stayed small, 0.28 percent of pairs against 0.16 percent, and the estimate is imprecise.

6
Drivers of Sales Performance: A Contemporary Meta-Analysis. Have Salespeople Become Knowledge Brokers?
Willem Verbeke, Bart Dietz and Ernst Verwaal · Journal of the Academy of Marketing Science, 2011 · the link opens the free 2010 working paper at Erasmus University, which has a shorter title

Across 268 studies of nearly 80 000 salespeople, sellers did best when they knew their product and their buyers and adapted their pitch. An unclear role was the biggest drag, more so when selling to businesses, and all the factors together explained about a third of the gap between sellers.

Do incentives make volume stay?

On the board: token incentives hold volume while they run, but only about 37 percent of it counts as theirs (range 12 to 62), because the rest would have come anyway. Little stays after they stop: about 5 percent of the peak a year later, and on a new chain 25 percent of what is left goes each month once the chain's incentives end.

7
STIP Retroactive Analysis: Spot DEX TVL
Blockworks Research for the Arbitrum Research and Development Collective · Arbitrum governance forum, 2024

Compared with similar exchanges that got nothing, Arbitrum's 2023 to 2024 incentives explained 12 to 62.5 percent of the deposits at four exchanges, about $2 to $12 of deposits for each $1 of tokens paid. Four to six weeks after the payments stopped, those exchanges held 47 to 78 percent of what they held when the programme ended.

8
Phantom Liquidity in Decentralized Lending
Andreas Park and Jona Stinner · WBS Gillmore Centre Annual Academic Conference, 2024 · the link opens the free conference slides, as no full paper was found

On Aave and Compound from 2019 to 2023, rewards drew activity and their end was followed by outflows. About a quarter of deposits was phantom, tied to accounts that deposit a stablecoin and borrow it straight back; on Compound these accounts held 18 percent of deposits and 31 percent of loans.

9
Decentralized Lending and Its Users: Insights from Compound
Kanis Saengchote · Journal of International Financial Markets, Institutions and Money, 2023 · the link opens the free version on arXiv

On Compound from 2019 to 2021, token rewards did little for deposits but lifted borrowing, as users borrowed to deposit again and earn more rewards. Loans lasted 31 days on average, and the top 100 accounts held 75 percent of deposits.

10
Airdrops: Giving Money Away Is Harder Than It Seems
Johnnatan Messias, Aviv Yaish and Benjamin Livshits · arXiv working paper, 2023, revised 2025

In nine large token airdrops, recipients sold up to 66 percent of the tokens quickly, often as their first move after claiming. In the Arbitrum case, the burst of activity around the airdrop did not turn into lasting use.

What moves lending volume?

On the board: only loans pay interest, so deposits that find no borrower earn the protocol nothing; 10 percent higher crypto prices bring about 7.5 percent more borrowing; and a higher protocol fee leaves depositors less yield, so it costs deposits.

11
Why DeFi Lending? Evidence from Aave V2
Giulio Cornelli, Leonardo Gambacorta, Rodney Garratt and Alessio Reghezza · Journal of Financial Intermediation, 2025 · the link opens the free working paper at the BIS

On Aave from late 2020 to mid 2022, a cut of 1 percentage point in the US policy rate went with about 35 percent more deposits, mostly from small users. Large users followed Aave's own rates instead, and a 1 percent rise in the ETH price went with about 0.75 percent more borrowing.

12
DeFi Protocols for Loanable Funds: Interest Rates, Liquidity and Market Efficiency
Lewis Gudgeon, Sam Werner, Daniel Perez and William Knottenbelt · ACM Conference on Advances in Financial Technologies, 2020 · the link opens the free version on arXiv

In early 2020, borrowing rates for the same stablecoin on Compound, Aave and dYdX moved together, and a shift in the rate at Compound, the largest, carried over to the other two for good. Money was concentrated: three accounts held about half of the DAI locked in Compound in June 2020.

13
Anatomy of a Run: The Terra Luna Crash
Jiageng Liu, Igor Makarov and Antoinette Schoar · Journal of Financial Economics, 2026 · the link opens the free 2023 working paper at the NBER

Anchor, a lending protocol paying high yields, held about 12 billion UST before the run of May 2022 and under 2 billion within a week. The largest holders left first and lost about 26 percent on average, against 76 percent for the smallest.

What does a sales team cost and do?

On the board: a person costs about the median pay in these reports, a new hire takes 6 months to reach full speed, and payback is held against the 16-month median. They are surveys of companies that chose to answer, so read them as ranges, not targets.

14
AE Models, Motions & Metrics: 2026 Research Report
Matt Bertuzzi · The Bridge Group, June 2026 · the link opens the free page of findings and method

In a survey of 158 business-to-business companies, new sellers took 6.2 months on average to reach full output, the longest the survey has recorded. Only 48 percent hit their quota, down from 51 percent in 2024, and the median pay at target was $200 000 against a median quota of $960 000.

15
Dragonfly 2024/2025 Crypto Compensation Report
Zackary Skelly and Chris Ahsing · Dragonfly, September 2025

Across 85 crypto companies and about 3 400 pay records from late 2024 to early 2025, both salaries and token grants were lower than a year earlier at nearly every level. Half of DeFi companies paid bonuses, against 71 percent of centralised crypto companies.

16
2026 SaaS and AI Metrics Benchmarks
Benchmarkit and Aleph · June 2026

In this benchmark of business-to-business software companies, the median company took 16 months to earn back what it spent to win a customer, while the best quarter of companies took about 6 months or less.

Does a known name make selling easier?

On the board: the brand is the protocol's rank among lending protocols by followers on X. At #8 of 13 it is the plan; at #1, ahead of Aave, first meetings rise about 45 percent from every source and 20 percent more of them qualify and sign. Sponsorships build it, slowly, through the year. The ranking is real; what a place is worth is an assumption, so the lever is a what-if.

17
The 5 Principles of Growth in B2B Marketing
Les Binet and Peter Field · The B2B Institute at LinkedIn, 2019

Campaigns from 1998 to 2018 in the databank of the Institute of Practitioners in Advertising: business-to-business brands grew best with about 46 percent of the budget on brand building and 54 percent on activation, and brand effects pass the short-term effects only after six months. The authors call this tentative, because fewer than 50 of the cases are business-to-business.

Much of the selling and advertising research above was done outside crypto. Read it as a direction for a lending protocol, not as numbers to copy. Links checked on 18 October 2026. Each lever and the studies behind it: the evidence. Back to the board.

Propose an improvement

An agent reads every request, writes it up as a build prompt, then builds and tests the next version of the board. It goes live once Victor has reviewed it, and shows up under Feedback and fixes. Your request carries the page you are on, your levers, your window size and the build, nothing else.