Traditional freight brokering involves a significant amount of negotiation and typically requires truck drivers (“carriers”) to build relationships with brokers to gain access to shipments (“loads”), which can be disadvantageous for those who dislike negotiation or lack relationships. Recently, Convoy developed transparent auctions, which employ both reserve prices and fixed closing schedules for its in-app bidding, while requiring no direct negotiation or broker relationships. Transparent auctions level the playing field for carriers, and facilitate competition on loads solely based on carriers’ operational efficiency and service quality. Carriers now have 3 to 7% more loads to select from because loads are no longer taken off the marketplace intermittently, and receive feedback 8 hours earlier because a message is sent to carriers at each auction closing.
With our industry-leading transparent auctions, carriers can search the Convoy mobile app for loads they want to haul before the universal auction closing time, and use the current reserve price as guidance for how much to bid. This provides several benefits to carriers:
- Faster and more predictable feedback: with fixed closing schedules, carriers receive the first feedback 8 hours earlier than the previous version of auctions (“legacy auctions”). Thanks to faster feedback, carriers can make more timely decisions on whether to confirm their winning bids, which leads to a 25% higher confirmation rate in transparent auctions compared to legacy auctions.
- 3 to 7% more loads to bid on with less checking: transparent auctions close and review carrier bids at the same time, so carriers bear lower cognitive burden because they no longer need to constantly check the Convoy app. And they can bid on 3 to 7% more loads before each closing time.
- Fairer access to loads: carriers are evaluated solely based on their bid amount and their historical service quality. They do not need to check the Convoy App 10 times a day to find loads they want to haul.
Why we developed transparent auctions
Outside of the freight industry, posted-prices and auctions are the most common market mechanisms used to discover prices and generate sales. In 2018, Convoy adopted in-app sealed bid auctions to enable carriers to reveal their desired rate. We chose this mechanism because loads could possess idiosyncratic values to different carriers due to differences in their operational efficiency, their schedules and locations at the time of booking, and occasionally “black swan” events like the COVID pandemic.
However, this version of auctions (“legacy auctions”) can close at an unpredictable time because the closing decision is optimized by a machine learning algorithm and human brokers. The two work together to balance truck cost and on-time delivery of loads. For carriers, there is also no guidance on whether their bids are competitive, leaving them to “guesstimate” how much to bid. Figure 1 shows what legacy auctions looked like.
Hence, legacy auctions cannot guarantee feedback time after carriers bid: they may hear back in an hour or a couple of days. In the latter case, their trucks may become unavailable, and therefore they cannot haul the load they end up winning. Not only is this an undesirable carrier experience, but it also increases broker burden: brokers need to answer calls from carriers who want to check their bid status, and spend time and energy re-matching loads if the initial winners indicated they no longer were able to haul loads due to delayed responses.
To summarize, legacy auctions faced three major problems:
- The process of price discovery can be interrupted by broker intervention, leading to incomplete price discovery.
- Carrier experience is not ideal due to uncertain feedback time and the lack of bid suggestions, causing lower confirmation rates of winning bids.
- Broker burden becomes heavier when more bids are placed and more carriers call in to check bid status or negotiate. This can in turn delay broker response where it is needed.
How transparent auctions work
To solve the three problems described above, we refer to consumer behavior research and auction theory (such as Ariely & Simonson (2003), Einav, Kuchler, Levin, & Sundaresan (2011), Jehiel & Lamy (2015), Borenstein(2002)) and other online auction platforms like eBay.com, auctionninja.com, liveauctioneer.com, etc. for ideas. Specifically, for the first problem, we propose a fixed closing schedule, which has up to 3 closings a day, based on the hours left until load pickup time. This prevents auctions from closing prematurely without discovering the price for long enough.
For the second problem, the fixed closing schedule guarantees feedback at each auction closing. To further address this problem, we also developed a machine learning algorithm to generate a reserve price that indicates Convoy’s willingness-to-pay at a given time. The reserve price reduces information asymmetry between Convoy and carriers, and between new and experienced carriers. Now carriers can explore hauling loads in new areas more easily and are more likely to win loads if they bid below the reserve price. Together, the fixed closing schedule and reserve price create a better auction experience for carriers. Figure 2 shows how transparent auctions make bid acceptance decisions in two different situations.
In general, the lowest bid below the reserve price is accepted after adjusting for carrier quality (left panel of Figure 2). If all bids are above Reserve Price 1 before an auction closes, as shown on the right panel of Figure 2, they will be carried over to the next auction closing and evaluated against Reserve Price 2. For example, suppose Reserve Price 1 is $900, and Carrier A, B, and C bid $910, $945, and $960, respectively. When the auction closes for the first time, none of the bids will be accepted because they are all above Reserve Price 1. As the auction reopens with Reserve Price 2 of $930, Carrier A’s bid is below the new reserve price but Carrier B and C’s bids are still above it. Therefore, we accept Carrier A’s bid when this auction closes. In all cases, carriers receive feedback regarding their bid status at each auction closing.
Lastly, with more loads closing automatically following the fixed schedule, broker intervention is less necessary when matching carriers to loads. This helps mitigate the third problem, and allows Convoy to attract and evaluate more bids, and match more loads without overwhelming the broker team. Figure 3 shows what transparent auctions look like.
How we tested transparent auctions
To test transparent auctions, we ran a carefully designed experiment for more than two months, where loads in different areas were assigned into either a treatment or control group in a given week. This ensures that more locations are exposed to the new auction mechanism, which increases power in detecting impact and reduces selection bias introduced by random assignment of a relatively small sample. It also mitigates the potential interference problem in a user split experiment, where only some users can see the product change.
When constructing the reserve price for a new auction closing, the machine learning algorithm considers hundreds of different options and chooses the values that it believes will optimize the expected outcome. However, the algorithm is not perfect, so we regularly explore values that were not chosen by the algorithm to observe how and if they lead to different outcomes. This balance between exploiting the optimal value and exploring new ones is at the heart of how we build an intelligent system that learns and produces the best possible outcomes, all in real-time.
The experiment results show that carriers hear back from Convoy more quickly, have access to more loads, are more likely to confirm their winning bids, and broker burden is significantly reduced. These results indicate a more efficient auction with a reduced level of post-auction reallocation. Being a win-win, we decided to launch the new auction mechanism to all eligible loads and now offer this new capability through the Convoy mobile app.
Join the Convoy Team
We strive to find the best matching mechanism for carriers. Since both carrier preferences and shipment mixes can evolve over time and location, we are always working on incorporating these changes into our marketplace design. If you are interested in joining the Convoy team and working on solving complex, impactful issues, apply here!
Ariely, D., & Simonson, I. (2003). Buying, bidding, playing, or competing? Value assessment and decision dynamics in online auctions. Journal of Consumer psychology, 13(1–2), 113–123.
Borenstein, S. (2002). The trouble with electricity markets: understanding California’s restructuring disaster. Journal of economic perspectives, 16(1), 191–211.
Einav, L., Kuchler, T., Levin, J. D., & Sundaresan, N. (2011). Learning from seller experiments in online markets (No. w17385). National Bureau of Economic Research.
Jehiel, P., & Lamy, L. (2015). On absolute auctions and secret reserve prices. The RAND Journal of Economics, 46(2), 241–270.