Intent-based trading is a trading model where the user signs an offchain message describing a desired outcome rather than broadcasting a transaction directly onchain, with actual routing and execution decided by competing offchain solvers. This differs from a regular DEX swap, where the route, Slippage setting, and pool selection are almost entirely written into onchain data and can be reconstructed step by step.
This model exists to solve two problems ordinary users run into when trading directly onchain: exposure to MEV bots (frontrunning or sandwich attacks), and rarely getting the optimal price when routing a trade themselves. By handing the complex task of "finding the best execution" to a set of competing solvers, users trade a signed intent message for better execution quality and MEV protection — at the cost of moving that process offchain, out of public view.
The concrete flow: a user signs an intent message (containing a minimum acceptable output and a deadline) → the message goes into the protocol's matching system, bypassing the public Mempool → multiple solvers each try to source execution for that intent from different places (DEX pools, CEX quotes, private inventory, cross-chain liquidity) and bid against each other → the winning solver packages the final result into a single onchain settlement transaction → the analyst only ever sees the final transfer result in that settlement, never the bidding process beforehand, the candidate routes considered, or the losing solvers' quotes. CoW Protocol's "coincidence of wants" mechanism is a special case: when two users' intents happen to satisfy each other directly, the match can bypass external liquidity pools entirely.
The practical implication: if you're using onchain data to evaluate the "real execution quality" of an intent-based protocol (or any system relying on solver bidding), looking at the settlement transaction alone isn't enough — it only tells you the final outcome, not whether the user could have gotten a better price, or how many potential matches failed simply because no solver bid competitively. Seeing that deeper layer depends entirely on whether the protocol chooses to publish offchain matching data (solver bidding history, auction records) — it's simply not visible in public onchain data alone.
A standard swap on a decentralized exchange leaves a reconstructable trail: the user's wallet initiates the transaction, picks a specific Liquidity Pool, sets a Slippage Tolerance, and the trade routes through a defined path of pools to complete — nearly all of this is written into the transaction data and event logs, and an onchain analyst can walk back through it step by step.
Intent-based trading protocols, CoW Protocol among them, work entirely differently. Rather than broadcasting an onchain transaction directly, the user signs an offchain message that essentially says "I want at least this much USDC for 1 ETH, by this deadline" — this message never enters the public Mempool; it's submitted to the protocol's matching system instead. The actual execution is handled by a set of competing "solvers," each trying to source the best possible execution for that intent from different places — DEXs, CEXs, private liquidity inventory, cross-chain liquidity — and whichever solver wins the matching packages the final result into a single onchain settlement transaction. By mid-2026, CoW Protocol had processed over $80 billion in cumulative volume — large enough that it can't be ignored — yet almost all the execution complexity behind that number happens offchain.
Opening a CoW Swap settlement transaction, an analyst can confirm: the final amounts transferred in and out, the settlement contract itself (which enforces the atomic swap), the identity of the solver that executed it (though solvers typically operate under anonymous or semi-anonymous addresses), and the gas cost of that settlement. This is "outcome-layer" information — real and verifiable.
What's invisible is just as clearly defined: the exact minimum acceptable output the user originally signed into their intent message (onchain you only see the final execution price, never the user's original floor); exactly where the solver sourced the liquidity to fulfill the intent — possibly a blend of multiple DEX pools, centralized exchange quotes, or even the solver's own inventory; whether this particular match happened through a "coincidence of wants" (Alice wants to sell 1 ETH for USDC, Bob happens to want the reverse, and the two are matched directly), a Dutch-auction mechanism, or some batch settlement optimization — the choice of matching mechanism is never written into onchain data; how many solvers actually bid on that intent before matching, and what each one quoted — that entire bidding history leaves no onchain trace at all; and finally, the actual spread profit the winning solver captured — the gap between the real price it sourced and the price the user received — is equally opaque.
Take CoW Protocol's core mechanism: if Alice wants to sell 1 ETH for USDC while Bob simultaneously wants to buy 1 ETH with USDC, the protocol can match the two intents directly against each other without routing through any external liquidity pool — which means no Slippage and no MEV exposure. But what's observable onchain is just two Token transfers settling atomically within the same transaction. The underlying fact that "this was a coincidence-of-wants match" is nearly impossible to reverse-engineer from onchain data alone, unless the protocol discloses it itself.
This structural opacity isn't a flaw — it's a side effect of the very problem intent-based trading is designed to solve. The protocol pulls the complex task of "find the best execution price" out of the user's hands and out of onchain competition, in exchange for better execution and MEV protection — but the cost is that a large portion of the onchain footprint analysts normally rely on to reconstruct market behavior has moved offchain. For researchers, this means that evaluating an intent-based protocol's volume and market share only shows "how many intents were successfully settled," not "how many bids failed" or "how far the actual execution quality fell from the theoretical optimum" — that layer of evaluation depends entirely on whatever offchain data the protocol chooses to disclose.