Are Sandwich Bots Profitable? Costs, Odds and What's Unknown

Sandwich bots can be profitable for well-equipped operators, but nobody can honestly promise it. Gross gains are visible on-chain; net results depend on gas, tips, failed attempts, infrastructure and competition. A vendor page that gives no performance figures, as CryptoMEV Bot’s sandwich page does not, is at least not inventing them.

Gross versus net

The first trap is accounting. A sandwich that returns a small gain looks like a win, until you subtract both legs’ fees, any tip or builder payment, the cost of the attempts that failed, the software licence and the servers. Because failures are invisible if you only look at successes, win-rate stories are unreliable.

A simple monthly ledger is the minimum:

Line	Include	Common mistake
Gross trade gains	Realised balance change after back-runs	Counting unrealised token holdings
Gas	All transactions that landed	Ignoring those that reverted
Tips and builder payments	Everything paid for ordering	Treating them as “optional”
Failed attempts	Cost of reverted or losing legs	Dropping them from the sample
Software and infrastructure	Licence, servers, RPC, data	Amortising too optimistically
Taxes	Per local rules	Forgetting them

Key takeaway: if you cannot compute net profit per week from your own records, you do not yet know whether the bot is profitable.

Why competition compresses margins

An opportunity visible to one bot is visible to many. When several searchers can build the same sandwich, the way to win is to pay more for ordering. In well-studied MEV auctions, most of the surplus tends to flow to the block producer or builder rather than the searcher, which is one reason searchers invest so heavily in latency and exclusive order flow. We do not quote a percentage, because it varies by period and strategy and we cannot verify a single figure.

The shrinking pool of exposed victims adds to this. As wallets and users adopt protected routing, described in how to avoid sandwich attacks, fewer large, loosely protected swaps reach the public queue.

What operators lose money on

1. Latency losses: being second to the same target.
2. Stale state: simulation no longer matches the pool.
3. Bad tokens: tokens that cannot be sold back, or that charge hidden fees.
4. Tip escalation: paying away the margin to win.
5. Overbuilding: buying expensive infrastructure for opportunities that do not exist.

CryptoMEV Bot’s explainer on what an MEV bot is claims that most MEV bot operators lose money, attributing the claim to research by Flashbots and EigenPhi. We could not verify that figure independently, so we do not repeat it as fact. The qualitative point, that losses from failed transactions and gas are common, is consistent with how these markets work.

Reading vendor claims

Questions that separate useful information from marketing:

• Is a profit figure tied to a period, a chain, a capital size and a method?
• Does it include costs, or only gross?
• Is it a typical result or the best one?
• Is it audited, or a screenshot?

The vendor page we reviewed describes the product’s strategies and simulation, lists a Performance section heading, and shows pricing, but states no profit numbers. That is a more cautious position than many in the sector. Pricing is covered in what a sandwich bot costs, and break-even maths follows from it.

A break-even way of thinking

Instead of asking whether bots are profitable in general, ask what your net margin per landed attempt would have to be, and how many landed attempts per month you need, to cover fixed costs. If a Retail Trader licence costs a one-off $1,999, as the vendor lists, plus servers and a funded wallet, you can compute the monthly net you need to repay it over, say, a year. Then compare that requirement with realistic conditions on your chosen chain. If you cannot see how to reach it without heroic assumptions, the answer for you is no.

Alternatives to compare against

A fair benchmark is any lower-risk use of the same time and capital. Cross-venue arbitrage, discussed in sandwich bot alternatives, has its own competition but does not require harming a specific user’s order. The vendor also sells an arbitrage-oriented product line for readers who want to see that approach.

Bottom line

Profitability is possible and unproven for any individual newcomer. Start small, keep records, treat the first months as a measurement exercise, and read sandwich bot risks before sending funds. Never risk money you cannot afford to lose.

Three honest scenarios

Rather than a forecast, consider three stylised outcomes, none of them predictions.

The infrastructure-rich operator. Runs tuned servers, has fast data and pays attention to tip economics. Can capture a share of the opportunities that exist, and keeps only what is left after paying for ordering. The edge is real but narrow, and it has to be maintained against rivals who copy it.

The licence buyer with a standard setup. Uses a shared, pre-built strategy on ordinary infrastructure. Competes for the same opportunities as every other user of the same tool, often arriving late. Results are likely to be small or negative once costs are counted, though the dashboard may show a few winning attempts that look encouraging in isolation.

The cautious tester. Treats the first period as measurement, keeps a dedicated wallet with a small balance and stops if net results are negative. This operator may not make money either, but is unlikely to lose more than the budget set for learning.

Which of these you are matters more than any average. It also explains why public profit screenshots mislead: they come disproportionately from the first group and from good weeks.

Questions to answer before scaling

• How many attempts landed, and how many were submitted?
• What was the net result per landed attempt after every cost?
• How much of the result came from one or two unusually large trades?
• Would the result have survived a doubling of tips?
• What happens to the strategy if protected routing becomes more common?

If you cannot answer these from your own logs, scaling is premature. The cost side of the equation is itemised in what a sandwich bot costs, and the failure modes are in sandwich bot risks.

This article is educational and not financial advice. Past results do not predict future results.