Two orders. Same store, same afternoon, near enough the same number of items. One offers $19. The other offers $11.
You can't find the difference because you're not allowed to see it.
There used to be a formula
Before 2020, Shipt paid $5 per delivery plus 7.5% of the customer's order total. That's it. That's the whole thing.
Which meant you could work it out yourself. A $200 order paid $20. You knew it on the offer screen, you knew it in the parking lot, you knew whether the day had been good before the deposit landed.
Shoppers who worked back then will tell you the pay wasn't always great. It was legible, though, and legible turns out to be worth a lot.
Then it wasn't a formula anymore
In early 2020 Shipt started rolling out a model it called effort-based. The published description is that offers carry an estimated pay based on the time and effort it takes to complete the order.
Order size, estimated shopping time, mileage. Those are the factors IEEE Spectrum reported going into it.
What Shipt has never published is the weighting. How much is an item worth against a mile. Whether a difficult store scores differently from an easy one, or whether the store matters at all. Where the line sits between a $19 offer and an $11 one.
Shipt said the change was fairer, and that it matched pay more closely to the work required. It may even be true. There's no way for anyone outside the company to check it, which is the part that matters.
Shoppers noticed the same week. Same routes, same stores, smaller deposits, no explanation.
So the shoppers built their own audit
This is the part of the story that deserves more attention than it gets.
A researcher named Dana Calacci, then at the MIT Media Lab and now teaching human-centered AI at Penn State, built a text message bot. Shoppers texted it a screenshot of their pay. Optical character recognition pulled the numbers out. The whole thing ran on Python and Twilio and cost almost nothing.
They called it the Shopper Transparency Calculator.
More than 5,600 screenshots came in from over 200 workers before collection paused in October 2020. Alongside it, Coworker.org analyzed 6,503 data points from 213 shoppers.
A company changed how it paid people, declined to explain how, and the people it paid reverse engineered the answer with a phone number and a scanner.
What the numbers said
Around 40% of workers were earning less than they had under the old formula. Half of those were down 10% or more.
Roughly a third of the shoppers in the sample were earning below their own state's minimum wage.
The Coworker.org cut of the data landed in the same place from a different angle. 41% earning less, an average reduction of 11% per shop, and most of the workers losing money were losing between fifty cents and three dollars on each one. By early October that share had climbed to 60%.
Here's the finding that explains why Shipt and its shoppers could both be describing reality. In aggregate, V2 paid slightly more. The money just wasn't spread evenly. Some shoppers gained, more shoppers lost a little, and the company average went up while a large minority went down.
Both sides were telling the truth. Only one side had the full dataset.
What Shipt said about it
Not much. The public statements described the algorithm as compensating for effort required, and noted that shoppers are free to decline any order.
That second point is technically correct and completely beside the point. Declining an underpriced offer doesn't tell you why it was underpriced, and it doesn't make the next one legible either.
Where this stands in 2026
Be careful with the numbers above. They're a snapshot of 2020, taken during the rollout, and Shipt has changed things since. Nobody has re-run that audit at that scale. Treat the study as proof of what happened then, not as a description of your paycheck this week.
Some things did improve. In 2022 Shipt published an earnings standard: no offer pays less than $16 an hour, every market, running past $27 in the busiest ones. Worth knowing that the current shopper pages don't restate that number. There are bonuses for on-time delivery, for claiming orders before promo hits, for specific retailers and time windows. Promo pay gets added to offers that go unclaimed as their delivery window approaches. Shoppers keep 100% of tips. And Shipt states there's no minimum number of orders you have to accept, so declining genuinely is free.
That's more disclosure than 2020 had. It's still not a formula.
And the earnings standard has a catch worth understanding before you lean on it. The $16 is measured against Shipt's estimate of the order, not your clock. If the estimate says 40 minutes and the store takes you an hour because half the list was out of stock, the floor held and your hour didn't. The drive to the store isn't in the estimate either.
Why it still feels random
Because from where you sit, it is.
Randomness isn't only about whether a process has a rule. It's about whether you can see the rule. A system with strict internal logic that you're never shown behaves, from the outside, exactly like a coin flip. You can't predict it, you can't plan around it, and you can't tell a bad offer from a bad day.
That's the actual cost of the black box, and it isn't the money. It's that planning stopped working.
Which is why the two things you can still measure, what you clear per hour and whether a given offer is worth taking, carry more weight here than on any platform that just tells you its formula.
The only audit you get is your own
Calacci's project worked because of one boring thing. Shoppers kept records.
Not opinions about pay. Records. Dates, amounts, stores, per order, enough of them to see a pattern that no single shopper could see alone.
You can do the small version of that for yourself, and it's the only move available:
- Log what every order actually paid, base and tip separately, against the store and the date.
- Log the miles, including the unpaid drive to the store. Offer pay is estimated partly on mileage, so mileage is the one input you can see both sides of.
- Give it a month. One order tells you nothing. Two hundred tell you which stores in your metro are systematically underpriced for the work they take.
- Then stop taking those, and stop wondering why the day went badly.
You will not reverse engineer the algorithm. That's not the goal. The goal is a personal record of what the algorithm has actually paid you, which is the only version of the truth you're ever handed.
Shipt knows exactly what it pays you. The gap is that most shoppers don't.
Auto Tip Map is an Android app that saves every order, tip, address, and mile automatically, including tips that arrive hours later. It builds the record this post is asking for without you having to remember to keep it.