Product1 publisher3 min readPublished
Atomic backer Jon McNeill says DoorDash runs about 90% of its purchasing on the startup's software
Atomic raised $12.5 million after its recurring revenue quintupled this year, according to Jon McNeill, whose firm incubated it. The case that operators now let its software reorder stock on its own rests on what McNeill and the founders say.
The Product Desk · Product desk
What happened
- Klass Capital and Madrona Venture Group led the Series A, which takes Atomic's total funding to just over $15 million.
- Atomic's software simulates scenarios to decide how much inventory a company should hold and where, then recommends a response or picks one itself.
- The founders built an early version at Tesla during the 2018 Model 3 ramp, when the automaker's spreadsheets could not keep up with planning changes.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- exposure A purchasing team that lets Atomic infer its unwritten rules can end up with software placing orders on logic no manager has reviewed or signed.
- decision Because the same product both advises and buys, a buyer has to decide order category by order category where the software may act without sign-off.
- constraint Without a starting revenue figure, fivefold growth cannot tell a buyer how many customers let the software act and how many only read its advice.
The moment McNeill describes is a planner asking to be taken out of the loop. "Then customers were saying, 'Okay, then you might as well make the decision and free my time up,'" he said [9]. It is a secondhand account, relayed by a member of Atomic's board [7] whose venture firm, DVx Ventures, incubated the company [2].
McNeill's pitch is a product that "not only gives recommendations but it makes decisions, so it's fully autonomous in that sense" [6]. His evidence for the practice is one customer: "DoorDash is running, I think, 90% of its purchasing across hundreds of sites," he said [6]. That hedged estimate is the only evidence in the report that an operator lets the software buy at scale. HelloFresh appears only as a customer name [10]. The report does not include a comment from either company.
McNeill said the board set Atomic the goal of "compressing their onboarding time and really making it a non-event for a customer to turn this on" [15]. He also said the software works out the decision rules a customer's staff follow even if nobody wrote them down [8]. Put fast onboarding next to inferred rules and a purchasing team can reach autonomous ordering before anyone has read the rules the software applies on its behalf.
The sales argument is speed. "When I got to know Elon, he said the thing that will separate us from all of our competitors is decision speed, because decision speed compounds," McNeill said [12]. For food customers such as DoorDash, Atomic says the software cuts waste and spoilage [11]. Spoilage rates and the share of orders planners override are the measures that would show whether faster decisions are also better ones.
Most of Atomic's money is new. Before this round it had raised roughly $2.5 million [1].
Two axes sort the decision for a purchasing team. One is whether a rule is written down and owned by a person, or inferred by the software. The other is what a wrong order costs. In my view the software should earn autonomy one quadrant at a time, starting where rules are written and mistakes are cheap. The tradeoff is that the planner time customers wanted back [9] stays tied up in the expensive orders for longer. Written rules on cheap orders, such as shelf-stable restocking, can run on their own. Written rules on expensive orders can run under a spend cap, with a person signing anything above it. Inferred rules on cheap orders can run once a planner has compared a week of the software's orders with the ones they would have placed. Inferred rules on expensive orders, including most perishable stock, stay in recommend mode until someone has read the rule and signed it.
What to watch
- A statement from DoorDash or HelloFresh on what share of purchasing Atomic's software places without a person signing off.
- Atomic disclosing an absolute recurring revenue figure, a customer count, or measured spoilage reductions.
- Whether Atomic lets customers read and edit the inferred decision rules before switching on autonomous ordering.