What Optimizing EV Powertrains Taught Me About Pricing Your Shopify Store

By day, I'm an optimization researcher at McMaster University. My master's thesis is a loss-minimization control system for the electric motors that drive EVs — and I'm still in the program, because the university has me continuing that work on an industry project with Stellantis. My research group publishes with engineers from Stellantis, Fiat's research centre in Turin, and TOFAŞ.

I'm telling you this not to flex, but because it explains a design decision in Observtics that I think matters more than any feature: your store's pricing problem and an EV powertrain's efficiency problem are the same problem. Once you see it, you can't unsee it — and you start noticing that most stores are running, in engineering terms, without a controller.

The same problem, wearing different clothes

In a PMSM motor drive, there are infinitely many ways to produce the torque the driver is asking for. Each one wastes a different amount of energy as heat — in the copper, in the iron core, in the inverter's switches. My thesis formulates that as a constrained optimization: at every operating point, find the current combination that produces the required torque with minimum total loss, without ever violating the voltage, current, and thermal limits that protect the machine.

Now look at a product in your catalog. There are infinitely many prices you could charge. Each one produces a different amount of profit — margin per unit times the volume that price attracts. The best price is a constrained optimization too: maximize margin × volume, without violating the limits that protect your business.

The mapping is almost embarrassingly direct:

• Loss minimization ↔ profit maximization. Same objective structure, opposite sign.
• Voltage and current limits ↔ margin floor and market ceiling. A motor controller never commands a current the windings can't survive. Our pricing engine never recommends a price below cost + 15%, and never chases a competitor more than 10% down. Hard constraints, not suggestions.
• The operating envelope ↔ the price envelope. In a drivetrain you don't leap between distant operating points — you move through a feasible corridor, because large instantaneous changes destabilize the system. Prices work the same way: we cap any recommended move at ±30% of the current price, because a whiplash price change destabilizes customer trust the way a current spike destabilizes a rotor.
• Physics-informed models ↔ demand models with structure. In my thermal-estimation work with the Stellantis team, pure black-box neural networks underperformed a hybrid: a physics model that gets the structure right, plus a learned correction on top. Pricing is identical. We don't ask a black box to hallucinate your demand curve — we anchor a constant-elasticity demand model to a real, observed competitor price, then optimize on top of that structure. Structure first, cleverness second.

Even the optimizer is shared DNA: the differential evolution search inside Observtics uses mutation and crossover parameters I first tuned on motor-control problems.

"No controller" is a choice — just not a deliberate one

Here's the part that actually changed how I think about ecommerce. In motor control, nobody debates whether to have a controller. Running a motor at fixed, hand-picked currents while the load and speed change constantly would be considered malpractice — you'd waste hundreds of watts and eventually cook the machine.

But that's exactly how most stores run pricing. The price was hand-picked once, at launch, under market conditions that no longer exist. Competitors move weekly. Costs creep. Demand shifts with seasons. And the "controller" — the thing that's supposed to respond — is a founder's gut feeling, applied sporadically, usually only when something already feels wrong.

The waste is just as real as in the motor; it's only less visible. An overpriced SKU bleeds volume quietly. An underpriced one donates margin on every order. In my thesis, adding a proper controller cut system losses by 5–26% depending on the operating point. I built Observtics because I kept seeing the ecommerce version of those loss curves in store catalogs — and nobody was measuring them.

Measure your losses first

Every engineering project starts the same way: instrument the system, measure the losses, then decide what's worth fixing. So that's the free front door of Observtics. Upload your Shopify product export, let it find competitor prices automatically, and the same constrained-optimization engine — margin floors, market ceilings, operating envelope and all — will show you the estimated monthly profit impact of every gap in your catalog. It takes about a minute: run the free margin audit.

If the losses turn out to be small — great, you're one of the well-tuned machines, and it cost you nothing to confirm it. In my experience, though, systems that have never been measured are never as efficient as their operators assume. Motors or stores.