Copy the homework: a competitor's decisions are free discovery

Early on, I worked at a one-person Canadian startup as the unpaid intern. The founder couldn't pay me in money, so he paid me in what he knew, and the lesson I still use most came up over coffee. If you ever want to open a coffee shop, he said, find the nearest Starbucks. Starbucks has already paid analysts to study the demographics, the income levels, the foot traffic, and the neighbouring businesses of every postal code it enters. If they opened there, the homework has been done. Open a specialty shop a few doors down and let their research pick your location. I laughed at the time. I've since decided it's one of the smarter discovery shortcuts I've been taught, with one catch I'll get to.

What Starbucks actually does

Starbucks runs an internal mapping platform, built on Esri's tools, that pulls together consumer demographics, population density, income, car traffic, transit stops, and the kinds of businesses already nearby, then projects the foot traffic and average spend a candidate site would produce. That description is from a Harvard Digital Innovation write-up, not from Starbucks' internal criteria, which aren't public. The exact thresholds don't matter for our purposes. What matters is that every open Starbucks is the visible output of an expensive, data-heavy decision, and the output is sitting on a street corner where anyone can see it.

Decisions are research you didn't have to pay for

This is the general principle, and it applies far beyond coffee. When a well-resourced company makes a decision, it usually did research first. You can't see the research, but you can see the decision, and the decision encodes most of what the research found. Discovery, in the sense I use it on this site, is the work of finding out what's true before you build. A competitor's choices are a large, free, slightly out-of-date discovery dataset.

For a software product the same reading applies. An incumbent's pricing page is a market-sizing study: it tells you what segments they think exist and what each one will pay. Their job postings tell you what they're building next, six months before it ships. The features they launched and then quietly removed are a list of things that didn't work, which is the most expensive kind of knowledge to generate yourself. The integrations they built first tell you which adjacent tools their customers can't live without. And the markets they have visibly chosen not to enter, after years of being able to, are a signal that something there is harder than it looks.

We built a product on this idea

Squint is this principle turned into software. It takes a prospect's website, finds their strongest competitor, and produces a benchmarked report that shows exactly where the prospect is losing. The value isn't the audit; agencies can run audits. The value is that the competitor's site is evidence of what works in that category, and putting the two side by side turns a vague "your site could be better" into a specific "here is what the company beating you does differently." Nobody had to run a study, because the study was already published, one URL at a time.

The catch

Here's the part the advice leaves out. Starbucks' site selection optimises for Starbucks' model: high volume, fast throughput, a lot of morning commuters who want the same drink in ninety seconds. A specialty coffee shop is a different product. Its customers sit for an hour, spend more per visit, come less often, and care about things Starbucks' model deliberately doesn't. The postal code that's perfect for a Starbucks can be exactly wrong for a place built around a slow pour-over, because the foot traffic that makes the first work is the wrong kind of traffic for the second.

So copying the homework gets you a shortlist, not a decision. The competitor's research answered their question. You still have to check whether it answers yours, and that check is the last mile of discovery you can't borrow. GridSentinel is the version of this I lived through. Electricity-theft detection is a real market with real buyers, and other companies are in it. Copying that market would have been reasonable, right up until technical discovery showed the physics don't close on the medium-voltage lines the product was designed for. The market was real. It just wasn't ours to take with that product. The first version was re-scoped to outage detection instead, and that decision came from doing the last mile, not from reading what competitors had chosen.

How to do it without fooling yourself

Treat every competitor decision as a hypothesis with a source attached, the same way you'd treat an interview quote. "Starbucks opened here, so this postal code supports a coffee business" is a hypothesis. "Starbucks opened here, and their model needs commuter volume, and my model needs people who linger, so I need to find out whether this postal code has both" is the version that survives contact with reality. Write down what the competitor's decision assumes about their customer, then write down where your customer differs. The gap between those two lists is your discovery plan, and it's usually short.

Nothing is original, and that's fine

The objection I hear to all this is that it isn't original. Almost nothing is. Most products are a known model applied to a segment the incumbent has decided is too small, too awkward, or too unprofitable to serve. Dwell is credit scoring, a model that has existed for decades, applied to the 1.2 million newcomers a year that the existing bureaus are structurally blind to. Whatever originality there is lives in the choice of who to serve, and in the willingness to check whether the borrowed model still works for them. Copy the homework, then do the one page of it that's yours.

Sources and further reading

• Starbucks: Brewing up a data storm (Harvard Digital Innovation and Transformation, 2015). Describes the Atlas platform and the inputs it uses for site selection.
• Squint case study. A product built entirely on reading a competitor's site as evidence.
• Dwell case study. A known model applied to a segment the incumbents ignore.
• GridSentinel case study. Where copying a market without checking the physics would have gone wrong.