Getting into PM

Software engineer to PM: what changes and how to switch

Lesson 35 of 45~7 min read
In this lesson3 sections

An unofficial lesson drawing on the methods in the Getting the Right Experience chapter of Cracking the PM Interview by Gayle Laakmann McDowell and Jackie Bavaro (2014). PracHub is not affiliated with the authors or publisher. Buy the book

Software engineers moving into product management start with a real advantage: they've spent years watching which PMs are worth copying, and they know what a feature costs to build. What hiring managers test is the part engineering rarely asks of you, which is starting from the customer and getting a team to back a call you made.

This lesson follows one invented engineer, Farah, who has spent four years on the backend of a payments API that online shops use to take card payments. When a payment succeeds, the API sends the shop's server a message called a webhook, which is lost if that server is down. Last quarter a partner shop asked for webhook retries, and how Farah handled it became her case for the PM job. A missed webhook, at one invented shop:

Tomas is the only developer at a small flower shop. On the Friday before Mother's Day he ships a fix at 6pm, and the server is down for ten minutes. Eleven payments go through meanwhile. On Saturday morning the owner forwards him three emails from customers who were charged and heard nothing.

Answering "why leave engineering?"

Expect some version of this in almost every loop when you're switching from engineering. Engineers miss it two ways. Some run engineering down ("I'm tired of building what other people decide"), and the hiring manager wonders how you'll treat your future engineers. Others stay vague ("I want more impact"), which every candidate says.

A stronger answer names a product problem that pulled you toward the work, what you did about it, and what you'll bring from engineering. Farah's:

"When a partner asked my payments API team for webhook retries, I went through three months of support tickets and joined two partner calls before anyone wrote code. The real problem was paid orders stuck as unpaid until someone fixed them by hand, and that changed what we built. I enjoyed that part more than the code, and I want it to be my job. From engineering I keep a sense of what each change costs, so I can tell a partner what's realistic on the call."

That's longer than the one to three sentences the why-are-you-leaving lesson suggests for a plain job change. From a switcher the question really means "why PM?", a motivation question, and the interviewer expects about 40 seconds of evidence.

To practise this step, try Explain your PM transition and growth. Answer only its prompt about moving from a technical role into PM, in under 45 seconds, and ignore the internship framing.

Showing the call you made and the evidence behind it

Before anyone wrote code, Farah went looking for the problem behind the request. She searched three months of partner tickets, 3,600 in total, for words like "webhook" and "unpaid", then read all 620 matches herself. Of those, 540 (15% of all tickets) were about payment notifications that arrived late or never. That's 180 a month.

The tickets showed what broke, not what it cost the shop. So on a partner call she asked:

"You asked for retries. What happens today when you miss a webhook?"

The order stays marked unpaid, the partner said. And then? The charged shopper hears nothing and emails the shop. And then? Someone fixes orders by hand every morning. Keep asking until you reach something the customer cares about for its own sake. The request was retries; the need was knowing every paid order ships without anyone checking.

The delivery logs gave her the size. Nine in ten failed deliveries went to a server answering again within an hour. Allowing for tickets with other causes, Farah assumes 80% of the 180 monthly tickets go away: about 144 fewer a month. At 25 minutes of support time each, that's 60 hours a month.

Tomas's Saturday shows who hears about a miss first: the owner, who isn't technical, and she hears it from customers. So if events still fail after an hour, the shop gets an email in plain words, and Tomas needs a way to resend what was missed. None of that was in the request.

To practise this step, try Customer obsession story: when first-hand customer insight changed a product decision, using a time you went below what a customer asked for.

Her scope for the quarter came from that evidence. Most failed deliveries hit a server that's back within the hour, so retries on by default, the email alert and a resend button cover most of what Tomas's Saturday needed, and those ship now. They leave long outages open: a server down for two days still loses events, and resending means someone noticing and clicking, event by event. A 30-day event log lets the shop's server fetch everything it missed with nobody doing it by hand. It moves to next quarter, with a morning email telling the owner that every paid order from the day before reached the shop. Had she started from "add retries", neither would have made the list.

To practise, try Plan how a 15-engineer Google team would make the internet healthier and Open-ended product design: envision and scope the footwear of the future. For each, say what you'd ship first, what waits, and the evidence that splits them.

Defending the call to your lead and former peers

Farah's recommendation drew fire from two sides. Her lead showed the 80% was too high, and her spec, which stopped retries after 24 hours, drew three objections: partnerships wanted three days, on-call worried about queue size, and one partner asked why not forever.

Her lead's point: most failed deliveries come from large partners with short blips, but most tickets come from small shops, and some stay down longer than a day of retries, say from Friday night to Monday. Farah matched each ticket to its delivery log: retries would have caught 70%. She opened her reply with the correction:

"I said 80%. Matched ticket by ticket, it's 70%: about 126 tickets and 53 support hours a month. Still worth building, and the event log matters more than I thought."

She sent it quickly and in writing, the new number first and every figure built on it redone. The miss cost her little because of how she had first stated the number:

"About 180 partner tickets a month, 15% of the total, are missing or late payment notifications. I read every matching ticket myself, so I'm confident in that number. Nine in ten failed deliveries reach a server that's back within the hour, so I'm assuming a day of retries removes 80% of these tickets, about 144 a month or 60 support hours. I'll check that two weeks after launch. Retries will send some events twice, so we'd ship an event ID with them. I think we should build it this quarter, and I'd like to write the spec."

She was sure only of the count she had checked by hand. The 80% rested on a log pattern and a guess about other causes, so she called it an assumption and gave the date she'd test it. Her lead knew how much weight it could bear; stated as fact, it would have given him reason to recheck her hand count too. Ending with her own call and an offer to write the spec gave him a proposal to answer.

To practise, try Learning from Wrong Data with a decision of yours that rested on a wrong number, and Behavioral: Receiving Blunt Criticism of Your Work and What You Changed, each told in under two minutes.

She treated the three objections as information about the product. On-call's worry was a cost her design hadn't covered, so it changed the spec: a cap per partner. Partnerships' three days was a claim about how long outages last, so she answered it with the delivery logs. "Forever" she declined, and said why.

The repeat risk needed the right words. Retries make repeats more likely, and Farah had watched two partners set up webhooks in a test environment; both treated a repeated event as a new payment. In the design review, instead of calling it an idempotency problem, she put it in the shop's terms, "a shop could ship the same bouquet twice", and an event ID went into the first version. With former peers, technical depth helps when it improves the decision rather than wins the argument.

To practise this step, try PM Behavioral: Using Technical Depth to Shape a Product Team's Decision.

In an interview, tell it that way: the wrong number and how fast you fixed it, then each objection, what it told you about the product, and whether it changed what you built. Our lesson on leading without authority goes further on bringing engineers with you.

To practise this, try Behavioral: Handling Team Pushback on a Product Idea You Championed with a real objection to something you proposed.

All technical product management questions