How PMMs find Product-Market Fit

This case study was created from an episode of the Product Marketing Adventures Podcast. It has been adapted into Dusted's case study format to preserve and showcase the guest's real-world experience. Written to preserve the voice of the true author, Pranav Deshpande.

Finding product-market fit is one of the hardest things a PMM will ever do—and knowing when you don’t have it is even harder. I’ve seen both sides across my career. At Twilio I worked on high-scale messaging and developer-first GTM after the company had already achieved insane PMF with SMS and voice; they went public within a year and a half of my joining. I later joined the early Autopilot team (an NLU chatbot/voice-bot API in beta, pre-LLMs, around 2019). After Twilio I became the first marketing hire at Modern Treasury (~14 people), built the product marketing function over two years while wearing every marketing hat, and faced the challenge of creating narrative and demand in a new FinTech category. Now at OpenAI I help define GTM for the API platform and frontier AI products that developers and enterprises use to build new businesses.

At Twilio Autopilot I carried over the mindset from SMS/voice PMF and didn’t sufficiently question assumptions. We had massive distribution and customer interest thanks to the Twilio halo—no shortage of pilots, especially in contact center and customer service—and the ROI story plus programmability looked strong versus legacy tools. The real problem was core product and underlying tech: the models required heavy hand-holding and hard-coding, were orders of magnitude less capable than even small modern LLMs, and the low-level customizability made flows brittle and created a barrier past initial trials. It wasn’t messaging or distribution; it was product readiness and form factor.

At Modern Treasury the first customers (startups and a few larger logos) came through founder sales, word of mouth, and VC networks. What I didn’t examine critically enough early on was whether those paying customers were acquired and using the product the same way. The business model was untested: a two-sided (effectively three-party) motion requiring the customer, Modern Treasury, and a bank all to align so companies could run cheaper/faster B2B payments on bank rails. We effectively had ~20 different ICPs across a few dozen deals. Product risk felt secondary (strong eng team, great demos); the bigger issues were business-model and market risk and the absence of repeatable sales.

I now treat PMF as a continuous journey, not a binary point-in-time declaration, and operate in two distinct modes as a PMM depending on whether we have it. The core questions I keep asking: Who is the ICP? Are they getting clear value? Is that happening repeatably?

When you lack repeatable sales you don’t truly have an ICP (or you have the wrong one). At Modern Treasury I would deprioritize polished website copy and perfect positioning docs—those only pay off once you know what to say and to whom—and instead wear more of a demand-gen/growth hat: test channels and assets to validate use cases, buyers, triggers, and willingness to pay, then group early deals into sellable segments. Differentiate user vs. buyer: for users, the simple test is whether they would be genuinely sad if you took the product away; for buyers, look for the same persona and budget owner showing up consistently across closed deals.

In a big-company setting the halo effect creates false positives (lots of pilots). I look past vanity metrics to willingness to pay and invest, frequency and quality of engagement with SEs/support, whether customers are pulling the product out of you versus needing constant prodding, and how much hand-holding is required beyond expected onboarding. Systematically I reverse from hypothesized ROI and total cost of ownership: how much should we charge, how long should customers take to realize that ROI, how many have actually crossed the threshold, and how many conversations stall even when the product is already under contract. Opportunity cost for the team also matters when deciding pivot vs. persevere. Across all of it the required mindset is pragmatic optimism—believe in the product enough to position it, but stay clear-eyed about flaws so you can see actual PMF strength.

Comparing the two earlier experiences clarified the diagnosis: Autopilot was primarily a product/technology and form-factor failure; Modern Treasury was primarily business-model and market risk plus missing repeatable motion. Those lessons compound at OpenAI, where PMF for the core models and developer experience is already extremely strong (alongside a small set of peers). The job shifts from convincing people to try the technology to education—helping developers and companies find measurable value and deploy it reliably across domains such as coding, customer service, research, and healthcare so they can build lasting products and workflows on top of it. The overarching takeaway I apply every day is simply to know where we sit on the PMF journey and focus only on the work that moves the needle most for that stage.