Wanting the wanting.

The bit I want to start with came on through one earbud sometime past midnight, on a walk home from a dinner in Gulshan 2. The podcast had cut to a comedy clip I had not been paying close attention to, and the part that made me stop on the footpath outside the Spectra building and rewind the audio twice was this one.

I have been a senior PM long enough that any line about wanting versus having lands in my head as a line about feature requests. The shape the comic was describing is roughly the same loop my B2B customers have been running on me for the last four years, except the object of the wanting is a row on a roadmap instead of a phone in a box. There is a retailer who emails me every Thursday asking when the endpoint for bulk uploads will be live. He is, in some quiet way he himself does not have language for, more attached to the asking than he will ever be to the using. When I ship the endpoint, his usage in the analytics will tick above zero for about three days and then settle back to the flat line it lived on before he ever asked.

The version of this argument I keep circling on calls with other PMs tends to come out sounding like a complaint, when it is closer to an observation about what a feature request actually is once you have watched it for long enough. By the time you have seen the same customer file twenty requests across two renewal cycles, the requests start reading less like a workflow problem and more like a way the customer is saying who he wants to be inside the relationship with the vendor. The cleanest evidence is the gap between request volume and usage after launch in any product mature enough to measure both.

The clearest example from my own work was a piece of functionality a retailer asked us for over almost a full calendar year, the SupplyLine lending dashboard that would let him see his outstanding credit lines against his supplier invoices on one screen. He brought it up on every QBR, again on the renewal call, and at one point sent us a Loom of his accountant in Wari toggling between four browser tabs and a printed ledger, with a polite note saying that if we could put all four tabs into one screen he would call us hero brand. We shipped fourteen months after the first request, with a small launch email and a walkthrough video the CSM recorded in Bangla. Adoption in the third month was around 14 percent of seats provisioned, and the retailer who had asked the loudest was inside that 14 percent for about eleven days before his sessions on the page dropped to zero. The feature lives in production today, carrying the integration cost of two upstream services and a nightly reconciliation job, and nobody on the team has the political budget to propose deprecating it because the same retailer would notice within the week and ask why.

The other example, kept generic because the customer is still on the platform, was a campaign that ran six months, mounted by a B2B customer for email subject lines generated by AI. They wanted, quoting one of the early tickets roughly from memory, the ability to "let AI write our subject lines so we can show our board we are an AI-forward company." We shipped it as a feature flag inside the existing campaign editor. Adoption peaked at maybe four percent of campaigns sent that week and then flatlined. The CSM told me, on a debrief call three months later, that the original push had been about being able to say it on a board slide rather than about the labour of writing better subject lines, and the operator who had been emailing us was not actually the person who wrote them. The person who did had been doing it for nine years and trusted her own ear over the model's. Every campaign that mattered to the company still carried her hand at the top of it; the AI toggle she had been told the company now owned never came up in her morning workflow.

What both examples have in common, and what I think the comic on the podcast was unintentionally diagnosing on stage, is that the asking itself was doing real work inside the customer's week even when the shipped feature was not. The retailer pushing for the lending dashboard got a quarterly call where he was the protagonist of his own working life, an artefact to wave at his accountant in Wari, and a recurring sense that the vendor he was paying actually heard him in the middle of a week mostly spent on the receiving end of supplier deadlines. When the dashboard shipped, the standing reason for that monthly call closed, and the place where the asking had been sitting at the centre of his week for fourteen months had nothing to sit in any more. The other customer got, out of his six months of pushing for the AI that wrote subject lines, a sentence he could say at the next board meeting, and the toggle behind the settings page never showed up in his actual operator's morning again after the second week.

Kahneman's Thinking, Fast and Slow has a chapter on what he and Schkade called the focusing illusion, the cognitive habit of believing whatever you are currently thinking about will matter to your future happiness much more than it actually will. The customer writing the ticket is running that illusion in miniature. The lending dashboard, in the moment the retailer is composing the request, occupies almost the entire screen of his attention, and the feature feels like the missing piece. On the morning the feature ships, his attention has moved on to the dispute over credit periods with his largest supplier, or to the cousin who needs help finding a job, and the dashboard now occupies about three percent of his attention, which is roughly the share of his week it ever deserved.

The right PM response here is not the obvious one of refusing the request on the grounds that the customer will not use the feature, because that move discounts the real value the customer was getting out of the act of asking. The retailer's year of asking was not a wasted year for him. He was getting a recurring sense of being heard by a vendor he depends on, and saying no on day one would have lost us the renewal long before we ever got to the question of whether the dashboard would be used.

The less obvious response, and the one I have been quietly running on the SupplyLine team for about two quarters, is to ship the smallest possible version of the wanted thing, so the customer's asking is honoured without the team taking on ten years of maintenance behind the artefact. For the lending dashboard, the smallest version would have been a weekly emailed PDF pulling the four data sources into one document and dropping it in the retailer's inbox every Sunday night, with no new UI and no nightly reconciliation job. The retailer would have had his artefact to wave at his accountant, the team would have spent three days of sprint work instead of three months, and the month his usage dropped to zero would have cost us a Mailgun bill in the low single digits rather than a permanent slot on the platform's surface area.

The harder version of the same move, which I have not figured out how to operationalise cleanly yet, is to surface the wanting itself as the conversation. The roadmap call where you ask the retailer not what feature he wants but what working week he is trying to escape from is a better call than the one where you take down requests and rank them by frequency. Christensen made roughly this point in the jobs-to-be-done framework, where the milkshake the commuter buys at seven in the morning is not really about the milkshake but about a commute that takes forty minutes and that he is trying to make less boring. The version I would write today, eight years on, is that the job is more often identity than workflow, and the cleanest sign you have not understood it is the adoption curve flatlining two weeks after launch while the next request from the same customer is already in your inbox.

The loop does not catch every request, of course. Some feature requests are real workflow needs, and the way you can tell is that the customer is already doing the work manually with a workaround that costs him measurable time every week, where the ask is to automate the workaround he can show you on screen. Those you take seriously and ship the full version of, because the workaround on his screen is the signal that the workflow underneath is real. When the customer cannot show you a workaround, the request is almost certainly identity work, and the cheapest version, the one emailed as a PDF, is the right answer regardless of how loudly the dashboard is being asked for.

The walk home from Gulshan ended at my own gate sometime after the comic had moved on to a bit about marriage I have already forgotten. I had paused the audio at the wanting line and never pressed play again. The retailer in Wari who had asked us for the lending dashboard for a year sent me a WhatsApp message two days later, asking when our team would be building an API for his ERP to push purchase orders into our system. I sat with the message for about ten minutes. The part of my job I had been doing wrong for four years had just been described to me by a stranger on a comedy clip. The reply I sent the retailer did not promise the API. It asked him first whether he was already doing the push manually today, with a script or a spreadsheet or a junior on his team retyping the orders, and what the actual workaround on his desk in Wari looked like, before I would write a single line of the spec.