OKR vs KPI, a 72 hour hackathon, and a psychologist on staff: Nitin Gupta of Prisma

Most companies pick a side in the OKR vs KPI debate. Prisma runs both, on purpose, with a rule connecting them that makes the whole system testable: stop anyone in the company and they can trace their current project back to a company metric in about ten seconds.

Nitin Gupta is COO of Prisma, which builds the next generation of Node.js and TypeScript object relational mapping tools. At the time of recording, around 250,000 active software engineers were using it. Gupta brings more than 20 years in high tech and now lives in France, having previously been in Silicon Valley.

Explaining a technical product to your mother

Michael's opening question is the one every operator at a deeply technical company faces, which is how you explain what the company does to your mom.

Gupta has actually done it, and treats it as a genuine test. His mother has a master's in science and always asks him to explain. If he cannot explain it to her, he says, he should stop.

His version is worth stealing. Pick her favorite app, in this case WhatsApp. Break it into three layers: the interface she touches, the logic that runs when she touches something, and the data layer where the information lives. Developers build the first two layers in one language, say JavaScript, but to store data they have to switch into SQL, because that is what most databases speak.

The analogy he uses for the switch is switching between Hindi and English, which for his family is easy, but for most application developers is genuinely painful because the second language is unfamiliar.

An object relational mapper sits between the database and the developer and handles the translation. You keep writing in your normal language and it produces whatever the database needs. The benefits are confidence, productivity, and the ability to swap databases later, with Prisma handling the migration.

Michael's response is to put his own mother on the call to confirm.

Gupta then explains what sits on top. The ORM is open source and has been in the market around six years. The commercial products, then in early access, apply the same idea of sitting between data and the developer: Accelerate, which gives developers a content delivery network through a function call without learning the underlying infrastructure, and Pulse.

Teaching the non-engineers, and the engineers

With a product this technical, Michael asks how you bring along everyone who is not an engineer, meaning sales, general and administrative, and customer success.

Gupta's first answer is hiring. Curiosity is a company value, and they test for it in interviews, because the outcome you want is someone who may not understand every detail but genuinely wants to learn.

Then the mechanisms. Prisma University is an internal education series, necessary because the pace of change in the edge and serverless space is fast. Technical team members record explanations and the knowledge spreads.

The part I liked is that it runs both directions. Most of the technical team does not understand how finance works either, so finance people present too. His example is the collapse of Silicon Valley Bank: what happened, and why.

Asked how that gets operationalized, Gupta names two things and warns they will sound basic.

The first is an internal mentoring program. People raise their hands to mentor on a topic, usually technical, and make themselves available in groups of one to three. His observation is that this surfaces hidden capability, team members who turn out to be unusually good at explaining concepts or building prototypes.

The second is the internal hackathon, where they deliberately pair people who do not normally work together.

How to run an internal hackathon that includes the ops team

Michael raises the problem he had at Automattic, the company behind open source WordPress, which is that as a business and operations person it was always hard to find a place in a hackathon.

Gupta's answer is that they addressed it explicitly up front: projects do not have to be code, and can be anything that improves either the user experience or the company's own experience.

He was on a team working on the product development process. Every team had its own way of building products, and you cannot design a coherent process without product managers and engineers in the room, but the output was not a line of code. It was a Miro board laying out an eight stage process and, for each stage, the artifact the team must produce to declare that stage finished.

The logistics are where the operational thinking shows.

The hackathon ran 72 hours, Monday evening to Thursday. Around 90% of the company was freed on their calendars, with recurring meetings and one on ones paused. People were told that if a priority item came in they handle it first, and some team members stayed watching for anything breaking.

It was planned by a member of the events team, whose job was to account for everyone's time commitments. Prisma is fully remote, so the planning ran down to how team members would get food, on the principle that they should be fed and energized rather than improvising.

Gupta describes it as close to a playbook now, because they want to run it repeatedly, and he says they intend to open source some of their non-technical process work.

His ambition going in was that some percentage of hackathon projects would survive afterwards. His expectation coming out was that all of them would.

All hands, with the questions first

Michael notes that finance explaining SVB to the company was a fast response, and asks what makes that possible.

Gupta says the call to action is an all hands, and the team knows that an all hands invitation means something important. The regular cadence is monthly, with a deliberately repeatable format, because otherwise all hands meetings become dull.

The structural choice worth copying is the running order. They open with anonymous questions from the team rather than closing with them, specifically because closing with questions means running out of time. The message is that team member questions come first. He notes the questions are challenging, and that answering them is the job, transparency being a company value.

On SVB itself, the approach was not to rush. They gathered information, told the team an all hands would happen four or five days out, prepared the content, and recorded the session so anyone who missed it could catch up and ask afterwards.

What Gupta owns, and why

Gupta's definition of the role starts from what the CEO is. In his experience CEOs come in three types: sales, marketing, or product. Prisma's CEO, Søren, is product, is good at it, and wants to keep doing it. So the leadership wanted a counterbalance.

That means Søren has product and engineering, and Gupta has the rest.

Inside what he calls business operations, the structure at the time was four groups. Developer connections, covering everything outbound including sales, marketing, events and developer relations. Legal and finance. People and HR. Data and analytics.

It was six months old at the time of recording, and he expects it to break apart as functions mature, with marketing, sales and developer advocacy each eventually standing on their own.

The counselor on staff

The part of the structure Michael cannot get past is the trained psychologist inside the people team.

Gupta's explanation starts with Star Trek, where every starship has a counselor who can overrule the captain because they know what the crew needs. That is the version he uses to explain it internally.

The actual origin is the television series Billions, and the character Wendy Rhoades, who lets people in an extremely high pressure environment decompress. Watching it, he wondered why companies do not have this.

His reasoning is the useful part. People run into these situations constantly and take them to their manager, who in most cases is not trained to handle them, particularly when the difficulty presents as an emotional reaction, or is masked and does not present that way at all. So you want someone trained to listen and understand, who can help the person decompress before it escalates. Because eventually, if needs go unmet, people leave.

The counselor sits in the people team, has a read on what is happening across the company, and team members are encouraged to talk to them. His summary: sometimes you just need to talk.

Michael, who has been building remote companies for well over a decade, says almost nothing surprises him anymore, and that this does.

Commercializing popular open source

Gupta's distinction is between open source and popular open source, and he puts Prisma firmly in the second category. Which means you are at the mercy of your users, who will tell you without restraint what you are good at, what you are bad at, and what you should be doing.

That informs the roadmap, but it requires judgment about what the user is actually asking for, which he says takes serious product management. Their open source product manager is willing to put a foot down and refuse something a thousand people are requesting, because there is a particular vision to execute. And equally willing to build something almost nobody is asking for, because it will matter later.

He is candid that decision making is evolutionary rather than solved. Add one new person, particularly a senior one, and the whole thing wobbles, because they arrive with their own ideas and expectations.

On the commercial side, Prisma has raised around $57 million through Series B, so the open source roots have to coexist with converting users into customers.

Their framework for that is the one two three model popularized by Adam Gross of Heroku, who is also an investor. One is individuals, two is teams, three is enterprises. The central variable in moving between them is value: are you delivering something into the market that people will pay for, while making sure the free offering is not so generous that nobody upgrades.

In practice that means early access, then preview, then general availability, with a generous free tier and an easy path to a commercial transaction for people who see the value.

The line he is firmest about: the open source ORM stays open source. He will not tell more than a million users that they now have to pay. His verdict on doing that is dry, calling it a good exit strategy for an open source product.

Michael's addition is the one to remember: trust is earned in drops and lost in buckets.

Where the company actually is

Prisma was conceived in London. The founders had friends in Berlin, went to visit, slept on some couches, and ended up incorporating there. The CEO is Danish rather than German.

On regulation, Gupta clarifies the structure. The German entity is a subsidiary of the American corporation, so legally they are an American company, though the German entity gets certain benefits and they adhere strictly to GDPR as a European entity.

Why an LLM is a database problem

Michael's framing is that Prisma seems less exposed to generative AI than most, and asks whether that is right.

Gupta confirms they did not panic, and his reasoning is elegant. Prisma sits between the user and the database. A large language model is a database, specifically a vector database. So the thing they are working on is something that sits between the user and a vector database, which is the same product they have always built.

The second angle came out of the hackathon. Tools like ChatGPT understand code structure well enough that Prisma can show users what is happening underneath, turning the product into a learning tool alongside a convenience layer. That project was built in two days and he expected to ship it.

Bottom up into the enterprise

The sales motion is developer-led, and the reference Gupta reaches for is a Twilio billboard he used to pass on the 101 in Silicon Valley, which read that they were Twilio, and to go ask your developers about them.

Products like that enter through a developer who sees a benefit and builds around it. What has to be engineered is the path from there into a purchase, and Gupta's read on enterprise buying is that it is largely compliance and risk mitigation, so the packaging question is how to compress the cycle by answering those concerns up front.

But the anchor is still value, and he wants it demonstrated rather than narrated. The team has calculators, and the ambition is that the analytics live in the product itself, so a customer using the CDN product can see how much faster their queries are.

At the time of recording most acquisition was self serve, the sales function did not exist yet, he expected to need one within about eight months, and he was handling enterprise conversations himself.

The hybrid: annual KPIs, quarterly OKRs

This is the section to read twice.

Prisma runs seven core metrics at the company level. They are annual and deliberately do not change quarter to quarter. The examples Gupta gives are annual recurring revenue, employee net promoter score, and customer churn, with target end states computed at the start of the year.

Teams then set OKRs quarterly. The rule that makes the system work is that every objective must align directly to a company KPI.

Which gives the test. Stop anyone in the company, ask how their project helps, and they can draw the line back to a company metric within ten seconds. His stated purpose is that nobody at Prisma should be unable to explain how their work moves the company forward.

The worked example he chooses is the hackathon. There is a target eNPS score. The people team owns moving toward it, so they set quarterly objectives with key results. The hackathon is one of those key results, run to support wellbeing, and the eNPS survey at the end of the quarter will reflect whether it worked.

The metrics are public internally, on a dashboard called MTM, for metrics that matter, which he says is a pinned tab in nearly everyone's browser. It updates reliably, and the business review runs weekly on Fridays, so people watch for what is progressing and what is not.

The planning cadence is scheduled a year in advance. In the third month of each quarter a fixed series of meetings begins. The first is a team self assessment of what is working, so they keep doing it, and what is not, so they improve or stop it. That feeds next quarter's plan, including anything carried forward. By the final week of the quarter the plan exists, and it is presented company-wide at an all hands in the first week of the new quarter.

Nobody is guessing when the planning meeting is, because it has been on the calendar since January.

His diagnosis of what this prevents is the reason to bother: without the alignment, teams form their own opinions, stay busy, and do not move the company forward together.

The logo that came off the screen, and then went back on

Gupta's crazy story is from Scandit, where he ran partnerships.

The deal was with one of the two largest mobile handset manufacturers in the world, who were building a solution on one of Scandit's SDKs. The negotiation took ten months. It was during COVID, so entirely remote, with no chance to shake a hand and build a relationship, and conducted through a translator.

His CEO's instruction had been unambiguous: the Scandit logo does not come off the screen.

He remembers the first time he raised it. A pause. The translation. And then laughter, which he took as a clear answer.

The thing he never expected to see is that Scandit's logo ended up on that app anyway.

His explanation is one word: logic. Stick to requests that are logical and sensible, and you can turn a company of that size around, because there is a reason you are asking and you can make them believe the reason. He describes it as learning to hold his position without getting emotional, pursuing the business relationship through logic first and converting it into a personal one afterwards.

He became friends with his counterpart remotely, sharing family photos, and has still never met him.

The 5 things I took away from this conversation

1. The OKR vs KPI question has a third answer. Nitin's model is annual company KPIs that do not move, quarterly team OKRs underneath them, and a hard rule that every objective connects to a KPI. That is not a compromise between the two systems. It is using each for what it is good at: KPIs for the stable definition of what matters, OKRs for what you will do about it this quarter.

2. The ten second test is a better audit than any dashboard. Stop anyone in the company and ask how their work helps. If they can trace it back to a company metric quickly, the alignment is real. If they cannot, no amount of cascading documentation will fix it. I am going to use this.

3. Announce that a hackathon does not have to produce code. As the ops person who has struggled to find a place in one, this landed. Nitin's team spent 72 hours designing a product development process and shipped a Miro board with eight stages and a defined artifact per stage. That is a hackathon output, and saying so up front is what let it happen.

4. Put the anonymous questions at the start of the all hands. Small change, real consequence. Questions at the end are the first thing cut when you run over, which teaches people that their questions are optional. Moving them first is a cheap way to make a transparency claim credible.

5. Managers are not trained for the conversation that actually matters. The psychologist on staff is the idea I keep turning over. Nitin's reasoning is that when someone is struggling they go to their manager, who has no training for it, and the situation escalates until the person leaves. Whether or not you hire for it, the gap is real at every company I have worked at.

FAQ

What is the difference between an OKR and a KPI? A KPI is a measure of ongoing business health that stays stable over time, such as annual recurring revenue or churn. An OKR is a time-bound commitment to move something. At Prisma the KPIs are annual and fixed for the year, while teams set quarterly OKRs, and every objective has to align to one of the company KPIs.

Can you use OKRs and KPIs together? Yes, and Gupta argues you should. Prisma's model puts seven annual company KPIs at the top and quarterly team OKRs underneath, with each objective wired to a specific KPI. The test that the connection is real is whether any employee can trace their current work back to a company metric in seconds.

How long should an internal hackathon run? Prisma's ran 72 hours, from Monday evening to Thursday, with roughly 90% of the company cleared on their calendars and recurring meetings and one on ones paused for the duration. Priority incidents were still handled, with a small number of people watching for anything breaking.

Do hackathon projects have to be technical? Not at Prisma, and they say so explicitly before it starts. Gupta's own team produced a process design rather than software. Making the scope explicit up front is what makes participation possible for people in operations, finance and other non-engineering functions.

How do you commercialize an open source project without losing the community? Prisma uses the one two three framework popularized by Adam Gross, moving from individuals to teams to enterprises, with value as the variable at each step. The free tier is generous but not so generous that upgrading is pointless, and the open source ORM itself is never charged for, because charging existing users is how you lose the trust the project depends on.

Also mentioned

• Prisma, its open source ORM and the Accelerate and Pulse products
• Node.js and TypeScript, the ecosystem Prisma serves
• Heroku and Adam Gross, source of the one two three commercialization framework and a Prisma investor
• Twilio, and the billboard that defined developer-led enterprise sales
• Scandit, and the ten month logo negotiation
• Automattic and WordPress, Michael's own open source background
• Miro, where the eight stage product development process ended up
• GDPR, and how Prisma's German subsidiary relates to the American parent

Listen to the full episode

Nitin Gupta on Between Two COO's

Between Two COO's is hosted by Michael Koenig. Subscribe on Apple Podcasts, Spotify, or wherever you listen.