पाठशाला Pathshala · विचार Vichār, The idea · Lesson 23 · Scale

Platform shifts and when to bet the company on one

Decide whether a new platform such as AI, ONDC or voice is a feature, a product line or the whole company, with five questions answered in numbers and a date to answer them again.

Pathshala, The Founder Library · 11 October 2026 · 7 min read

Rusty railway tracks diverging through a green forest. — Photograph: BREAKS OUT · Pexels

Every few years a new platform arrives and every founder has to decide what it means for the company they already run. In 2026 the candidates in India are AI, ONDC and voice in Indian languages. Most of the decisions made about them will be made by mood: excitement in one room, fear in another. This lesson replaces the mood with a sorting rule and a review date.

It sets out the three possible answers, explains why incumbents tend to answer late and start-ups early, gives five questions that place a platform in one of the three, and works through the cost of betting the company too early with the best-documented case there is. The closing section is the review cadence, because the answer changes as the platform does.

Three answers: feature, product line, company

A feature. The platform makes one step of what you already do cheaper or better. You integrate it, ship it, and your customers barely notice beyond the improvement. One person owns it. Most platform shifts are this for most companies, and treating a feature as a strategy wastes a year.

A product line. The platform lets you serve a job or a customer you could not serve before, alongside the business you have. It gets a small team, its own revenue number and a quarterly review, and it competes for resources with the core on evidence.

The company. The platform changes the job itself, or cuts its cost so far that the old way of doing it will not survive. The roadmap is rewritten around it, the old business is run for cash, and the founders accept that the next two years will look worse before they look better. This answer is rare, and when it is right it is the most important decision the company will make.

Why incumbents answer late, and start-ups answer early

Joseph Bower and Clayton Christensen explained the pattern in Harvard Business Review in 1995. Established companies do well with innovations that serve their mainstream customers’ next needs. They struggle with technologies that at first perform worse for those customers and serve small or emerging markets, because every process and incentive inside the company directs money towards the customers who already pay. Their prescription was blunt: the only way to pursue such a technology is to create organisations that are completely independent of the mainstream business.

The start-up’s error is the mirror image. With no mainstream business to protect, it bets the company on every platform that arrives, and some platforms are features. Jeff Bezos’s 2015 letter to shareholders gives the distinction to use. Some decisions are consequential and irreversible or nearly so, one-way doors, and deserve slow and careful thought; most are reversible, two-way doors, and should be made quickly. Making a platform a feature is a two-way door. Making it the company is a one-way door. The same letter describes AWS, which began as an unusual bet, reaching $10 billion in annual sales faster than Amazon itself did. The bet was right, and it was placed as a separate business beside the retail company rather than in place of it.

The five questions

Does it change the job your customer hires you for? A platform that changes one step is a feature; one that changes the whole job is a candidate for the company. Does it cut the cost of that job by ten times or more? Twofold improvements are absorbed by incumbents; tenfold ones remake markets. Are your own customers already using it, by a number you can count? Not the market, your customers, measured monthly. Could the incumbent bolt it on inside six months? If so, the platform is not your advantage, because it will be everyone’s. How much of your revenue does it put at risk if you ignore it? This is the question founders skip, and it is the one that decides how fast to move.

Score each answer none, some or most. The figure adds them up and places the platform. The thresholds are a judgement, not a law. The discipline is answering each question with a number from your own business, and writing down the date on which you will answer them again.

Answer the questions with two people in the room who disagree: the person who runs the core business and the person most excited by the platform. Each scores alone, then the two compare. Where they differ by two points on a question, the disagreement is the work: find the number that settles it, from customer calls, from usage logs, from deals lost in the last quarter. Most teams find that the excited person is right about cost and wrong about adoption, and that the person running the core is right about adoption and wrong about how quickly an incumbent can copy.

A feature is a two-way door. Betting the company is a one-way door. Know which one you are walking through.

Netflix and the price of betting early

Netflix is a well-documented case of a company that bet itself on a platform shift, and the cost is better documented than the triumph. In July 2011 it split its plans so that streaming and DVDs by post were sold separately at $7.99 a month each, and later announced that the DVD business would be spun off under a new name. In the quarter that followed it lost 800,000 US subscribers, falling to 23.8 million; the company had abandoned the spin-off shortly before the results, and its share price fell from near $300 before the price change to about $71 after the results. Its own letter said long-standing members had been shocked by the pricing changes.

A black and white photograph of an old railway switch lever on gravel with tracks behind it. — Throwing the switch is a one-way door. Netflix threw it a year before its customers were ready. Photograph: Emmanuel Codden · Pexels

Netflix was right about the platform and wrong about the speed at which its customers would follow. That is the usual shape of a correct bet placed too early: the direction is vindicated and the year is lost. The lesson for a smaller company is not to wait. It is to make the bet as a product line first, with its own number, so that the evidence for making it the company comes from your customers rather than from the strength of your conviction, and so that the core keeps paying for the transition while it happens.

AI, ONDC and voice in 2026

AI scores high on cost for many service jobs: drafting, summarising, first-line support and classification fall in cost by far more than tenfold. It scores low on the incumbent question, because every incumbent can bolt a model onto its product within months. For most software companies that makes it a feature or a product line, not the company, unless the job itself changes: if the customer no longer needs your software because a model does the job, the revenue-at-risk question answers itself. Watch margins too; Andreessen Horowitz found AI companies running gross margins often of 50 to 60 per cent against 60 to 80 per cent for comparable software.

An aerial view of railway tracks crossing and branching in a city. — An open network is only as useful as the traffic it carries. Count the orders before you rebuild around it. Photograph: Krisna Satmoko · Pexels

ONDC scores low today on adoption for most sellers. The network processed about 1.2 crore orders in July 2024, 44 lakh of them mobility, with about 6.3 lakh sellers and service providers; in December 2025 the government told the Lok Sabha that more than 1.16 lakh retail sellers were live across more than 630 cities and towns. For a seller-side software company that is a feature now, with a quarterly review against the order number. For a mobility company where a third of the network’s orders already sit it may be a product line.

Voice in Indian languages is the platform with the widest gap between belief and evidence. Run the five questions against your own customers: how many already speak to a product rather than type, by a number you can count, this month? Until that number exists it is a feature to test, not a company to build.

A review date for every platform

Keep one row per platform in the strategy document: the five answers, the score, the placement, the owner, and the date of the next review. A feature is reviewed every six months, a product line every quarter. Before each review collect three numbers: the share of your customers using the platform, the cost of the job with and without it, and the revenue in deals where it was asked about. Move a platform up a level only when two consecutive reviews say so; move it to the company only after a product line has shown, in its own revenue, that customers follow. The [pivot lesson](/library/pivot-decision-evidence-not-exhaustion) covers what to do with the old business when that day comes.

Figures were checked in October 2026 against the sources below. The scoring thresholds are this library’s judgement. Nothing here is investment advice.