One of the most common misconceptions about running a platform business is that it’s all straight and easy once the initial chicken and egg problem of getting both producers and consumers is solved. Ironically, in an age where kicking off new platforms isn’t quite as difficult as it used to be, the most important platform management issues really come up once the platform starts working and delivering value.

But let’s start this at the beginning:

The original platform hassle

Everyone loves a good network effect. The problem is often that it is quite difficult achieving it. But without network effects, there is little or no value on the platform (unless the platform has some alternate form of standalone value).

As we so often discuss on this blog, the goal of the platform is to enable interactions between producers and consumers. Think of Dribbble, TaskRabbit or Salesforce, you see this theme repeatedly. As the platform manager, you want producers to create value and consumers to consume value for the platform to fulfill its purpose. Additionally, the platform needs some mechanism for curation, that ensures quality.

Network effects are realized when there are enough producers and consumers with overlapping intent for interactions to spark off between them. In such a scenario, the platform starts fulfiling its role of enabling interactions.

From here on, the goal of the platform is to scale its ability to enable more and better interactions.


A scaling manifesto for platforms

If we were to condense it in one line:

A platform’s goal is to scale the quantity and the quality of interactions that it enables. 

The first part is obvious but involves several nuances. The second part is less obvious and is often ignored at the platform’s own peril.

Let’s peel this further.


Scaling interactions

To start with, it goes without saying that the platform wants its core interaction to be repeated as often as possible. More rides on Uber, and more tasks booked on TaskRabbit, is good news.

The Core Interaction consists of three actions: Creation, Curation and Consumption. Depending on the type of the platform and the stage of evolution it is in, scaling interactions may involve scaling one or more of these actions.


Scaling Quantity: Creation

Some platforms scale by focusing solely on scaling creation of new value. A true blue classifieds platform like Craigslist largely cares about the depth and breadth of its listings.

A platform can stall despite having strong network effects if producers stop creating value on the platform. Thankfully, this is one of the most obvious goals and metrics to watch out for and platforms rarely fail because of an inability to focus here.

There are several ways platforms encourage a constant stream of value creation, a framework for which is discussed in detail in the article here.


Scaling Quantity: Consumption

Scaling creation and consumption often go hand in hand. This is because producers are likely to participate on a platform only when there is active participation from consumers as well. No one wants to speak to an empty room. Hence, efforts to scale creation of new value are not going to be sustainable unless complemented by efforts to scale consumption as well.

For demand-driven platforms like Elance-Odesk and 99Designs, where the value is created in response to a request, scaling consumption is a critical first step to scaling interactions. But even for a creation-centric platform like YouTube, scaling consumption is very important, especially when producers participate with a need to self-express or self-promote.

Of course, to scale consumption, the platform needs to capture better data about the consumer and use that to make more relevant recommendations. No amount of marketing and alerts/notifications can substitute relevance for consumers. This is also largely why we see the role of data scientist rapidly emerging in importance, not only in the tech industry, but also outside it.

A consequence of this is the fact that platforms need to start acquiring data about users the moment they sign up. This translates itself into simple tweaks in the on-boarding flow. E.g. Pinterest will ask you to like a few boards and a few topics as part of its on boarding flow. It’s data that helps it understand what to serve you next so that you continue to consume. Increasingly, marketplaces aren’t focused on transactions alone. Additionally, they encourage users to follow topics and listings, in the hope that such data on users’ interests can help create opportunities for transactions in the future without the user explicitly initiating one.


Scaling Quality: Curation

Finally, as more value gets created and consumed on the platform, the platform also needs to get better at its ability to curate and differentiate high quality from low quality. To ensure value and a desirable experience to its users, the platform needs to ensure that it encourages actions that result in high quality creations and discourages actions that result in low quality contributions.

In its early days, YouTube ran competitions where the most upvoted videos were rewarded. Wikipedia blocks IPs and accounts which generate a lot of suspicious activity. SitterCity and TaskRabbit do intensive background checks on service providers on their platform. Whether social feedback or editorial pre-screening, curation is a necessary part of running a platform.

Curation is typically done in one of three forms:

1. Editorial: An editor, admin or community manager approves and disapproves contributions to the platform.

2. Algorithmic: Algorithms take decisions on what’s desirable and what’s not based on certain parameters.

3. Social: The community curates through signals about quality, like rating, voting etc.


Scaling curation may mean different things depending on which of these aspects are being scaled.

Scaling editorial curation: The brute force method to scale editorial curation, traditionally, has been to get more editors on board. That never works well for a network effects platform. Editorial actions scale only when they are gradually moved out to the community over time. The editors do not become redundant, they simply take on more abstracted roles. This may involve educating the community on how to curate and ensuring that the tools of curation (e.g. rating, review, reporting etc.) are being used correctly and often enough. In the case of platforms like Viki and Wikipedia, it may also mean creating a hierarchy of sorts in the community to differentiate highly reputed users from less reputed ones to phase out actions gradually, from internal editors to highly reputed users, and so on.

Scaling algorithmic curation: Very briefly, algorithmic curation may scale by improving the algorithms themselves or improving the inputs to the algorithms. The inputs to the algorithms are provided by editors or by the community and hence scaling algorithmic curation works very closely with scaling the other two forms of curation.

Scaling social curation: Social curation scales either by permeating a reputation model through the community or by merely relying on the strength of numbers. In the former, the opinions of experts are given more weight than that of novices. In the latter, all opinions count towards the same. This is of course a continuum rather than a duality and most platforms lie somewhere along the continuum. The more the expertise required to make a judgment on curation, the more likely is it to rely on a reputation model. Users who have curated well in the past have a greater say in future curation.


Scaling Quality – Overall Governance

Finally, This specifically involves looking for corner cases and situations where the platform is used/abused in ways that the platform creator hadn’t planned for. Specifically, this involves identifying undesirable interactions and ensuring that those are not repeated. A murderer using a dating site to find his next victim is an undesirable interaction as is a contributor defacing the Wikipedia profile of a public figure.

In some cases, scaling quality through governance may have limits. No matter how much you govern, you may still have the odd traveler trashing an Airbnb apartment. In such cases, creating centralized trust mechanisms becomes critical to scaling the platform and ensuring widespread adoption among mainstream users. This brings us to our final point on scaling platform activity.


Scaling Quality – Mitigating Interaction Risk

The final element of scaling quality of interactions has to do with the risk inherent to either side in participating in an interaction. In an age where people actually die when meeting people through Craigslist, mitigating risk becomes an important aspect of driving platform adoption. Not all interactions are created equal. Some are riskier than others. Participating on Twitter doesn’t involve much risk for either side but participating on a platform for discovering home cooked food may have higher risks associated. Depending on the degree of risk involved, the platform may have to invest heavily in offering centralized guarantees and insurance. Most sharing economy platforms invest in creating insurance and trust mechanisms to ensure that users aren’t discouraged from participating.



There are many unique challenges to scaling platforms, not counting the many challenges of scaling a network effects platform technologically, as Twitter would well bear testament too. There are significant management challenges alone while scaling a network effects platform, which are often underestimated. Using the above as an overall framework of thought helps to scale in a manner that ensures repeatability and sustainability of the core interaction on the platform.

About the Author:

  • http://www.friv2ol.com/ Ngoc Anh

    Relax with the entertainment you here okay.

    Thank you. Wish you a very effective working days.

    Friv 200

  • https://contractiq.com Ashwin |ContractIQ|Tapiphany


    We can look at ‘expertise on demand’ platforms like Ask.fm & Zintro for answers. Their central unit is a ‘high touch’ conversation.

    Also one’s social network (in contexts like enterprise outsourcing) often is an unreliable channel to ask for introductions to producers (not all friends of yours may understand your context before making recommendations) but when it comes to evaluation, it can be turned into a good, reliable source.

    So, there are ways to scale even ‘high touch’. At scale, when there is trove of data, high-touch could potentially be scaled down (but we’ve not gotten there yet to speak out of experience).

    PS: BTW met Sridhar (ex-inMobi) a while back and it was interesting to know that you both were at Yahoo!. Small world.

  • http://platformed.info/ Sangeet Paul Choudary


    This is a very interesting comment. I agree that the unique points that apply to your marketplace make it difficult for social curation to work. How would you scale such a match-making model though?

  • https://contractiq.com Ashwin |ContractIQ|Tapiphany

    Sangeet, in marketplaces where the consumption is a business service (like ours – ContractIQ – An enterprise version of oDesk) curation based on consumer ratings/comments becomes unreliable, as the ratings are often with-held in the case of negative experiences, as there is some lock-in to the vendor.

    If someone builds an app for a customer, chances are that he has to go back to the same producer to fix it in the future (or) at-least keep such an option open.

    Open criticism or bashing or negative ratings become increasingly difficult when the value and the length of the transaction grows (which is the case in enterprise-y outsourcing)

    Governance and mitigation of execution risks are also difficult in B2B service marketplaces unlike say a taxi-sharing economy – If a taxi driver fails (execution) you can send another and the loss is minimal and revocable in many cases unlike in outsourcing transactions where the damages linger for quarters if not years and there is very limited you can do to govern the execution and mitigate execution risks.

    Even if I go so far to create an ‘execution insurance’ product that I sell to consumers, it only compensates for monetary loss in the transaction and not the ‘time to market’ loss.

    The philosophical (and unconvincing) argument is that the levels of curation of quality, quantity and execution even if they are imperfect, are better than ad hoc transactions outside the marketplace that fail more often.

    One of the reasons we still run ContractIQ as high-touch match-making service (apart from giving the option for the producers and consumers to do it themselves) is that, the lack of documented data can be compensated by inferred insights of an expert on the other side.

  • http://platformed.info/ Sangeet Paul Choudary

    Absolutely, Steve! It’s rarely an either-or on curation. THere are multiple points of curation on a platform and you can have a combination of different curation mechanisms at all these points.

  • http://www.linkedin.com/in/stevenomchong Steve Nomchong

    Thanks Sangeet for another insightful article.

    Re scaling quality through curation, I’d suggest that a more robust strategy is to combine more than one of the forms you outline rather than an either/or approach. For example, use editorial curation to ensure access to only high quality producers to the platform and social curation to provide the measure of ongoing quality.

    Of course, this approach will not be applicable to all models but may be appropriate in a case where the producers are established businesses for which the platform initially provides an additional distribution channel AND they are able to grow with the growth of interactions on the platform.