Kiran Kulkarni

Startups and Google Cloud: A match made in heaven

By Kiran Kulkarni on July 11, 2022

Startups lead change. They are building prototypes and trying to learn quickly what works and what does not. The cloud is a tool that allows for quick prototyping without huge upfront costs. It is an ally to all the entrepreneurs out there.

Unfortunately, over the last few years, the time it takes for an entrepreneur to get his/her startup from inception to an exit (either in the form of an acquisition or an IPO) has kept increasing. It currently stands at around 6+ years. In addition, the current environment of economic uncertainty and war has made it more difficult for the startup ecosystem.

At google, I work with a program to help startups. Here are a few highlights of the program that a startup can benefit from.

Financial benefits:

During the bootstrap stage, when startups are testing out an idea, they can use Financial Credits worth two thousand Euros to test them out on the google cloud platform. For example, they can use serverless or no code technologies on the google cloud to if something works.

Enabling Entrepreneurs:

If a prototype shows promise, the entrepreneurs need an environment and the necessary skills to develop it further. Google offers a “startup school” to help entrepreneurs develop the right muscles. For example, what should their hiring strategy look like? How should they make key architecture decisions? OKRs (Objectives and Key Results)

More Financial benefits:

Hopefully the entrepreneurs use their skills and knowledge to build their prototype and get funding from angel investors. They can get additional 2 years of cloud credits worth hundred thousand dollars per year.

Higher level of support:

In addition to the financial support, Startups get support from dedicated Google cloud engineers (like myself!). We hope to help them avoid the common mistakes by bringing our learnings from our past experiences.

I hope many startups are able to avail this support and successfully cross the six year hurdle it takes to be acquired or do an IPO.

The alternative for me would be to spend many more years working with large enterprises with legacy applications and legacy mindset – and it’s not the most exciting part of my day.

If you want more information, you can find it here.

There is also an excellent podcast here, where my colleague Ryan Kiskis talks to Kevin Horek about startups and more.

Websites to Platforms

By Kiran Kulkarni on June 28, 2022

The internet started as a network of networks. We already had local networks (computers in a given building or organization connected physically) and then we found a way to interconnect the local networks with each other in one giant mesh. Hence the word interconnected network. Seen from a network perspective, the internet has 3 major components. The computers that host the content, the companies that lay intercontinental cables to bring the content closer to you, and the edge providers.

The internet initially hosted websites. It served static content. Then came commerce, or e-commerce we called it. Finally we had social media. Over time, social media monetization models have co-evolved with e-commerce models and we now have “social commerce”. Social media and e-commerce feed on each other in a self-serving feedback loop of data. It goes as follows.

The more information you have about people, the better your machine-learning process to help you understand your users and predict their needs. Better understanding leads to better emotional engagement with the users, and what will make them act. The more you can engage users, the longer they will use your service, enabling you to gather more information about them. Knowing what makes your users act allows you to convert views into purchases, increasing your economic power. This economic power can be directed to building more services on the website, restarting the self-serving loop.

As this virtuous cycle continues, the websites evolve into platforms. For e.g., Google, Amazon and Facebook are platforms. I would argue that every successful digital business is a platform. Platforms can take advantage of the network effect. As more people connect with a platform, the value of the platform increases exponentially because the information held within the network increases exponentially.

In theory, the internet is still a network of networks. But in practice, the Sandvine report says that in 2022, six companies (Google, Facebook, Microsoft, Apple, Amazon, and Netflix) generate almost 57% of Internet traffic – more than everyone else, combined. Those who are located at the center of the network can gather the most information. They can offer unique value-added services from the data they have.

How does a traditional company become a platform? The good news is that if you have the content, the intercontinental cables and edge services can be rented out.

The journey from a website to a platform is what happens on the cloud.

Leaders as change agents

By Kiran Kulkarni on June 5, 2022

In my previous post, I talked about licensing costs and rent-seeking behavior. I asked the question – why do companies pay licensing costs on older technology when the same thing is available without the licensing costs in the cloud?

There are, of course, many reasons for it. I believe an important one is the role of middle managers. My hypothesis is that the digital era companies need a very different kind of middle managers as compared to pre-digital era companies.

In the pre-digital era, the companies were producing goods and services. Workforce productivity was important. If the workers produced more widgets with better quality, the firm made more money. The middle managers were there to drive productivity improvements among the workforce. They implemented 6-sigma, kanban and other projects.

In the digital era, companies are building platforms. The workforce has shrunk and work has gotten smarter. Earlier productivity metrics are useless. The more sophisticated platform has less lines of code, not more. The focus is on Innovation, not productivity improvements. The cloud is the tool for innovation. For e.g., In a recent google hackathon, just 3 developers built the winning product in 10 days! The solution speeds up ML model deployment using BigQuery ML, stores it in google cloud storage, finally deploying it in cloud run using cloud build. That is the power of serverless computing, which is offered in the cloud.

In my experience, most middle managers of today still take their job to be that of improving worker productivity through kanbans (albeit, I must admit – they use digital kanbans these days). They focus on daily standups and set priorities. They try to mitigate operational project risks and communicate project status accordingly. Thanks to newer tools, they are left with a lot of free time, which they spend in unproductive meetings.

Meanwhile, the organization is facing an existential struggle to digitally transform themselves.

In the era of digital transformation, middle managers today should be, primarily, change agents. The rest is operational detail.

Their focus should be to become the pivot for digital transformation. I think every middle manager’s task today is to identify opportunities to introduce AI/ML features into whatever they are doing. This means building scalable platforms, modernizing the applications and building data lakes. They need to see legacy software, inflexible architectures and avoidable licensing costs as the biggest challenge to their transformation journey. Once we establish that, saving licensing costs would be actually the low hanging fruit, not a risk.

Profit seeking Vs. rent seeking behavior

By Kiran Kulkarni on May 29, 2022

There are two ways of making money. Profit seeking and rent seeking. Profit seekers increase their wealth by creating new wealth (usually through innovation). Rent seekers increase one’s share of existing wealth without creating new wealth.

The classic example of rent-seeking, according to Robert Shiller, is that of a property owner who installs a chain across a river that flows through his land and then hires a collector to charge passing boats a fee to lower the chain. There is nothing productive about the chain or the collector. The owner has made no improvements to the river and is not adding any value. All he is doing is finding a way to make money from something that used to be free.

In the software / IT business, the pace of change is so astounding, that cutting edge softwares of the last decade have become commodities today that can be offered almost for free. Take for example, relational databases. They were pieces of innovative software in the past, but today you can get a fully managed relational database for free.

In the cloud, you do not pay anything for a relational database except for the computing costs. That is because the cloud providers have automated most of the management tasks and the open source versions are quite stable and reliable.

The licensing costs of a typical on-premise database are a significant component of the overall costs – sometimes up to one third of the total costs. These costs are a classic example of rent-seeking behavior. They are able to do it because the databases are linked to operational (day-to-day) business and companies find it tough to replace them easily.

To me, it is not surprising that the companies engage in rent-seeking behavior. What is surprising is that their clients allow them to get away with it. Unlike the river or the railroad, software is not a physically scarce resource. The companies can switch to a license-free cloud version anytime they want. Heck, the cloud providers may even pay for the migration costs.

Even more shocking is that they realize that they will have to do switch at some point anyway. Unlike their software vendors, their customers will pay only for the value that they get. They still need to deliver value for their customers.

Cyber attack: The side effect of abstraction

By Kiran Kulkarni on March 13, 2022

In 2015, just before Christmas, the lights started going out in Ukraine. There was a cyber attack on its power grid, plunging entire neighbourhoods into darkness. Unlike conventional wars, cyber attacks offer dictators the option of plausible deniability. The IT infrastructure can become a source of vulnerability.

What makes one’s environment susceptible to hackers?

Unlike popular perception, 99% of our software vulnerabilities come not from the code that we write. They are embedded in the software supply chains that we created. Software supply chains are the outcome of the abstractions we create as the industry matures over time.

The evolution of computing can be seen as a trend towards abstraction. We first went from a physical machine to a virtual machine, where a piece of software abstracted parts of a physical machine. The operating system abstracts the underlying virtual machine and offers us and the applications a neat interface to work with. More abstraction allows us to buy targeted softwares. Our code is eventually the top most layer in a complex interaction between several layers until it reaches the physical bits and bytes.

Think of it like a car. I interact with the steering wheel, the gas and the brake pedal. Whatever happens below the gas pedal is abstracted away, so that I can focus on the traffic.

Unlike the automobile industry, the technology world is still nascent. Standards continue to evolve rapidly. Older, outdated standards become a vulnerability as time progresses.

The cloud can help in the following ways.

Buy everything as a service

If you are a small company, buy everything as a service. You write your own code if you need to customize it, but beyond that, you do not have to worry about the supply chain vulnerability.

Back up in the cloud

If you have a legacy infrastructure, you can create an alternative, relatively clean infrastructure in the cloud for the most critical operations.

Use the cloud organization policies to control your vulnerabilities

In the cloud, you can write your infrastructure as a piece of code. You can set policies that prohibit the use of softwares with known vulnerabilities.

Make updates less risky

It is not laziness but fear that stops organizations from regularly updating their infrastructure and legacy applications. The applications are interconnected. There is little visibility of their interdependencies. Hence the maxim “If it ain’t broken, don’t fix it”. The cloud can offer a parallel “training ground” for you to test your update and the interdependencies before you implement it.

A horizontal plane for vertical growth

By Kiran Kulkarni on January 30, 2022

Peter Thiel, in his book “Zero to One” differentiates between horizontal and vertical growth. He refers to horizontal growth as something that builds or expands on the existing ideas and innovation (going from 1 to n). Vertical growth, on the other hand is about creating something new, something that did not exist earlier (going from 0 to 1)

By definition, almost all start-ups are trying to execute on a vertical thought. They are a bet on a future that exists only in the innovator’s mind.

The cloud – with its scalable, on-demand computing platform – is an ideal horizontal plane on which the vertical progress can be made. It offers speed, scalable costs, real time insights and a host of software vendor partners to supercharge the growth of the startup.

But to succeed, a startup needs more than a modern computing platform. It needs access to a global startup community for its infectious energy, events to spread the word, accelerators to cross-pollinate ideas and mentorship to avoid the most common missteps.

Recently, Google cloud launched a program to provide all of this and more. The details can be found here.

We live in a world of easy access to capital. Every halfway decent idea will get funded. The successful ones will be those that have access to an ecosystem where ideas can be nurtured, tested and fine-tuned. The google startup program intends to offer such a nurturing ecosystem. It is the same ecosystem that generated 9 products with a billion plus users in the last 2 decades.

Data driven Innovation: How enterprises are hamstrung by their pre-cloud architectures

By Kiran Kulkarni on January 23, 2022

When Doordash did an IPO in Dec. 2020, the food delivery company was valued at more than 70 Billion USD, more than the combined market capitalization of Domino’s Pizza and Chipotle.

Investors believed that Doordash could innovate much faster because it had data driven decision making. Let’s peel the layers to see why a traditional enterprise is hamstrung competing with a young startup.

Limited access to Insights

Every organization would like more decisions that are based on data. If a store manager had access to a predictive forecasting model output every day, she would be much more efficient in inventory planning, staffing etc. Furthermore, she could automate the restocking of cola (a low-impact decision) and focus on higher impact decisions such as upselling.

In practice, the required data and analytics is not accessible to a large majority of people in the way, shape and form that can be useful. A very few people have access to analytics.

Manual processes

What stops these enterprises from making data driven insights accessible to all of its employees? The reason is that the process of building insights is very manual. A great majority of analysis are diagnostic dashboards built by running batch jobs on data. The predictive analytics teams serve higher level strategic decisions for which the impact is in quarters or even years, not days and weeks.

Data silos

Why are processes slow and manual? Because the data exists in silos. I do not mean just the physical location. I mean siloed use cases. In the pre-cloud era, organizations built different systems for different use cases. For e.g. there is one system for online transaction processing (OLTP) and another for analytics warehouse (OLAP). Similarly, there are different systems for Streaming Vs. Batch analytics, for fast and cheap vs. secure and accessible data. This creates silos – across products, across clouds and across teams.

The cloud breaks down these data silos.

For example, in Google cloud, you can automate the process to ensure that all changes to the transaction systems (change data capture or CDC) are updated in the central warehouse, BigQuery. You can also seamlessly query from BigQuery, any data sitting in any other database in the cloud. The data silos between the systems are broken down.

The innovation leaders are going one level further. They are setting up data exchanges (called Analytics hub on google cloud) where subdivisions publish their data and other sub companies and vendors subscribe to it. This gives them the opportunity to include their vendor ecosystem to drive innovation.

Deconstructing machine learning models

By Kiran Kulkarni on January 10, 2022

Most executives that I speak with – from large enterprises to startups – mention that experienced data science talent is the biggest bottleneck to their machine learning efforts. A lot of them imagine the process of building a model as a complicated, intellectually challenging process that requires very experienced folks. They see movies like matrix and visualize their brilliant data scientists typing away fervently at their computer terminals.

They would be right, if this were 2012.

In the intervening decade we got the cloud. And while it has improved a lot of things, I would argue that its biggest impact has been the democratization of the machine learning process.

Take the google cloud, for example. Google cloud has a petabyte scale analytics data warehouse called BigQuery. It offers a service called BigQuery ML. Users can create and execute machine learning (ML) models in BigQuery using standard SQL queries.

One can build a model with a simple query in 3 lines of code.

CREATE MODEL numbikes.model
OPTIONS
(model_type='linear_reg', labels=['num_trips']) AS
WITH bike_data AS
(
SELECT COUNT(*) a num_trips,
...

One can make a prediction with another query.

SELECT predicted_num_trips, num_trips, trip_date
FROM
ml.PREDICT(MODEL 'numbikes.model...

In just a few lines of code, your data analysts can build a prototype. I used linear regression in the example above, but it could be logistic regressions, clustering algorithms, deep neural networks – probably 90% of all the algorithms that we have available in the public domain.

Using service like BigQuery ML, teams can focus on building prototypes instead of writing machine learning algorithms from scratch. The product managers can evaluate the prototypes to identify the most promising ones. The data scientists can then focus on fine tuning the most promising prototypes.

They say all good writing is average writing that was rewritten. All good models are prototypes that were fine-tuned.

The Impact of the cloud

By Kiran Kulkarni on December 19, 2021

People who have a datacenter usually compare their datacenter running costs with renting the same thing in the cloud. They are confused when I tell them that they are comparing apples with oranges (or pears, if you speak german). The reason is, that you would never run your applications in the cloud the way you would run it in a datacenter.

When you use a cloud provider like google you don’t just rent machines, you also use the services and automation features that google has built for itself over the years. One such feature is called containerisation, which packages software code in a way that it can run in any environment. Google offers containerisation as a server-less service called cloud run on the cloud. You do not need to configure or maintain the computing resources you need. Just write your application and give it to google and it will provision the required computing.

We need a different framework altogether to understand the financial impact of services like cloud run, since they are not available on premises. A recent forrester study interviewed people who moved from their datacenter to the cloud run service and found some interesting things.

95% faster deployment, saving 180 developer hours

The IT teams mentioned that deploying a new update was done in a matter of hours on cloud run, where it took them weeks earlier.

98% fewer critical errors

Prior to using Cloud Run, their organisations had consistent issues with services not being able to scale to meet changing traffic demands, and this caused critical failures and direct loss of revenue and profit. With Cloud Run, services were able to scale quickly, and the number of critical errors due to platform issues essentially dropped to zero.

50% more efficient labor

Approximately half of the employees who previously monitored services were able to transition to other activities due to the automation that cloud run brought.

40% reduction in developer recruiting costs

As the word spread about the use of automation features spread, more developers joined from the networks of existing developers, thereby reducing recruitment costs

Avoided costs of pre-provisioned and on-premises platforms

The usage costs for cloud run were up to 75% lower than on-premises platforms.

These are, of course, just the quantifiable benefits. There are, in addition, several unquantifiable benefits.