5 things I learnt after 1 year in an AI automation startup - KeyReply

In the past year, I watched the startup i joined grow tremendously from a 3 person team to now a team of 20 members. Throughout this journey there were many challenges that we had to overcome. Engineering resource was always lacking, our customers were always requesting for more, and there was nobody really experienced to guide us. The average age of the company (even now) is less than 30, so we just had to learn to figure things out along the way.

To overcome these challenges, we employed several strategies to help us punch above our weight. We try to leverage as much tools and technology to reduce our effort and allocate our time and resources strategically to maximise the impact we bring to the company.

Scour the web

Our CTO has always reminded us that it is very helpful to maintain awareness of the latest and greatest out there, so that when the need arise we already know what we have in our toolbox.

Starting from a team of just two software engineers, we needed as much help as we could get. With things in the technology space moving so quickly, there is a myriad of open source libraries available that could help us achieve what we need twice as fast, ten times as good.

For example, we can easily add a neural network on the browser, or easily add polling, publish-subscribe, and optimistic UI to our web application through GraphQL Apollo (so that it becomes alive, like Google Docs). These are features that traditionally take up costly effort to design and develop but can now be done by your typical developer.

By keeping up-to-date with all the available resources out there, we were able to quickly churn out great and interesting features without having to sleep over in the office.

Leverage the cloud

We were able to scale quickly because Azure is just so powerful and easy to use. Many of our clients are large enterprises, so we had so much security and infrastructure requirements to meet. Database redundancy and backups, monitoring and alerts for application health, application logging, provisioning of SIT/staging/production environments, all within a few clicks.

The speed and maneuverability at which we fulfil the infrastructure requirements has sometimes surprised some of my enterprise clients’ counterparts. “So fast?”

“Perfection is the enemy of profitability” - Mark Cuban

With so many things to do, we always have to prioritise our time so that we are always creating the most value for our time. It is always tempting for us to want to spend more time touching up on our pet feature, but at the end of the day we have to admit that small finishing touch may not even matter to most people. So, the 80/20 rule is always at the back of our head. With so much work waiting for us we knew we could not afford to spend time on low impact work.

We understand that we are not perfect, and that we need to get help for things that we have trouble with. We sometimes find that insightful opinions or suggestions are brought in by someone we do not expect, such as someone from another functional team. Inclusive teamwork helps us ensure that our blind spots are covered. If someone else can do it twice as effectively as myself, why not let him or her handle the job while I focus on what I am better at?

You don't need a PhD

The most academically qualified person in our company is our CTO, who has a Masters in Applied Information Systems (Data Analytics). However, we have products that can match or sometimes exceed the capabilities of some our competitors who house highly decorated data scientists.

We realised that this is because leading researchers in tech giants such as Google have already done the work and even published their work for free. If Google, the company who handles 90% of our search queries, provides Google Books service, and created the best performing AI Assistant provides their research for free, why not just use their work?

AI is just a tool. What is your business case?

In the end, for us, technology are just means to provide value to our client. There is no point in having fancy intelligence or algorithms if they do not help our clients improve their sales, provide understanding of their customers, or reduce their costs.

Sometimes our clients come to us not knowing what they really want. We learnt from experience that if we had just blindly listened and followed their instructions, our products would not help them solve their problems. Now, we always try to understand their problem statement first before providing suggestions on how we can use AI to improve their circumstances.


These strategies have helped my company grow rapidly within the span of a year. I am sure that as we move on to the upcoming year, there will be new challenges as we start the next phase of our growth. I hope that these strategies will also come in handy for you, as they have for us.