My Investment Takes: LandingAI – Artificial Intelligence and a data-centric approach

My Investment Takes is a weekly dive into the startups I find the most innovative & interesting.

Do you use data on a regular basis? 

If your answer is yes, its capabilities will not be lost on you. Over the years, society has found hundreds of ways to use and abuse this capability.

For better or worse though, data can be important, useful, and insightful, as long as its quality and quantity uphold some standards. After all, if you decide to build an algorithm running on data that is irrelevant and using fundamentally incorrect assumptions, you won’t be able to go very far. You then have two one of two solutions: spend hours and money to clean up the data – and believe me, it’s a very frustrating thing to do: around 80% of a data scientist’s time is spent cleaning data.

Or, you can take a look at the start-up I decided to focus on this week.


LandingAI is a US-based start-up which has created a platform designed to help other companies to become AI-driven. Its founder promotes the concept of ‘Data-centric AI’, emphasising the quality of the data over the optimisation of the AI model.

What is AI?

Let’s give some clarity

Artificial intelligence, deep learning, Big Data, machine learning, algorithms, models and other glossy buzzwords used by startups and VCs on a daily may sound impressive, but the simple truth is that most people don’t really know what it is. So let’s start with some clarification.

I initially tried coming up with my own definition of AI, but seeing as I’m not an expert in the field, it seemed more apt to use one already created by one of the foremost experts in the field, Dr Kai-Fu Lee:

Artificial intelligence (AI) is smart software and hardware capable of performing tasks that typically require human intelligence. AI is the elucidation of the human learning process, the quantification of the human thinking process, the explication of human behaviour, and the understanding of what makes intelligence possible.

Now that we’ve established what AI is, how about deep learning and machine learning? Deep learning is a sub-field of machine learning, which is itself one of the sub-fields of AI. The biggest difference (if you can call it that), is human interaction.

Let’s start with an example of Machine Learning. Say you like Jazz, and you use a music streaming service that tailors its suggestions to your taste. The service will come to learn your taste, and continues to suggest several other Jazz records on top of those you already like. The data it feeds on is yours and it will improve over time. But if at some point, to your dismay, the service starts recommending Reggaeton, you can report it and an engineer will step in to make adjustments.

With deep learning on the other hand, human interaction is almost not necessary. The idea for it came from our own brains and in some ways, it replicates our web of neurons with artificial neural networks composed of thousands of layers. Engineers will only meddle with two of its thousands of layers: the input and the output. If you want for instance an AI to become the foremost expert on recognising purple pebbles (God knows it’s needed), you’ll feed it millions of images as either ‘purple pebbles or ‘not purple pebbles’, and the machine will adjust its parameters to maximise the chance that a correct result is achieved.

All this to say: data is critical. For each deep-learning AI, hundreds of thousands, millions, and sometimes billions of data points are necessary. And that is mostly why early tech giants are so powerful: they have an enormous amount of data, and they know how to use it.

Is all data created equal?

Short answer: no.

Well, internet companies have such vast quantities of data that it doesn’t matter much: they will get the quality data they need one way or another. And to be fair, they have been widely successful at using it: we are all spending more and more time on our screens and on social media thanks to the AI optimised from our own data.


Other companies are also data-rich, and they’ve started to realise it. That is mostly why we’re seeing Fintechs and Insurtechs being created every day, and more established players shift their focus towards a tech-dominant future.

However, there are other industries that have yet to make the shift, including anything related to producing physical products (cars, medical devices, packaging, etc.), the main reason being the lack of data points.

And that’s where LandingAI comes into play with its data-centric approach.

What I love

How they are doing it

‘Data-centric’ might sound like I’m adding to the sea of tech-y buzzwords, but it’s actually relatively simple. Rather than focusing on the quantity of data, it focuses on quality with clear labelling, complementing this with human subject matter experts. And the great thing about it? Academic research found that data quality makes a greater impact than AI optimisation.

LandingAI caters to the production industry such as manufacturing, automotive and electronics, ones that are heavily impacted by the cost of quality control. Visual inspection, while using simple machine vision solutions to analyse products, still heavily relies on human inspection. Results? Large upfront costs, less accuracy, production and accidental selling of defective products that will need to be recalled, draining even more resources.

All in all, not great.

LandingAI deploys its easy-to-use data-centric platform, LandingLensTM in the very short term. There’s no need to hire teams of developers and to teach and feed an AI a ton of inaccurate data to learn how to do its job. The idea is to improve the inspection accuracy of their clients products, to do so with the clean, non-noisy and limited data provided, and to keep the model flexible to changing standards and metrics.

Domain experts can now build an AI system with a few mouse clicks in a day, basically rendering the whole process low-code/no-code.

The path forward

LandingAI raised a whopping $57 million Series A in November 2021, only a year after the launch of their flagship product, LandingLensTM. The round included McRock Capital, Taiwania Capital, Insight Partners, CPP investments, Intel Capital, Samsung Catalyst Fund, Far Eastern group’s DRIVE Catalyst, Walsin Lihwa and the AI fund.

The money raised will help scale the team further and make their product better. The startup claims that Product-Market Fit has been found.

Of course, it does help that the Founder, Andrew Ng, is a world-renowned AI specialist, co-founder of Coursera, and founding lead of Google Brain. Not too shabby as far as CVs go.

The Questions I have for the Team

• Are there new products in your pipeline you plan on releasing?

• Which pain points do you think you’ll encounter as LandingAI develops?

My Conviction

LandingAI caught the early winds in the Data-centric AI race, and I believe they are very well positioned.

They have an impressive team, a product showing traction, and clients desperately needing it to stay competitive in tomorrow’s world. That’s a winning combo.
Other startups have also received interest, such as Scale AI, Defined.ai or Snorkel AI, and it seems like others might join the race as things pick up and more innovators realise the incredible potential the method has.

If you’re excited to check them out, take a look at their website https://landing.ai, and leave a comment on this post if you have any thoughts or suggestions!