title

date

category

slug

summary

description

cover

showtoc

Have a Startup Idea? These 5 Tips Should Help You a Long Way Through Your Journey

2020-04-01

Business

these-5-tips-should-help-you-with-your-startup-journey

Consider this list opinionated tips before you set foot onto your AI startup. I assure you a thorough read will give you a better perspective to follow.

image	alt	caption	relative
https://miro.medium.com/max/1458/0*BDlSxjeXPGScC-Lm	Crumpled paper inside a bin	Photo by Steve Johnson on Unsplash	false

true

Most startups fail. In fact, 90% of all startups are bound to fail. They fail to even hold on for the first few years, reports Patel on The Forbes. [1]

And believe it, startups related to Artificial Intelligence are most susceptible to failure. Marketing to “revolutionize” the market adds up even more to that potential threat. Marketing AI startups as revolutionary are so rampant now, the entrepreneurs should feel considerate about their pitch!

As countless AI startups fail, investors grow wary of casting away their capital. Check out what a panel of investors had to say in an interview about investing AI startups — AI Startups Will Fail For The Same Reasons Other Startups Do.

Here’s one excerpt from one the interviewees [2]

We focus on the business first before we even look under the hood at the AI. --— Jean Xin, Investor at DCVC

So, you’ve got a plan involving AI & you believe it’ll be profitable? Great.

Do you think it’ll be successful? Do you think investors will be willing to vest their money into your idea?

Well, hard to say. There’s a lot of variables involved to answer that question. Hence, throwing around the term “AI” into the pitch will not buy anyone in. Your strategy & plans have to be rock solid & foolproof.

Thus, I curated a list of suggestions, your business idea will gain from.

Identify the Root Cause & Analyze If AI is the Solution For It

Most AI startups are not going to fail on the basis of their actual AI, They fail because they fail to identify a problem that needs solving. --— Flomenberg [2]

As Flomenberg states in their interview with VentureBeats. Startups fail not because of the actual implementation of the AI system/product. Rather it goes downhill when the company fails to identify the core problem it’s trying to provide a resolution.

In fact, it shouldn’t come as a surprise if the funding gets rejected the moment you mention AI in your product.

So my advice is;

Focus on identifying what your customer’s problems are & act.

Quite often, you wouldn’t even need to put in AI into your product. As suggested in the xkcd strip, it’s not easy to apply Machine Learning to every simple problem. Ask yourself, would you use a forklift to pick up a matchbox?

Figure Out The Right Implementation for the Solution

You’ve identified the root cause of the problem?

Wonderful. So what now?

You see, the field of Machine Learning is evolving faster than one can comprehend. Less than a decade ago, CNNs could classify between a dog & a cat, 40% of the time! [3] Right now, even a layman can build a simple Neural Network (NN). A few lines of Python code & voila classified 10 hand-written digits!

Not to forget the power of Transfer Learning & cheap, available cloud resources to train our models. Opportunities like these made our lives easier. Now, all that’s left to do is to spin up a remote GPU instance, download a pre-trained model on it, train our model with Transfer Learning and be done.

Mark my words, this way you would still achieve a surprisingly high accuracy rate with minimal effort & knowledge. So if this is not the perfect cue to build your business right now, then I don’t know what is.

Hence, second on the priority list?

Figure out the tools & resources already available at your disposal. This will be pivotal for your business’s finances & on its long-term growth as well.

Should You Compromise, Speed Over Accuracy or Vice-versa

Training a model takes time, often days if not weeks! For example, it took Airbnb 3 days to train their model for Object Detection on images of the listings from their hosts. And they achieved an mAP of 50% which is very good! [4]

Surprising isn’t it? For a company with the time & resources, we assume it wouldn’t be much of a hassle for them to train & deploy an ML model.

On the brighter side though, progress in ML research has come a far way in the past decade or half. CNNs built for deploying on mobile devices are being developed & open-sourced as you read this article.

MobileNet is one such CNN. Termed as an “Efficient CNN for Mobile Vision Applications”, they enable the user to train their models with a relatively lower accuracy but at the cost of extreme speed. [5]

So coupled with the power of Transfer Learning, training your model on MobileNet can be exceptionally efficient.

But on a side note, the Accuracy of your model might be paramount to your business. It could be a matter of going kaput the day after training the model it doesn’t perform accurate enough. For example, high stake automated trading is where the consumers of your product will be reliant on the accurate prediction of your software.

So the takeaway here? The type of product/service you want to provide will decide the degree of compromise to make between either, Accuracy or Speed. [6]

Production Is As Important As Marketing the Product

Let me share an honest confession. Based on personal experience, I can vouch, many of you entrepreneurs either get too lost in marketing the product or ignore it. So my advice? Divide 50/50 of your time & efforts into both.

Get it straight, No Product == Nothing to Market, No Marketing == No Awareness About the Product.

Personally, I strongly believe in at the least developing a Minimum Viable Product & analyzing the response from the consumers based on that. Besides, it’s so much easier to gain a prospective investor’s attraction, as now you wouldn’t have to just sweet talk about your plans but rather show the product(s) first-hand!

Thus my fourth advice. Get out of your “[Donkey Problem aka Buridan’s Ass” situation & just work on creating a feasible product first & foremost. Everything else is secondary & will eventually fall into its place.

Do You’ve Enough Data

Perhaps, the most important factor in the decision-making process.

Deep Learning models are infamous for requiring huge amounts of Big Data. Even though the reliance on a big data set to train the model is changing with the advent of Transfer Learning. It’s safe to assume that for your use case, training a model from scratch might be necessary.

Thus the question “Do you’ve enough data to train the model on? If not where/how are you going to get it?” arises.

The response from the remove question & the segment your business serves in might add up further operational overhead. So it’s essential that you consider the available options here before diving head-on.