10+ Questions you Should ask Yourself Before Developing an AI Solution

Photo by  on 

So, you’d like to spice things up in your current business strategy and add an extra layer of high technology with a sophisticated AI solution. Let me just ask you one simple question first:

Are you sure it is the best move?

If your eyes sparkle with excitement as you think “Yes!”, you can skip this article and go find your perfect AI vendor.If, however, you have even a shade of doubt — stick around. I promise that you won’t regret it!

What I have for you is 10 questions answered by Miquido’s Head of Innovation that should help anyone understand whether adopting AI technology is the right move or financial suicide for your business. Ok, no more waiting — here are the questions:

1. Can you state your problem clearly?
2. Can you solve your challenge without AI?
3. What do you understand by “doing your task right”?
4. Do you accept that your solution will never be perfect?
5. Do you possess enough data to get started?
6. Is your data relevant?
7. Will you have continuous access to new data?
8. Can you label your data correctly?
9. Do you need an interpretable model or is accuracy enough?
10. Do you have enough resources to keep the project going?

Sound simple enough? Well, it gets a bit more complicated. So we’ll look into each of them in more detail in just a second.Get comfy, grab something to take notes with, and go along with this article, or simply save it for later.Let’s dive in!

1. Can you state the problem clearly?

A clear vision of what you want to achieve is a must when it comes to applying AI in business. You need to have a specific problem that requires high-tech solutions that can be solved with Artificial Intelligence and/or Machine Learning.

Introducing AI for the sake of bragging about having it won’t fly here.

To help you answer this question, try thinking about specific challenges you want to address and always have your target audience in mind. Ask yourself:

• Which repetitive tasks are you trying to automate?
• Who will benefit from this automation?
• Will it be worth the investment?

Pro tip

If you’re not sure where to start — start with data mining. Its ultimate goal is to help you get inspiration and once your idea is stated, you can come back to Machine Learning. Focus on what you can have right there and right now — the cornerstone of Artificial Intelligence, data.

And then ask yourself the next question:

2. Can you solve your challenge without AI?

Artificial Intelligence is great because it offers seemingly simple solutions to complex problems. The only issue is that, in fact, it’s far more demanding than it appears. So the ugly truth about the development of AI technology is that if you can get on without it — just do it.

Consider these questions:

• Is your problem complex enough to engage machine learning?
• Could it be represented by a mathematical equation?
• Can you think of a step-by-step recipe for an output?

If you answered “yes” to the last two questions — congratulations, you don’t need artificial intelligence!

If, however, you couldn’t quite wrap your mind around the correct patterns, advanced machine learning algorithms might be your only chance to succeed.

Pro tip

It’s quite simple, really: if you can recognise the pattern yourself, you don’t need Machine Learning to do that for you.

By now, you should have a pretty good understanding of whether you actually need AI in your strategy or if you simply want to follow the trends. Let’s see if Artificial Intelligence actually can help you out.

3. What does it mean to do your task right?

First of all, you’ll have to define what “right” and “wrong” mean to you and your business. The world of technology is still pretty binary, and if something isn’t “true”, it is false by definition.

So, before investing in any solution, make sure you understand what it is you hope to gain from it.

Artificial intelligence is, first and foremost, a complicated algorithm. And in order for it to learn, you need to be able to evaluate its performance. Think along the lines of:

• How will you evaluate that a task is done correctly?
• Which mistakes will be more harmful than others?
• How many mistakes per 1000 results can you afford to have?

That’s another big revelation about working with AI: mistakes are unavoidable.

Pro tip

Be careful not to aim too high with accuracy. Placing an excessively high bar may result in you missing some profitable opportunities.

That brings us straight to question #4:

4. Do you accept that your solution will never be perfect?

If your reply to this question is “no”, I’ve got some bad news for you: you’re not ready to work with machine learning. There will be mistakes. Sometimes more, sometimes less, but there’s no chance in the world for your solution to run error-free. Even if you have Elon Musk on your team.

At this point, think about yourself, your mental health, and the daily struggle you’ll have to accept from now on in your business:

• Can you live with occasional mistakes in your model?
• What are the actual consequences of such mistakes?
• What does it all mean from an ethical standpoint?

Once again, there are two potential outcomes here:

• You’ve evaluated the risks correctly, talked yourself into accepting them, and are on your merry way to hiring an AI software development company.
• You’ve reached the conclusion that the stakes are too high, so you can either give up on the idea of working with machine learning altogether.

Pro tip

To minimise the chance of costly and dangerous mistakes, simply have someone who will double-check the results. This is known as a “human-in-the-loop approach” and can save you some headaches.

With that in mind, you must have a pretty good idea of whether or not the AI onion is worth peeling. Now it’s time to see if you’ll be able to actually build your model.

5. Do you have enough data?

The most common and broadly discussed limitation of Artificial Intelligence is its heavy dependency on datasets. There is simply no machine learning without data.

To put it bluntly: if you don’t have the data to keep your project running, your chances of launching it in the first place are slim. Some questions that might help you think are:Do potentially useful inputs even exist? Can you gain access to them (e.g. build them, buy them, etc.)?Do you have enough examples?

Pro tip

It usually takes at least 10 thousand samples when training the model from scratch. Yet, the more examples you have the more reliable your AI model will be, and don’t we all thrive for perfection?

In data science, however, it’s both quality and quantity that matter, so if you do have enough materials to proceed with, let’s check if they hold any value.

6. Is your data relevant?

As you could’ve guessed from the previous point, without having a well-thought-through plan for data collection, you’ll stumble across multiple serious issues with your AI solution pretty quickly.

The best way to ensure it doesn’t happen is to double-check the actual relevance of your data.

There’s no point in having 1000+ features with no practical use, they’ll only eat up your precious storage space. The important questions to keep in mind to ensure that you’ll only work with the highest quality data points are:

• Are your data points significant?
• Is all the data clean enough?
• Is the data you have relevant to your target audience?
• Is it free of bias?

Pro tip

To learn from examples, AI needs good examples to learn from. In order for everything to work in an orderly fashion, you need to ensure your sample data is well-balanced, clean and free of inconsistencies.

And if you’ve got that covered, it’s time to think about the further evolution of your AI solution.

7. Will you be able to access new data continuously?

We are almost done with data questions, I promise! But while we’re on the subject, let’s think about the scalability of your project.Having an accurate initial model might suffice for obtaining reliable predictions for a relatively short period of time. Yet, you’ll soon notice there are numerous factors that may negatively affect its performance. These threats include social events, seasonal changes, shifts in demographics, the geographical location of your users, etc.

When there are factors that can influence your target audience over time, your AI model needs to be constantly retrained.

Here, you will have to face new challenges and try to answer questions like:

• How sensitive is your dataset to changes?
• Do or will you have access to new data continuously in order to update it?

A great example of such unexpected threats is Covid-19. It has pretty much rendered a massive amount of data obsolete, as people’s behaviour has changed drastically.

Pro tip

Make sure that your AI solution will be weather-proof. Or, in case of a disastrous setback, like a worldwide pandemic, at least can be easily updated.

That brings us to the final question about data.

8. Can you label your data properly?

We’ve covered some paramount issues like obtaining, updating, and navigating Big Data, so now it’s time to talk about data management.

In order for your AI solution to correctly understand the data, it requires proper labelling. Some datasets, such as image recognition systems, inherently contain labels based on the logged user actions. However, if your plan is to build a classification from the ground up, you’ll need to come up with a system for correct data labelling.

The important questions to think about here are:

• Do you need to label your dataset?
• Can you get a human to do it?
• How much time and money will that process require?

Pro tip

When working with advanced data, such as ECG signals or medical images, hire experts to correctly classify each case before your model could learn from examples.

With that, we’re ready to move on to the next question!

9. Does your model have to be interpretable?

Interpretability is one of the primary issues with machine learning. But what does it even mean?

In the simplest terms, the higher the , the easier it makes it for a human to understand the processes behind the algorithm’s decision making.

It is crucial for some businesses to fully understand the flow due to external policies and regulations. However, more often than not, having a model that is accurate yet not entirely interpretable, is enough to get your AI project up and running.

So, the two major questions that arise here are:

• Are there any regulations governing your model’s interpretability?
• What is your trade-off between accuracy and interpretability?

Pro tip

Testing helps to launch many projects regardless of their complexity. If you can’t explain how something works, run as many tests as needed to prove that it does work.

Looks like it’s time for the final and the most important question you should ask yourself before committing to an AI project:

10. Do you have enough resources to implement & maintain your AI solution?

As you may already know, machine learning projects aren’t particularly cheap, are not that easy to implement, and require an experienced team behind the wheel. So before you jump into your next idea headfirst, try to evaluate with great care whether you have enough resources to really pull this off.

Some of the questions that might help you at this stage are:

• How much does building and updating the database cost?
• Do you have access to enough processing power?
• Are you aware of the costs of training and retraining your model?
• Are the expected benefits higher than the estimated costs?

Depending on the complexity of your model, all these numbers can vary drastically.

Pro tip

Go with Cloud, as these solutions generally tend to be much more cost-effective than self-hosted solutions.

Final remarks

With all that in mind, you should be perfectly ready to make a final decision regarding your AI project.If you’re still up for the challenge — that’s great news! All there’s left to do is to find the right vendor to deliver your AI solution.And if you’ve discovered it may not be the best time for one just yet — don’t worry! The world of technology is evolving rapidly, so while something might have been too complicated yesterday, it may as well become the new normal tomorrow.

Just remember — as long as you can get on without Machine Learning, there’s really no pressing need to spend resources on it.

About The Author

This article was written by Anna Kvasnevska, and has been posted on her behalf, with her permission. Anna is a Millennial, an ex-pat, a writer, a marketer — in that particular order. On a path to understanding the world through personal & spiritual growth.