#BizTrends2019: 5 Tips to get your business ready for AI in 2019

Brett St Clair, CEO at Siatik

Many businesses spent the year taking sample data from marketing campaigns, customer insights and production lines, assigning a team of very clever data scientists to explore how things could be done better. We had varying degrees of success, but one thing is for sure: South African business is ready to take the next step towards AI and Machine Learning (ML) in 2019.

But is AI suitable for every business?

Yes, it is. Examples are proliferating around the world. We can use tools like ML to do propensity models for more accurate predictions; use Vision AI on production lines to faster classify fruit; diagnose brain scans for medical abnormalities with a far higher degree of accuracy; and use ‘bots’ to deal with FAQs in call centres before routing to a human agent for more complex scenarios. Every single business can offer a better service using AI - so why aren’t you?

First up, you need to be clear on the business problem you want to tackle. This is not a traditional IT project where you hand over a poorly-detailed spec to the “techies” and hope they do it quickly and cheaply. To make proper use of ML algorithms, it’s essential for business and IT functions to work very closely together. Understanding the constraints of your legacy systems and related business challenges is the secret here.

Face the challenges

Next, the biggest hurdle is data. We all have databases holding significant information about our accounting and internal systems. The problem is that this represents only around 10% of existing data: the rest is siloed data in older warehouses and littering our hard drives in unstructured formats like Excel and Word documents.

2019 is the year to make use of modern toolsets and services to revamp our data and fully incorporate AI into our day-to-day business operations. As you get your data ready for AI, beware not to put the cart before the horse. We can easily start training machines on inadequately prepared data sets but, to politely rephrase a much-loved saying about horse excrement, what you put in is what you get out. If there’s just one thing to aim for, know your data before taking the next step.

If you’re ready to ramp up for AI, here are some simple tips to boost your chance of solving business problems quicker than ever before:

1. Leave your onsite servers and traditional data centres behind

   Instead, think along the lines of hyperscale cloud, and do it like a Google. Traditionally, businesses had to run batch jobs and reports in quiet periods, because servers did not have enough power or storage space to run during peak times as well as do reporting. With hyperscale, you have unlimited access to storage and compute at any time. Plus, you only pay for what you use, which reduces your costs.

2. Ever heard of Data Pipelines?

   These are important because - like oil pipelines - you need to link and transport from your existing systems and data sources to the cloud database. Cloud streams data as and when it is ready, which creates real-time data known as Reactive Data Streams. It’s essential to pipeline your data and keep it in transit.

3. Next, get familiar with ETL: Extract Transform and Load

   This is part of the Data Pipeline process and helps translate multiple datasets into one central source in a common data model or structure. Imagine how hard it would be to work with a dozen people all of whom spoke different languages: it’s the same with data, so it’s critical to make sure everything can understand your data.

4. Think about your database as a service

   This what cloud computing is all about: you don't need to install the database on a server because hyperscale service providers like Google manage the databases. All you need to do is insert data using your Data Pipelines and query it. These modern databases are called Data Lakes because they store data in an unstructured way, allowing you to store vast quantities of data and query it quickly. SQL is great, but when it comes to your Big Data it just won’t cut it.

5. Finally, visualise your data

   There are multiple tools that can link to your database, such as Tableau, QlikView, Google’s Data Studio, to name just a few. Businesses can learn so much by just understanding their data. Only then will you be ready to start figuring out what problems you can solve with AI.

Once again, developing AI and ML projects is not like a traditional IT project. You cannot just plan a project and deliver it 6-18 months later. The chances of getting it right the first time are slim. Working with cloud tech is about constant iteration - making mistakes and trying again. But if you go cloud, you can rerun your models again and again, at high speed - so you’re not waiting weeks between each adjustment. Plus, you make a great saving on capital outlay.

You need to get going now because the world is moving fast - but starting with the recipe above means you will be well ahead of the pack in terms of taking your business to the next level