Machine Learning for Small to Medium Enterprises

Introduction

Ok you’ve got it.

By the time you’ve read your 10th LinkedIn article, received your 5th forwarded email and seen your 3rd Slack message all proclaiming that Machine Learning (ML) and Generative AI are a MUST for modern businesses, you’ve got it: something is happening. Yet you can’t help but wonder: What does this mean for my Small or Medium Enterprise (SME)?

The fear of falling behind is real and is amplified by a relentless barrage of success stories -companies transforming their entire operations through the power of ML. Or at least they say so. This creates an undeniable sense of urgency and fear of missing out that’s impossible to shake off, particularly when the stakes are high for profitability and productivity in these highly competitive times.

So, how can SMEs, often constrained by limited resources and lacking specialized tech talent, break into this brave new world of machine learning?

ML + SMEs = ❤️ ?

The bright side

There is a reason for this renewed hype of ML and AI since the release of the infamous ChatGPT. Recent technology breakthroughs in these fields (from foundational models to more mature platforms) have made it easier than ever to use ML models without a dedicated team of data scientists. The development cycle is accelerated, and the entry barriers are significantly reduced, allowing SMEs to jump on the ML bandwagon without breaking the bank.

And indeed ML can be quite a versatile tool, offering a range of applications that can drive unique business value.

• Optimization/Productivity: Automate routine tasks, optimize supply chains, and improve operational efficiencies.
• Knowledge Organization/Summarization: Use natural language processing (NLP) to summarize large volumes of text or categorize customer feedback. For example, ML can help digitize low-quality handwritten printed documents.
• Customer Satisfaction: Analyze customer behaviour and preferences to personalize product and marketing strategies. For example, a call centre company can benefit from automated analysis of calls, leveraging speech-to-text and text processing technologies such as sentiment analysis or conversation segmentation.
• Demand Forecasting: Predict sales and usage trends based on historical data and market conditions. For instance, ML can help a retail business to predict inventory needs based on historical sales data.
• Content Generation: Use generative algorithms to create marketing content and product descriptions.

Given this wide range of applications, it looks like everyone should be doing ML, right?

ML for everything?

If all you have is machine learning, everything looks like a dataset.

ML is not a panacea. Here is when you should NOT use ML:

• Simple decisions. A simple heuristic works fine? Keep it!
• No data or data that is too hard to acquire. Data can be hard to collect because of regulation or it can be hard to use because of its heterogeneity (e.g. coming from source with incompatible standards, for instance different hospitals).
• Traditional software development is enough. Keep it simple and deterministic!
• High-stakes decisions. If errors are not allowed or too costly, then ML may not be worth it.

You can see challenges of making ML real with what we could call the beta bait. It is one thing to share a shiny 10s video showcasing a smooth end-to-end process with amazing ML-powered features. It is another to make a business out of it. This is why most companies releasing features based on the latest OpenAI models are still in beta.

Even with more traditional ML methods, a lot of things can go wrong.

Are you ready for ML?

Before you jump into the world of ML, it is crucial to assess whether your organization is ready.

Here’s a simple framework to evaluate your ML-readiness:

Process. Functionalities. Data.

1. Process
   • Do you have current processes that could benefit from ML? Ex: If your customer service team spends hours sorting through customer queries, an ML algorithm could automatically categorize these queries, freeing up valuable time for your team. The added value of ML is to automate parts of the process which are repetitive, time-intensive and somewhat boring so that you can focus on more interesting and complex tasks.
2. Functionalities
   • What new functionalities could ML bring that would be useful for your internal stakeholders or customers? Ex: An ML-powered recommendation engine could personalize the user experience on your e-commerce site, increasing sales and customer satisfaction.
   • Are they worth exploring from an ROI estimate? As fancy as they may look, ML-based projects also need to be analyzed to make sure they will bring value to your business.
     • Cost: What functionalities will you get and at what price? The cost of an ML project is mainly influenced by 3 factors:
       • The difficulty of the task, whether existing solutions can be reused/customized and the required computation power to train/use the model
       • The cost of data collection and annotation
       • The need for accuracy
     • Implementation time: How long before you can use these functionalities?
     • Expected benefits: With these new functionalities, how much <time will be saved> / <revenue will be generated> / <cost will be reduced> / <customer satisfaction be improved> ?
3. Data
   • Do you have data that can be used, or the potential to collect it? Data is the lifeblood of any ML project. It can range from emails and Excel sheets to databases, Google Drive files, customer ratings, and even code repositories. The scale and quality of your data will significantly impact the effectiveness of your ML initiatives.
   • Is it structured or can it be structured? Structured data is easily searchable and can be readily analyzed. Think Excel sheets, JSON files, or Notion tables. Unstructured data, like written documents or audio files, will require additional processing to extract useful information or metadata. Generative AI is so popular because it provides a new powerful way to interact with unstructured data.
   • Are there legal constraints or privacy considerations? Ex: If you’re considering using customer emails for sentiment analysis, you’ll need to be aware of data protection laws like CCPA or HIPAA that could impact your ability to use this data.

Build or Buy?

So you’ve passed the ML-readiness test and are eager to start incorporating machine learning to empower your business? That’s fantastic! Your next big question: how do you go about it?

A major decision you need to make is whether to “Build” or “Buy.”

• Build: in-house development. Choosing to build means leveraging or hiring internal resources to develop and maintain ML-based functionalities. This can be a good option if:
  • Long-Term Strategic Investment: You see machine learning as a long-term strategic asset and are willing to invest in its ongoing development and maintenance.
  • Expertise: You already have team members with machine learning expertise or are willing to invest in hiring and training personnel.
  • Customization: Your business needs are unique, and off-the-shelf solutions won’t meet your specific requirements.
  • Control: You want complete control over your data and algorithms, perhaps due to sensitive information or unique business logic.
• Buy: external solutions. On the other hand, buying involves leveraging existing software or APIs to implement ML features. If you’re not sure what an API is, check this out. This can be a good option if:
  • Strategic Focus: Your core business isn’t technology, and you’d rather focus on your main operations than divert resources to develop ML capabilities.
  • Speed: You want to implement ML features quickly without the time-consuming process of building from scratch.
  • Cost: Your budget doesn’t enable hiring of specialized staff or long development cycles.
  • Scalability: You’re looking for solutions that can easily scale with your business without requiring constant tweaking, maintaining and updating.

Leveraging external consulting expertise can be useful to know whether “building” or “buying” is the right decision. Technical consultants are aware of the effective capabilities of current technologies and can help you identify and implement the most promising solutions.

Your path to ML success

Navigating the world of machine learning can be a daunting task, especially for SMEs with limited resources and tech expertise.

However, the barriers to entry are lower than ever, and the potential benefits are significant. From automating mundane tasks to gaining deep insights into your customer behaviour, the applications are vast and varied.

But ML is not a one-size-fits-all solution. It’s crucial to assess your organization’s specific needs, capabilities, and constraints before diving in.

Still not sure if ML is the right solution for you nor how to implement it?

Feel free to reach out to us to schedule a free consultation with our experts to help you make the right decision, with or without ML!