Introduction

Artificial Intelligence is yet another tool that is an overnight success, decades in the making. The Fear Of Missing Out (FOMO) seems huge since ChatGPT brought AI from what Gartner’s Hype Cycle calls the Trough of Disillusionment to what might be the beginnings of the Slope of Enlightenment. But what do you really need to do with it, if anything? Do you need to build anything strategically core to your customer offerings? Use it as a tool to enhance existing offerings? Or use something more operational; from process management to marketing. One challenge right now is people seem to want to just sprinkle some AI on things to make sure the buzzword is in a product checklist or annual report.

According to Statista’s “AI Trends & Predictions Roadmap to 2025,” AI market value will increase from US$244 billion in 2025 to US$827 billion by 2030. You can find predications of various magnitudes from other sources. But what does it mean? Chances are it means your competitors are deploying AI. Whether it’s for something useful with solid ROI or not is the question. If your competition is deploying things, you’d better at least be doing research.

So where do you start? Too many posts I’m seeing talk about AI Strategy go right to implementation. That’s AI at tactical level. There’s value in that. And it’s possible to think of general capabilities as strategic. But being strategic means starting with your business drivers. Your AI strategy, (just like all the others), should be based on the value you’re trying to create. At the very highest level, as always… there’s basically two goals: Increase Revenue. Or Decrease Costs. Unless you’re actually building and selling the AI tools themselves, chances are the value you’re creating comes from elsewhere. In which case AI is “just” a tool; another means to an end. Being skilled at it may be a strategic capability, but it’s not a business strategy in and of itself.

What Are Your Strategic Drivers

There’s an expression in safety critical environments, “It’s often not the emergency that kills you, it’s a panicked response.” For those scrambling to “get some AI in here!” to check some boxes, that might not work out so well. A calmer approach is to look at this as just another tool. Yes, it’s potentially powerful and your competitors may be making moves. However, AI has non-trivial costs and risks as well.

Let’s look at strategic drivers first. Perhaps there are areas that jump out as useful to experiment with a test project. This could help build organizational muscle with the technology before making larger bets.

Strategic drivers for AI can be categorized into operational efficiency, customer experience, innovation, decision-making, risk management and more. Here’a one view with examples:

1. Operational Efficiency

• Routine Task Automation: Automate tasks such as data entry, transaction processing, and handling customer inquiries, freeing up personnel for more complex activities.
• Process Optimization: Algorithms can analyze processes to identify bottlenecks and suggest improvements, leading to more efficient workflows.
• Cost Reduction: Reduce operational costs; labor, materials, and energy consumption.

2. Customer Experience

• Personalization: Analyze customer data to provide personalized recommendations, offers, and communication, enhancing the overall customer experience.
• Support: Chatbots and virtual assistants can provide 24/7 customer support.
• Sentiment Analysis: Analyze feedback and social media to gauge sentiment.

3. Innovation

• Product Development: Assist designing products and features with trends, preferences, and historical data.
• R&D Acceleration: Process vast amounts of data accelerating R&D.
• Creative AI: Contribute to creative such as content, design, and marketing, generating new ideas and approaches.

4. Decision-Making

• Data-Driven Insights: Analyze large datasets to uncover patterns, trends, and correlations, providing valuable insights for strategic decision-making.
• Predictive Analytics: Forecast future trends and outcomes based on historical data, helping businesses make proactive decisions.
• Optimization Algorithms: Solve complex optimization problems, such as supply chain management and resource allocation, to enhance decision-making processes.

5. Risk Management

• Fraud Detection: Detect fraud in real-time by analyzing patterns and anomalies.
• Cybersecurity: Enhance cybersecurity by detecting and responding to threats faster.
• Compliance: Help ensure regulatory compliance by monitoring transactions and processes for adherence.

6. Human Resources

• Talent Acquisition: Streamline the recruitment process by screening resumes, assessing candidate fit, and even conducting initial interviews.
• Employee Engagement: Monitor employee engagement and satisfaction, providing insights to improve workplace culture and reduce turnover.
• Training and Development: Offer personalized training programs and career development plans based on individual employee needs and performance data.

7. Supply Chain and Logistics

• Demand Forecasting: Predict demand patterns, helping businesses manage inventory levels and reduce stockouts or overstock situations.
• Route Optimization: Optimize delivery routes, reducing costs and delivery times.
• Inventory Management: Manage/automate inventory for optimal stock and reduced carrying costs.

8. Marketing and Sales

• Targeted Marketing: Analyze consumer behavior to identify target audiences and tailor marketing campaigns for better ROI.
• Sales Forecasting: Predict trends and buying patterns, helping plan and strategize.
• Customer Segmentation: Segment customers based on various criteria, allowing for more precise and effective marketing efforts.

Each of these strategic drivers highlights AI’s potential to transform business operations and strategy, leading to improved efficiency, innovation, and competitive advantage.

What are your strategic imperatives?

Back to the P&L

For all of our business complexities, gravity from the P&L can pull us back to simplicity… we’re either making money or spending money. Calculating ROI on AI can be challenging. It’s likely easier to do on the cost side. For example, you may have a business driver to implement AI for customer service expecting savings of 20% per year. You could be wrong, but your goal and means to achieve it are rational and there’s models for this business case. On the top line side it might be a harder sell that AI can increase innovation pipeline cadence by segmenting customers better. It might be true, but how you design and build clustering and classification models for segmentation is still art plus science; so there’s plenty of room to get this wrong. You’ll likely need to experiment. Even if you get it right, costs are fuzzier here from data acquisition to clean-up to modeling. Not to mention the gains themselves.

It’s possible you’re in a zero sum game here. If competitors are using AI to good effect, you may have to follow. Keeping up may be a strategic driver, but in what sense? Just “having AI” isn’t likely it. What’s the business case? In the face of competitive pressure, your strategic benefit might just be staying in a game you’d otherwise get knocked out of.

Execution – MLOps

This is a vast topic of it’s own with an increasing body of knowledge. For the purposes of this article, I’m going to simply leave this as saying, “You’ll need extensive consultation with your technology leadership on organizational readiness to do anything in this area.”

Generally speaking, here are main topics of concern.

• Are you going to use a service for a point solution, or do you need to build out a full Machine Learning Operations framework. (Referred to as MLOps.)
• People: Can you find people to implement given this is such a hot area right now?
• Ongoing Maintenance: You cannot typically just launch a project and be done. There will be ongoing costs from basic DevOps upkeep to model adjustment, possibly re-training, fine tuning and so on.

Though again, building AI alone is not a business strategy, MLOps is arguably strategic from a capabilities consideration standpoint.

For a solid and expansive overview of ML Ops, consider seeing this paper published on IEEE, Machine Learning Operations (MLOps): Overview, Definition, and Architecture by Kreuzberger, Kühl, and Hirschl of IBM. See the paper itself for an easier to expand version of the image below. Depending on what your goals may be, you won’t have to necessarily build out every aspect of this level of workflow. And there may be other special components not depicted that your situation dictates. But to get an overview, this popular paper can give you a sense of what may be involved for a full end-to-end effort. It’s probably one of the best general overviews out there.

Evaluation

Depending on who you talk to, AI evaluation is arguably still in early stages. How effectively you can define success criteria will depend on the category of thing you’re trying to do. For example, an implementation aiming to enhance customer service, (to be more effective while reducing costs at the same time), may be relatively easy to measure. Revenue improvements based on segmenting customers and offering more effective messaging may be more challenging. And for some levels of AI performance, there may be no clear success metrics. You may be able to attribute top and bottom line success to project implementations, but at the product usage level, it’s still sometimes hard to evaluate low-level drivers. At a product level, we have issues including: Data Quality and Availability, Model Performance Metrics, and the usual usage characteristics. At a strategic level, we have Return on Investment (ROI), Ethical Considerations, Regulatory Compliance, Security and Privacy, and more. Further looking at ROI; quantifying ROI of AI initiatives can be difficult. It may take time to see results, and the benefits are not always straightforward to measure. As with any effort, there should be some kind of specific and measurable goals, with the understanding that you may need to adjust how you measure experiments as you move through the process.

Conclusion

New AI/ML tools are a step change in what was available as recently as just a few years ago. While a variety of ML tools may have been available for some time, (decades in some cases), the popularity of Large Language Models (LLMs) has brought all of these tools further out of research and into mainstream business consideration. And yet, the same business considerations as for any tool apply. What’s your vision, mission, strategy? Where might these tools further your goals? From a high-level strategic perspective, nothing about these kinds of considerations have changed.