Approximately 54% of logistics and supply chain companies have implemented AI into their operating processes, and around 37% of companies are using machine learning and AI for predictive analytics. Many more are relying on this tool for everything from chatbots to social media. There is no doubt that AI is a disruptive force in logistics and beyond, as it is already performing a host of important functions, ranging from automation of repetitive tasks to trend forecasting. One of the most effective ways to make the most of AI in business is prompt engineering—the process of designing specific prompts (or asking the “right questions”) to enable AI to generate the responses you need. Ask a question that is too wide and you will receive a response that is too generic; a narrow prompt, meanwhile, may leave out information you need to receive an answer that makes a difference. Below are a few tips on how to make the most of this handy tool.

Read also: Transforming Supply Chains: The Rise of Artificial Intelligence

If you are already using Perplexity, ChatGPT, or Bard, you can improve the outcome of your searches through curated prompts for more reliable outputs. Prompt engineering works by developing rules for your chosen AI systems. In effect, it involves “training” AI by providing it with strict output format rules and teaching it how to respond correctly through pre-prepared examples in the prompt. Through practice, extensive testing, and constant updating, you can refine your process until you have a curated list of effective prompts.

Training AI to give you precise, bespoke answers can be very helpful in areas like supply chain optimization, predictive analytics, and automation. In supply chain optimization, for instance, prompt engineering can be used to identify inefficiencies and recommend improvements in tasks such as inventory management and transportation. In predictive analytics, prompts can be used to ask AI for predictions for aspects such as supply chain disruptions and necessary operational adjustments. The right prompts can also enable AI to automate repetitive tasks like writing product descriptions or responding to technical support questions.

Many companies are investing in the services of prompt engineers. These professionals can work alongside you to understand your needs, select the best platform, and advise you on whether to use a major provider like OpenAI or host your deployment of an open-source model. For lower-volume tasks, you can often get the results you seek by improving your prompts and strategies for given data. At other times, you may receive the best results from existing models like Bard and ChatGPT by blending chain-of-thought prompts with other tools (such as the Internet, databases, and other software containing pertinent data).

Try a few prompt engineering practices yourself to see whether doing so will satisfy your needs, or whether you need a little help. To do so, first understand the respective strengths and limitations of the major models (including GPT-3 and -4, Bard, and Perplexity). For instance, ChatGPT is known for its ability to produce descriptive language, while Perplexity is fast becoming an indispensable research tool. You will also need to hone your writing skills, trying out several contextual prompts until you create a precise list of prompts that leads to your chosen output. If you go beyond the most popular AI tools, you may need to expand the number of skills you currently possess. For instance, you may need to hone skills like programming, software development, and more.

We mentioned supply chain optimization, predictive analytics, and automation, above. Useful prompts for supply chain optimization include, “Optimize the logistics route from Miami to Las Vegas, considering traffic and weather patterns.” Another prompt might be, “What is the best way to distribute products from our Oregon warehouse to multiple stores in San Diego?” For predictive analytics, a useful prompt may be, “Predict the likelihood of a shipment being stopped owing to snowy weather over the next 24 hours” or “Analyze data from the past three years to identify supply chain disruptions and propose successful mitigation strategies.” For automation purposes, try the following prompts: “Utilize the following shipment details (place of origin, destination, size, weight, and delivery date) to identify budget freight carriers and book the cheapest, most effective option,” or “Analyze the current warehouse stock, orders, and available staff, then create an optimized schedule for staff, allocating them specific packing, loading, and transportation duties.”

Prompt engineering is a magnificent way to get the most out of AI for your business. For the ultimate bespoke prompts, prompt engineers can help you work either with existing frameworks or a dedicated one created for your needs. Of course, you can also experiment by formulating prompts and refining those that produce the closest results to those you are seeking.