Getting Better Results with AI Chatbots: 5 Best Practices for Purchasing Teams
AI chatbots like ChatGPT and Copilot are helping purchasing teams save hours each week.
A recent report by AI at Wharton showed that 94% of purchasing professionals surveyed reported using AI chatbots at least weekly.
That’s up from just 50% the year before!
We’ve found the most powerful uses of AI chatbots for purchasing fall into 3 areas:
- Sourcing (see example prompts here)
- Negotiation (see example prompts here)
- Spend analysis (see example prompts here)
In our AI Chatbots for Purchasing course, we go through 150 examples and use cases for these areas, as well as other important areas like Communication, Risk management, Supplier management, Strategy development, Contract management, and Operational procurement.
Example prompts are a great start for procurement professionals to get better results with AI Chatbots.
There are also traps to avoid and best practices to follow. Here are five simple ways to get better output from AI tools:
1. Be specific in your prompts and questions
Vague prompts lead to vague answers. The more detail you give, the more useful the response will be. For example:
Instead of: “Write a supplier scorecard”
Try: “Write a supplier scorecard template for indirect MRO suppliers that includes cost, delivery, and quality KPIs and a rating scale from 1 to 5.”
Include examples, context, or formatting preferences when possible. You’ll reduce the number of follow-ups and get closer to what you need the first time.
2. Explore GPTs and use “Keep in sidebar” if helpful
Custom GPTs are free, purpose-built chatbots trained for specific tasks – kind of like apps on your phone – with many available for common purchasing tasks.
Access OpenAI’s GPT store by clicking on GPTs on the left sidebar in ChatGPT. Currently, the most popular GPT for procurement is “AI Procurement Consultant”, with a user rating of 4.5 stars.
If you find a GPT you use often, click “Keep in sidebar” to make it easier to access in future sessions. You can even create your own GPT with your preferred instructions, tone, and datasets—no coding required.
3. Don’t upload sensitive information—opt out of data sharing
Never upload contracts, pricing, supplier data, or any proprietary files unless you’re 100% certain you’re using a secure, enterprise version with no data retention.
For personal accounts, go to Settings > Data Controls > and turn off “Chat history & training” to prevent your chats from being used to improve the model.
Anonymize sensitive information when testing prompts or analyzing files (think Supplier A and Supplier B rather than supplier names).
4. Ask the chatbot to construct a better prompt for you
If you’re not sure how to ask something, just say so.
For example: “Help me write a prompt to analyze supplier quotes in Excel and highlight the best total landed cost.”
The chatbot will ask follow-up questions or build a reusable prompt you can refine. This is especially useful when you’re new to a tool or trying to automate a complex task.
5. Break tasks into steps and use prompt chaining
Chatbots handle multi-step tasks best when you give them instructions one step at a time. This is called “prompt chaining.” For example:
Step 1: Summarize an Excel file
Step 2: Identify top cost drivers by category
Step 3: Suggest supplier negotiation strategies based on those drivers
This helps maintain context and accuracy, especially when working with large or technical content.
Bonus Tip: Always verify outputs
AI tools can produce confident but incorrect answers. The technical term for this is ‘hallucination”, and all the chatbots do it – but they are getting better.
Always spot check the data you get from chatbots, and click into source links to check that reliability of any recommendations. A good rule of thumb is to check anything you are going to present as your own work.
APD offers a wide range of insightful resources, including expert-led training, on-demand webinars, and blogs on leveraging AI for purchasing!