Prompt Optimization: Better Results, Lower AI Costs

We've all worked with someone who can't answer a simple question without explaining everything that happened leading up to it. Ask why an order is late, and 15 minutes later you're hearing about a supplier change in 2019, a warehouse reorganization, and someone named Bob who retired three years ago.
Sometimes we treat AI the same way.
We give it lengthy instructions, attach every document we can find, and hope that somewhere in all that information is what it needs to produce a useful answer.
But more information doesn't automatically mean better results. And when you're using AI across hundreds or thousands of business transactions in systems like Microsoft Dynamics 365 Business Central, all that extra processing can add up.
If you read my previous article in this series, you may remember Laura. Her AI initiative was a success, but managing the growing costs became another challenge. And one place to look for savings was how everyone was actually using AI.
That's where prompt optimization comes in. It's not about writing the shortest possible prompt. It's about giving AI enough information to do the job well, without making it work through everything Bob left behind.
I. How can better prompts reduce AI costs?
Better prompts can reduce AI costs by helping AI complete a task with fewer unnecessary instructions, responses, and follow-up requests. The objective isn't simply to use fewer words. It's to get the right result without needing several attempts to get there.
Think about a manufacturing manager who asks AI to review a delayed customer order.
A vague prompt might be: "Tell me what's going on with this order."
AI might produce a lengthy summary of the order history, customer information, and related transactions. Useful? Maybe. But the manager really just wanted to know three things: Why is the order delayed? What needs attention? And who should handle it?
A more effective AI prompt would be:
"Review this order and identify the reason for the delay, any action required, and the person or department responsible. Summarize the findings in three bullet points."
Now AI has a specific task, a defined scope, and a clear output format. There's less room for irrelevant information and fewer reasons to ask follow-up questions.
This is where prompt efficiency becomes a business consideration rather than just a writing skill.
One poorly designed prompt probably won't make much difference to your AI budget. But when the same request runs hundreds of times across sales, purchasing, production, and customer service, those inefficiencies can multiply.
That's why I encourage businesses to evaluate prompts based on more than whether they produce a good answer. The better question is whether they produce the right answer consistently, without unnecessary processing or repeated attempts.
II. Do longer AI prompts use more tokens and cost more?
Generally, yes. Longer prompts consume more input tokens, which can increase costs in AI systems that charge based on token usage. But a shorter prompt isn't automatically more efficient, especially if it leaves out information AI needs to complete the task.
That's where AI prompt optimization requires a little judgment.
Let's say you're asking AI to explain why production costs exceeded the budget. You could give it six months of financial reports, production records, purchasing data, and every spreadsheet someone thought might be useful.
Or you could provide the relevant budget figures, actual costs, and a few key production details.
The second approach gives AI a much better chance of finding the answer without sorting through the digital equivalent of someone's overstuffed filing cabinet.
But there's a catch. If you omit important information, AI may produce an incomplete explanation, requiring additional prompts or corrections. All of a sudden, your attempt to reduce AI token usage has created more work.
That's why AI context optimization isn't about stripping away information indiscriminately. It's about deciding what AI genuinely needs to complete a particular task.
I encourage businesses to review frequently used prompts and ask three questions: What information is essential? What's helpful but optional? And what's included simply because it's available?
The same thinking applies to the response. If you need a three-sentence explanation of a variance, ask for one. There's little value in generating a three-page report that nobody has time to read.
The objective is simple: provide enough context for an accurate, useful answer, and request only the level of detail the business truly needs.
III. Should businesses standardize AI prompts for repeatable workflows?
Absolutely, especially when the same AI task is performed repeatedly. Standardized prompts can improve consistency, reduce unnecessary AI consumption, and make business workflows easier to manage.
Prompt optimization becomes part of the process
Think about a customer service team using AI to draft responses to order inquiries. If every employee creates their own prompt, the results may vary considerably. One response might be three sentences. Another might read like a formal apology from the company's legal department.
Instead, businesses can develop reusable prompts with defined instructions, inputs, and output formats. Customer and order details can change without requiring everyone to reinvent the process.
Microsoft's Copilot Studio supports this approach by allowing organizations to build, test, and reuse prompts across agents and workflows.
This is where AI prompt standardization connects to broader AI workflow optimization. As I detailed in my earlier article on AI automation and Business Central, the goal isn't simply to automate individual tasks. It's to make business processes work better.
Once AI becomes part of a repeatable workflow, the prompt is no longer just something an employee types into a chat box. It's part of the solution's design.
Better prompts are just the beginning
Prompt optimization isn't about making every prompt as short as possible. It's about giving AI enough information to do the job well, without wasting resources on instructions, context, or responses that don't add value.
For businesses using AI across everyday operations, those small improvements can make a meaningful difference. Better prompts can lead to more consistent results, fewer retries, and better AI cost efficiency as usage grows.
But what happens when AI has to process the same information over and over again?
That's another opportunity to improve efficiency, and one I'll explore in the next article in this series, when we look at tokens and prompt caching.
In the meantime, if you're wondering how to make AI work more efficiently within your business processes, let's talk. Sometimes the best place to start is simply asking better questions.
About Matt Keyes

Matt Keyes is a visionary leader, founder, and CTO of Key Partner Solutions. With over two decades of experience in Microsoft Dynamics, he is passionate about driving digital transformation for businesses through innovative technology solutions.
His deep technical expertise, combined with a strategic approach to solving business challenges, makes him a sought-after thought leader in the industry.
Today, Matt is focused on empowering companies to unlock new levels of growth and efficiency through cutting-edge software development and consulting.
Connect with Matt on LinkedIn.




