*In Operational work, writing is rarely just about writing.
*
We process thousands of accounts daily, with multiple agents utilizing large sets of tools. A single account can require an agent to locate approved pre-defined content, adding or subtracting the content to align with the consumer, correcting the grammar, adding empathy, and many other items.
Apart, these single steps are simple. Together across multiple accounts in one day, it becomes a digital scavenger hunt, with each iteration causing friction and frustration.
With the advent of Microsoft CoPilot access being given to our department, I began experimenting with it as a way to reduce that frustration; not by asking it to make the decisions for these scenarios, but as a controlled-text transformation tool.
The Problem: Great Content, Too Many Steps
My workplace already has content for a variety of situations. However, something you learn in healthcare is that you never stop learning, and our predefined content lends credibility to that by never having all the information for an issue in one place. Yet, the goal of this type of content is twofold; it creates a consistent language for the company and helps employees communicate in clear ways amongst each other.
The traditional process can look something like this:
- Locate the first piece of predefined content.
- Copy or insert it.
- Locate a second piece.
- Add it to the first.
- Remove redundant language.
- Rearrange the information.
- Correct grammar and formatting.
- Rewrite portions so the final response reads naturally.
- Review the finished text for unnecessary personal information.
You can see that just for one response, you may have to click and manually edit it multiple ways to form it into a cohesive message.
In this scenario, the content is not the bottle neck. It is the assembly wherein the issue lies.
My Solution: A Master Prompt as a Safety Layer
It would be just as unfeasible to simply use Copilot as a middleman; copy and pasting all of this information to Copilot to assemble it into what you need just adds extra steps to the process.
To combat this, I wrote what I called a "MasterFile" as a DocX; essentially it is a large scale prompt that Copilot can parse.
In conjunction with this Masterfile is the custom agent which Copilot allows you to build. Using these two tools, I am able to create a single chat agent that will not work unless slotted with the most up to date Masterfile, and establishes protocols and boundaries with how to handle ouput.
For example; the custom agent is asked to ensure that if any kind of information such as a name, location, or phone number is entered into the chat, it instead addresses that and does not create a response. Another example is that I store the Masterfile Version in the Agent so that when you upload the Masterfile, it can advise that it is not the most recent file or it is too far out of date to use.
So, What Does It Actually Do?
At its core, the agent makes approved content faster and easier to use.
Instead of opening several menus, locating multiple pieces of predefined content, and copying and pasting them together, the employee can simply type something like:
"provide ______ pdc."
The agent then pulls together the relevant predefined content and presents it as one response.
In some cases, I have been able to take a process that may have required four clicks and three separate copy-and-paste actions and reduce it to one typed request. That may not sound like a huge difference for a single account, but when the same process is repeated throughout the day, those saved steps begin to add up quickly.
Another issue the agent helps with is capitalization and grammar.
In chat environments, agents often work from internal information that is written entirely in capital letters. The content may be accurate, but it is not always ready to be sent directly to a consumer. Manually rewriting it takes extra time and can also lead to inconsistent wording between employees.
Using the agent, an employee can simply type:
Rewrite:
They can then paste the sentence or paragraph underneath it. The agent converts the all-caps internal language into something that is grammatically correct, easier to read, and more appropriate for a consumer-facing conversation.
The employee still reviews the response before using it. The agent is not deciding what information belongs in the conversation, and it is not making decisions about the account. It is taking content the employee has already selected and helping turn it into a cleaner response.
I was not trying to build something that would replace the employee or automate the entire interaction. I wanted to remove some of the repetitive clicking, copying, pasting, rearranging, and rewriting that slows the employee down between understanding the issue and helping the consumer.
The predefined content still provides the approved information. The employee still provides the judgment. The agent simply helps close the gap between the two.
Conclusion: A Few Clicks Matter More Than You Think
This project started with a simple question: if employees are repeating the same clicks and copy-and-pastes throughout the day, is there a better way to reach the same result?
The goal was never to have Copilot do the employee’s job. The employee still chooses the correct information, keeps personal details out of the prompt, and reviews the final response. The agent handles the repetitive work between those steps.
One saved click does not seem like much. Neither does one saved copy-and-paste. Repeated across multiple accounts and multiple employees, however, those small improvements begin to matter.
The MasterFile and agent are still developing. This was not intended to be a massive AI transformation project. I noticed a frustrating process, started experimenting, and found a way to make it easier.
That makes me curious: what repetitive process at your job seems small on its own but becomes a major source of friction throughout the day? Have you found a way to improve it with Copilot, another AI tool, or even a simple automation?
Top comments (1)
This is a good use case for Copilot-style automation because the value is not “AI writes for me.” It is turning a repeated multi-click ritual into a repeatable workflow where the human still owns judgment and final copy.