Field story 4 min

# AI Made the Request Better, Not the System

A healthcare technology company built a simple chat assistant that asked requesters the questions intake reviewers used to ask days later. Request quality improved immediately. The intake system itself did not change, and being precise about that difference is what made the result useful.

The condition

Every incoming request has to be translated by someone downstream before it can be evaluated, because each one arrives in a different shape.

## A faster person in an unchanged process

Most organizations approach AI the same way. They buy licenses, train people on the tool, and ask everyone to be more efficient. Individual productivity does improve. A person drafts, summarizes, and analyzes faster than before. But the gain stays local. The faster person hands the work to the next person, who is working inside the same unchanged process as always. A faster person inside an ineffective process produces a faster ineffective process. Mandates make it worse: people use the tool for everything, including activity whose main purpose is to show they are using it.

The more useful question is where in the flow of work a deliberately designed assistant would make the whole system better. At one healthcare technology company, the answer started at the front door.

Requests there arrived from every direction and in every shape. Some were detailed and some were vague. Some were complete thoughts and some were fragments. Every one of them needed someone downstream to translate it into the format required for evaluation. One of the company's leaders decided to build an assistant to take that translation work off his colleagues. He gave it the intake template, a set of well-formed example requests, and clear instructions on writing goals, which we helped him draft.

## Say it the way you normally would

He demonstrated it to a company-wide audience. On one side of his shared screen sat a blank chat window. On the other sat the company's intake template. Most people in the room, he pointed out, had submitted a request and had it bounced back for being in the wrong format. So he asked for a request exactly the way someone would normally say it, not the way they thought the form wanted it.

Someone called one out: the analytics team needed to add a column to the weekly utilization export so contractor hours could be seen separately. He typed it in word for word.

The response surprised people. The assistant did not reformat the request. It asked two clarifying questions: who was requesting the change, and what business problem was driving it. Those were the same questions the intake reviewers asked. The difference was timing. The assistant asked them while the request was fresh, instead of days later in an email thread. He answered both, and the assistant returned a structured request in the agreed format, with an outcome statement, a requesting owner, a business driver, and an urgency rating.

## It does not submit anything

A hand went up near the front. Did people still have to fill out the form? They did. The assistant did not submit anything and did not touch the intake system at all. It helped people create a better request before they submitted it, removing the part everyone disliked: turning a messy version into an acceptable one.

The next question was what happened if it got something wrong. Then you fix it, he said, the same way you would fix your own draft. It was a thinking partner, not a decision maker. In the two weeks he had been testing it, it had gotten the format right almost every time, because the format is the easy part.

By the end of the session, people were using the assistant to prepare their own requests. Incoming requests began arriving in a consistent format, and much of the translation work reviewers had been doing downstream disappeared. The same approach then found a second use. People could paste rough notes or meeting transcripts into a companion assistant and get back the decisions made, action items with owners, open questions, and next steps, so fewer commitments were lost between the conversation and the backlog.

## Precise about what changed

The improvement was real, and it is worth being exact about it. The assistant made people more effective. It did not change the system.

Someone still had to realize there was a need, open the assistant, provide the context, and submit the finished request. The assistant reduced effort. It did not remove the process, and it did not decide whose request mattered most or create any capacity to do the work. That makes it an excellent first step for an organization that wants value quickly without adding technical complexity. It is not a redesign.

It also worked for a specific reason. This organization already knew what a good request looked like. The template, the examples, and the expectations existed before the assistant did. Put the same assistant on top of a broken intake process and it simply produces a faster version of the dysfunction.

## The next rung, and where to stop

We had the same intake problem ourselves. Requests reached us through email, chat, referral calls, and the occasional hallway conversation, and some of it never entered our intake process until someone remembered it had been promised. So we went a step further than a chat assistant. First we connected an internal application, where a whole category of our own work lived out of sight, to the backlog where everything else was tracked. Then we connected an agent to the inboxes, shared drives, and chat channels where most new work arrives. It notices when new work shows up, turns each request into a record with the fields we require, and routes it into the backlog so all of it is visible in one place.

There is a step beyond that, where an agent acts on its own across systems on a schedule or a trigger with little human oversight. It is real and it works, but it is not where most organizations should begin. Autonomous action compounds error. An agent taking a hundred actions a day will eventually make a mistake that spreads before anyone notices. That level of autonomy belongs to a process that is well understood, stable, and cheap to get wrong, not one that is still being redesigned.

For most organizations, the right starting point is an assistant that makes hidden work visible and organizes it for a person to confirm. In every version, the structure came first and the efficiency came second.

AI can make a sound work system faster. It cannot supply the structure the system is missing.

Know someone running into this? Send it to them.

[Share on LinkedIn](https://www.linkedin.com/sharing/share-offsite/?url=http%3A%2F%2Fagileon.test%3A64380%2Finsights%2Fai-made-the-request-better-not-the-system%3Futm_source%3Dlinkedin%26utm_medium%3Dshare%26utm_campaign%3Dinsight-share%26utm_content%3Dai-made-the-request-better-not-the-system)

Email

### Share With Your Team

Choose who it is for. You send it yourself from your own email; Agileon never writes to your colleagues.

Your Message

Open in Email

Share

[Share on LinkedIn](https://www.linkedin.com/sharing/share-offsite/?url=http%3A%2F%2Fagileon.test%3A64380%2Finsights%2Fai-made-the-request-better-not-the-system%3Futm_source%3Dlinkedin%26utm_medium%3Dshare%26utm_campaign%3Dinsight-share%26utm_content%3Dai-made-the-request-better-not-the-system)

Email

Components

[Surface](https://www.agileon.com/insights?component=Surface)[Scope](https://www.agileon.com/insights?component=Scope)

Industry
Health Tech, Professional Services

Seen in practice
[The work that fixes this Work Management Systems](https://www.agileon.com/professional-services/work-management-systems)

Keep reading
[Surface The Seven Components of Work](https://www.agileon.com/insights/the-seven-components-of-work)[Surface Professional Services The Next Reorg Will Be AI-Driven. Most Will Do It Wrong.](https://www.agileon.com/insights/the-next-reorg-will-be-ai-driven)[Surface Health Tech Half the Portfolio Was Keeping the Lights On](https://www.agileon.com/insights/half-the-portfolio-was-keeping-the-lights-on)[Scope Professional Services Scope Is Where Hidden Rework Begins](https://www.agileon.com/insights/scope-is-where-hidden-rework-begins)
