Our First Women in Automation Session: A Live Agent Demo That Answered the Right Question

A recurring pattern kept showing up across networking sessions: people would get into a conversation about what they were working on and want to keep going. The inaugural Women In Automation session gave that energy a stage. This user group meets monthly, and the format is meant to be hands-on rather than lecture-style.

Shreya Kumar, Technical Community Specialist at Automation Anywhere, and Shrigouri Jumnalkar, Team Lead and Solution Architect at Tangentia and our MVP, walked through something they had built together and showed it working live.

Where the Idea Came From
Shrigouri had been reading a Gartner article about women growing into leadership roles in the supply chain. It got her thinking about what people in those roles are actually doing with most of their time, and specifically about the kind of manual, back-and-forth work that tends to fill the day before the higher-level decisions even get made.

Vendor price changes were a strong example. A vendor emails to say a price is going up. Someone on the procurement team then has to find the relevant inventory data, calculate what the increase actually means for the business, figure out whether approving or negotiating makes more sense, and write a response.

That process often takes longer than it should, runs through multiple tools, and depends heavily on whoever is handling it having the right context at the right moment. The use case was real, and solving it well gives someone more room to focus on the work that actually requires their judgment.

What They Built
The workflow starts with an email. A vendor writes in to say a price is changing on a specific material, effective on a specific date. In most procurement teams, that email sits in an inbox until someone has time to deal with it. The agent they built picks it up immediately.

It reads the email and pulls out key details: vendor name, material, current price, proposed price, effective date, and whatever reason the vendor gave. Then it calls the inventory system via an API to retrieve the product data that the email does not include, such as stock levels, annual demand figures, and lead time for that material. With that in hand, it builds an impact analysis covering the percentage change, the estimated monthly and annual cost impact, and a read on whether the inventory position gives the buyer any negotiating room.

The analysis also includes suggested paths, each with trade-offs noted. The agent does not make the call directly rather it brings human in the loop. A form surfaces for the person handling the request, and they choose: approve, negotiate, or hold. After that, the agent acts on the human approver’s decision: if negotiation is chosen, it drafts and sends the response email with analysis numbers; if approved, it sends the acknowledgment and updates the price in the inventory system.

Watching It Run
To show it working, Shreya put it to the test. She sent an email raising the unit price of a material from $13.00 to $14.50, effective March 1, 2026. A few seconds later, the agent had read it and returned the full analysis. The increase was 11.54%. The estimated annual impact was around $180,000. The inventory data showed the company did not have enough stock on hand to absorb a long gap if they walked away from the vendor, so the suggested path was to approve.

Shrigouri looked at the analysis and chose to negotiate instead. The AI agent reffered to the high cost impact numbers and suggested a lower price that could potentially work for both parties. It pulled the relevant figures into a negotiation email and had it ready. The vendor would receive a professional response that referenced the cost impact numbers and made the case for a different price, written in seconds rather than minutes.

For the second scenario, Shreya sent a revised email bringing the price down to $12.00, which Shrigouri approved. The agent updated the price in the inventory database and sent the vendor an acknowledgment. Both the communication and the system record were closed without switching between tools.

What the Room Asked
The Q&A went straight into the mechanics. Attendees wanted to know what actions and tools were used inside the build, how the agent decides which task to run next, rather than following a fixed branching structure.

The agent uses AI to decide which process to call next, based on how the tasks are described and what inputs and outputs they expect. That is what makes it possible to handle variation in vendor emails without writing a separate branch for every scenario. For production use, the approval form can surface through Automation Anywhere Copilot, and inside Microsoft Teams, so the person making the decision does not have to leave the tool they are already in.

Getting something like this into production involves more than what was visible on screen, and the questions from the room reflected that people understood the gap between a demo and a deployed workflow.

What Is Coming Next

The group meets monthly. Session details, demo recaps, and community updates go out through three places:

• Follow the Pathfinder Community on LinkedIn for announcements and post-event content.
• Join the Pathfinder Community forums at community.automationanywhere.com to ask questions, share what you are working on, and stay connected between sessions.
• Keep an eye on our events page for our next user group sessions.