Agentic Workflows for Operations

Context

I reviewed how the Operations team worked to find places where AI could remove manual steps. Two useful opportunities emerged: resource allocation and RFP management.

Challenge

A key challenge for Operations was matching people to projects while considering both their experience and the work they wanted to do.

The process relied heavily on one-to-one conversations and the Operations team’s knowledge of people’s preferences and experience. That became harder to maintain as the organisation grew and teams changed.

The RFP process had a similar problem. Teams spent time finding old proposals, choosing useful examples and copying material into a first draft. Generic templates did not remove the need to search through past submissions.

Both problems depended on making the underlying information structured enough to search and use in automated workflows.

Approach

Resource Optimisation

We used Databricks for the resource allocation work because AI_QUERY could structure free text and Genie could provide a natural-language interface to the results.

As part of the quarterly review process, employees described their interests, career goals and preferred project types in free text. We loaded those responses into Databricks and used AI_QUERY to extract consistent attributes.

This gave Operations a record of employee preferences that could be updated and used alongside its existing knowledge when allocating projects.

We exposed the data through a Genie space so the team could query it in natural language. Custom tooling handled the more specific operational queries.

We then added timesheet data and RFP descriptions to show both past experience and the upcoming work pipeline. Operations could use that view when matching people to opportunities.

RFP Management

For RFP management, we first brought historic proposals, approved templates, reusable content and examples of 4most’s writing style into one knowledge base.

The RAG workflow could retrieve relevant material and combine it with the right template and writing guidance. Users could create a starting point based on the client, sector, service line and proposal context without searching through folders themselves.

Databricks could have supported the workflow, but Microsoft Copilot was a better fit for this use case. It was a straightforward RAG implementation for a broad group of users, and Copilot made it available through Microsoft Teams, where they already worked.

The system made approved RFP material searchable through Copilot and provided reusable starting points for new proposals.

Results

The work gave Operations two practical capabilities: a searchable view of employee experience and preferences for resource allocation, and a central source of approved material for RFP drafting.

Resource allocation

• Captured employee interests and project preferences alongside experience data
• Combined timesheet history with the upcoming project pipeline
• Let Operations query the combined information through Genie

RFP workflow

• Centralised previous RFPs, approved templates and reusable content
• Retrieved relevant material using RAG rather than manual folder searches
• Made the workflow available to users through Microsoft Teams and Copilot