AI For Business
My team wants to use AI. But where do I start?
Most organizations do not know which workflows are ready, whether their data can support AI, or which tools fit the work. I help teams figure that out before they invest time and money in the wrong solution.
Scenario 1: Is my data ready for AI?
AI cannot reliably use data that is duplicated, inconsistent, outdated, or stored in different formats. I help teams identify where data needs to be cleaned, matched, normalized, and standardized before it is used in an AI workflow.
Scenario 2: Which team should start using AI first? Are my employees ready?
Not every task needs AI. I help teams find the workflows where AI could provide real value, identify the risks and dependencies, and choose a practical place to start. I also consider the very real human impact of introducing this ‘new employee’ and the dangers of AI slop on a group dynamic.
Scenario 3: How do I choose what AI model or agent to build?
A team may know it wants to use AI, but not whether Copilot, Claude, or another tool is the best fit. I help teams compare options, design a small test, and follow practical best practices when building Copilot agents and multi-agent workflows.
About My Work
I help nontechnical teams translate messy business processes into AI workflows that can actually be tested and used. My focus is not just the prompt. I look at the data, the people, the workflow, the tool, and the decisions that still require human judgment.
Choosing the AI model is the fun part.
The real work is preparing your data, workflows, and people to use it well.
Case Study: Multi-source extraction
Challenge: A company wanted several AI agents to gather information from different sources and create an executive brief.
Tool: Copilot Enterprise, Studio, Powerautomate
What I did: Tested source permissions, identified field matching, cleaned and normalized data, generated human-review verification protocols, and added synthesis guardrails.
What I found: More instructions did not always improve performance; data-source problems could look like AI failures; source-specific agents needed different guardrails; human verification was still necessary; humans wanted visibility at each touchpoint to retain trust.
Result: Multi-agent, transparent workflow, reducing manual compilation, allowing team to curate the final briefs.
Methodology:
I broke the existing briefing process into three specialized AI tasks covering internal data, official company information, and external press research, then built and iteratively tested separate Copilot agents for each source. I worked directly with users to compare agent results against manually prepared briefs, identify retrieval and accuracy failures, refine search and verification rules, and add meeting context so research is relevant to the specific audience and purpose.
The resulting design uses a primary orchestrator to call the specialized agents and combine their outputs into one briefing workflow, while keeping structured data retrieval, calculations, and document generation deterministic where appropriate. I also identified human verification points, source limitations, authentication concerns, and opportunities to reuse the underlying architecture for additional briefing workflows.
My role covered workflow decomposition, agent design, prompt and instruction development, testing, user feedback, guardrails, and translation of the business process into requirements for the orchestration build.
Update August 2026: Microsoft now lets you add an MCP server straight into Copilot Studio through Tools → Add a tool → New tool → Model Context Protocol. You provide the server URL and authentication, and Copilot can discover the tools that server exposes. This is a possible optimization to this workflow.
Case Study: Preparing Human Data for AI
Challenge: A team wanted to automate a daily briefing using information from an Excel schedule they had relied on for years. The file worked well for people, but its visual structure was difficult for AI and automation tools to interpret reliably.
What I found: The workflow needed standardized, machine-readable data. But simply rebuilding the spreadsheet would create a different problem: employees would have to learn a new system, and the existing file was a historical record and source for other business processes.
What I did: Kept the familiar, human-facing spreadsheet unchanged and created a separate AI-ready data layer behind it. Formulas automatically translated the existing schedule into standardized tables the automation could reliably read.
Result: The team kept the workflow and source document they already knew, while the AI received the structured data it needed. No duplicate data entry, no new maintenance process, and less disruption for the people expected to use the new system.
Services
AI use-case discovery works better when you start with the work people are already struggling to do, rather than asking them to invent an AI idea from scratch.
Book an Appointment
I’m opening a small amount of independent capacity for AI workflow testing. I help teams test whether a proposed Copilot or AI workflow is using the right data, following its instructions and producing results people can verify. I’m looking for one or two pilot projects starting Aug 31, 2026.
Example Starter Package
AI Workflow Test Sprint
I test one proposed AI workflow, document what works and fails, and give the team a practical improvement plan.
Possible scope:
one intake call
review of one workflow
5–10 structured tests
findings and recommendations
one results call