Using AI tools to support project coordination, reporting, & research communication workflows

In my role as Centre Manager at CIDER within the School of Public Health, I oversee a broad range of operational, administrative, and logistical functions that support both the Centre and its research portfolio. My responsibilities include spearheading the planning and management of the Centre’s administrative and financial operations, ensuring effective coordination of research projects in compliance with timelines, funder requirements, and university policies. This includes developing and implementing systems to track CIDER’s various research related activities.

More recently, I have been involved in developing a project tracking application, with the longer-term goal of expanding it into a fully interactive dashboard environment to support project oversight and reporting. This work requires balancing detailed operational and administrative responsibilities with broader strategic coordination across multiple projects, timelines, and reporting requirements.

The challenge is to develop a single, integrated system capable of tracking the full lifecycle of a grant or project. This includes monitoring income and expenditure, managing employment contracts and staffing timelines, and supporting overall project oversight, compliance, and reporting requirements. The system should not replace UCT systems but support and cover the information gaps. I also wanted to explore how AI could support more complex project coordination tasks without disrupting my existing workflow. The challenge was not simply speeding up tasks but finding ways to work more intelligently while still remaining critically engaged in the process.

To support these responsibilities, I use a range of AI tools for different purposes, depending on the workflow requirements and the type of output needed. I use ChatGPT and Gemini primarily for search support, writing enhancement, and prompt development. NotebookLM assists with the development of slide decks, posters, and infographics, while Consensus supports research activities and literature reviews. I use Claude extensively for prompt development, creating custom skills, dashboard design, and application development.

I found that providing detailed context upfront significantly improved the quality and relevance of the outputs generated. My prompts would typically begin with an explanation of my role, the overall objective, the operational or project context, and the specific task requirements to ensure that responses were aligned with the intended purpose and audience.

For example, one prompt included instructions such as:

“I am the Centre Manager of a research group based at a university. I need an Excel unit cost template built from our existing financial data for FY2025, with projections through to 2028.”

I then specified the source files (with anonymised data), staffing assumptions, worksheet structures, cost categories, formatting requirements, and technical instructions such as freezing panes and avoiding formula errors.

This approach helped produce outputs that were more aligned with the operational realities of the work. The more detailed and intentional the prompt became, the more useful the AI-generated output tended to be.

The combined approaches used varied broadly to guide the output more clearly as exemplified in the below instance:

“I am the Centre Manager of a research group based at a university (i.e. role-based prompting). I need an Excel unit cost template (.xlsx) built from our existing financial data for FY2025, with projections through to 2028 (goal-oriented prompting).

Source data: Pull 2025 actual costs from our existing workbook ‘file title’ in my selected folder — specifically the Budget Unit Costs 2024-present and PIV2025 sheets (contextual and source-grounded prompting).

Staff count: 30 FTEs (used as the divisor for all per-staff unit cost calculations) (constraints-based prompting).

Structure the template across three sheets (highly structured instructional prompting):

Sheet 1 – Unit Cost Template: Organise all costs under three sections, each showing Annual (R), Monthly (R), and Per Staff (R) columns for 2025 actual and 2026–2028 projections:

• Section A – Facility/Rental: Office/Offsite Rental (154.27m² × R250/m²/month)
• Section B – Operating Expenses: Telephone, Office Consumables, Stationery, Photocopying & Printing, Internet Connectivity, Vehicle Management, Office Equipment (R&M)
• Section C – Infrastructure Charges (per staff): Library Expenses, Network Costs, IT Support

Ensure zero formula errors, freeze panes on each sheet, and save to my selected folder (precision prompting).”

What made this prompting approach effective is that it combined role context, task goals, source grounding, operational constraints, formatting instructions, and technical quality controls, to reinforce that the quality of the AI outputs depend heavily on the specificity and structure of the instructions provided.

Before using any AI-generated outputs, I reviewed them carefully for accuracy, formatting consistency, and alignment with the intended purpose. In financial or reporting tasks, I checked calculations, projections, and spreadsheet structures manually before relying on the outputs.

For written outputs I also checked whether the content aligned with the intended message, research context, and professional tone appropriately.

I also found that refinement was often conversational and iterative. If an output was not suitable initially, I adjusted prompts and clarified requirements until the result was closer to what I needed.

I am mindful of privacy and data management throughout the process. By creating and managing my own files and folders, the AI tools could only access the information I provided.

No sensitive or personal information was included in the examples described above.

AI proved particularly effective for structuring documents, generating templates, assisting with summarisation, and supporting the development of presentations. It also helped to streamline repetitive setup tasks and provided strong foundational outputs that could be refined and adapted further as needed.

One of the greatest advantages was that the process became iterative rather than disruptive. I was able to refine and improve outputs conversationally while continuing with my normal workflow, instead of interrupting work processes to recreate documents manually from scratch.

I also found considerable value in using different AI tools for different purposes, depending on their individual strengths and capabilities

The quality of the outputs depended heavily on the quality and clarity of the prompts provided. Where prompts lacked sufficient detail or contextual information, more engagement and refinement were required before the outputs met the necessary standard.

Critical review, contextual understanding, and professional judgement remained essential throughout the process. While AI was highly effective in supporting drafting, structuring, and content development, it could not fully replace the human understanding required to assess whether an output was appropriate, accurate, or aligned with the specific institutional, operational, or project context.

Human oversight was therefore necessary to validate outputs, refine content, ensure relevance, and apply the level of nuance and judgement required in a research management environment.

AI has become integrated into several aspects of my workflow. The use of these AI tools has improved efficiency in a range of operational and administrative processes, while also supporting more structured and streamlined approaches to information management and content development. The process has also provided a valuable learning opportunity. Through engaging critically and responsibly with these tools, I have developed a stronger understanding of both the capabilities and limitations of AI. My knowledge and skills in AI-assisted workflows developed progressively through practical application and experimentation across a range of operational and project management tasks. 

I would avoid using AI in situations involving sensitive or confidential information, particularly where privacy considerations or institutional restrictions about sharing data apply. I would also avoid relying solely on AI outputs for work requiring specialist interpretation or final decision-making without careful human review.

Develop a strong prompting style. Clear instructions, contextual information, and specific goals make a significant difference to the quality of outputs. It is also important to engage critically with AI tools rather than assuming the first response will always be correct or complete.

Treat AI as a collaborative support tool rather than a replacement for professional expertise.