Most employees who receive access to an AI tool start without instruction on what it does well, where it fails, and what company data should never enter a prompt. ChatGPT and Copilot training for companies through HighSpark covers this before daily use begins. Hands-on sessions are conducted by a group of 8-20 Singaporeans. In this program, participants can practice prompts relevant to their current position rather than generic ones based on real-world scenarios.

Tool strengths and limits

During the first month, employees wasted their time on tasks that the tool couldn’t handle. It is possible to improve the analysis of problems, the drafting of drafts, the summarising, and the reorganization of long documents by using language models. Sources are required in order to verify data, obtain live data, and produce accurate figures. Using ChatGPT, finance analysts can get plausible, but potentially incorrect, answers to numbers. An analyst can compose a first draft of a narrative in less than two minutes when using verified data. A training program should start by defining the differences between tools that save time and those that generate risk.

Prompt writing for real work

Most employees open a prompt box and type something short and vague. The tool returns a generic result. An employee soon realizes that the program is of no use to them. Tools are specified by their roles, tasks, output length, and constraints. It is easier to summarise this meeting transcript for a non-technical director than to summarise this. HighSpark participants receive prompts to write on their own files during the session. Three to five tested prompts are matched to their actual weekly tasks.

Data safety first

Contract text, client names, financial projections, and internal strategy documents should not enter an unapproved AI tool. Most employees do not know which plan their company holds or whether data they enter is used toward model training by the provider. Training covers this directly. During the training, participants will learn which plan the company uses, its data policy, and which categories are not included in any prompts. In contrast to an email circulated by the team, one group session develops a consistent standard across the team. Various interpretations resulted from each employee reading it once and interpreting it differently.

Copilot inside microsoft 365

Copilot works differently from a standalone chatbot. Based on files, calendars, and emails previously created by the user, context-aware output is generated. People who are unfamiliar with it treat it as a text input without knowing what they are entering.

Company sessions use Microsoft 365. Using Outlook, the response is drafted, PowerPoint is used to create slides, and Word is used to extract the summary document from the role’s personal file. Copilot allows participants to quickly start writing emails and documents after graduation by using their calendars, email inboxes, and recent documents.

DBS, Abbott and IDC have sent staff through this model. During a half-day session, you will learn more about one tool in depth, which is suitable for teams implementing ChatGPT or rolling out Copilot. There will be a second coaching round for each tool during a full-day session. It does not require any coding knowledge. Practising skills from one session the night before can be transferred to a later session, and the data handling habits cultivated in the session are carried over to future participants at no additional cost.