AI is only as good as the data that feeds it – businesses who don’t have that in place will never reap the full benefits of an investment in AI.
This is according to Peter Clarke, CEO LanDynamix, who says: “Bad data and poor prompts produce bad AI results. Poor-quality data leads to flawed insights and biased decisions,” says.
Clarke emphasises that before scaling AI, businesses must assess the quality of their input. “Business leaders need to understand that AI thrives on predictability. Standardising formatting and naming conventions is essential. Defining the conventions for all types of data ensures data can be matched correctly,” he says.
Clarke lists the following as his top five tips for businesses to adopt if they want their data to be AI-ready.
- Standardise formatting and naming conventions. AI thrives on predictability. Customer ABC is not the same as Customer ABC (PTY) LTD. The same goes with formatting, Smith, John isn’t the same as John Smith. Defining the conventions for all types of data makes sure data can be matched correctly.
- Establish a single source of truth: Eliminate information silos and establish a single source of truth. When data sites are located in multiple places such as email, spreadsheets and paper notes, AI can only see a fraction of the picture. Consolidating data into a single system that becomes the source of truth allows AI to see the big picture. Using a centralised CRM system can bring all the information together.
- Implement strict data hygiene: Incomplete, duplicated and outdated data will skew AI results. It’s essential to possess well documented processes that deal with obsolete data, merging duplicates, and correcting typos. Check if the systems you are using have built in de-duplication tools to help clean up the information and maintain integrity.
- Define and enforce access controls. Before allowing AI tools entry to company data, access controls need to be inaugurated to ensure sensitive data does not leak. Modern AI tools read whatever they get access to. This means that if an employee asks the AI assistant a question, it may pull the answer from sensitive data the user should not have access to if the permissions are not in place. Authorisations must be reviewed regularly.
- Establish metadata and a defined file structure. Move away from vague file names such as invoice1.pdf and draft_final_v2.docx. Use a structured name like YYY-MM-DD_ClientName_DocumentType_Vxx.pdf. Also add metadata to documents, such as authors, tags and creation dates. This will provide the AI tools valuable data about the documents it’s reading.