Ask any organization, employee, or customer what they want most from a chatbot, voicebot or AI assistant and accuracy will be in the top three of answers. But accuracy is not a given. It is directly correlated to the usability and quality of the knowledge base. And often, the quality of information is poor.
AI accuracy requires quality, structured and governed knowledge.
Better not more
Most enterprises don’t need more information; they have an enormous amount already. It can be found spread across many systems: from SharePoint libraries to CRM systems, emails, Wikis, PDFs, policy documents, and teams’ conversations, to name a few.
There is a lot of information in organizations. Quantity is usually not the issue. Its findability, reliability and usefulness are. AI doesn’t fix this. What it does do, however, is expose it.
Duplicated, out of date and contradictory information is of no use to an employee. And it’s of no use to AI if accuracy is a priority. Being able to retrieve an answer is easy, but to retrieve the right answer requires the right information.
Governance, structure, and quality = useful and usable
In organizations where multiple documents contain contradictory information and/or documents are out of date (e.g. old policies, organizational structures, or workflows), this is the information that the AI will retrieve. It is not inventing it; it is using what it has access to. And it doesn’t know if it is outdated.
Without the correct knowledge governance, accurate AI is compromised. AI is not going to solve this problem, that requires knowledge management and important ownership.
Better knowledge equals better AI accuracy
Every organization serious about AI and committed to ensuring accurate, responsible, and ethical AI must pay attention to its content. Knowledge management must be a governed, controlled and priority discipline practiced organizational-wide.
From content governance, ownership, metadata, and version controls to document lifecycle, taxonomy and permissions, organizations must commit resources and budgets to knowledge management. Without this focus the return and business value that AI has the potential to deliver will be minimized.
Fit for purpose knowledge base
The first step in any conversational AI initiative must be to look at knowledge management.
Carrying out an information audit to identify duplicated content, obsolete documents, conflicting guidance and correcting this must be a priority.
Reducing duplication and ensuring there is just one authoritative source rather than multiple copies will result in better quality knowledge. And assigning an owner for every important document with responsibility to regularly review and update will ensure ongoing accuracy.
Quality content needs structure and reviewing and implementing improvements in headings, metadata, tagging and searchability will contribute to greater AI accuracy.
Whilst these steps are critical at the outset of a conversational AI initiative, information needs continuous governance. Knowledge management must be an ongoing, continuous discipline with dedicated resources.
Just because information exists doesn’t mean that it is trustworthy, useful, or usable.
Well governed knowledge produces better AI outputs which means more accuracy, trustworthiness, and consistency. AI outputs mirror knowledge quality.
Don’t blame the AI
Most organizations are embedding AI into workflow. The quality of conversational AI solutions and the value it brings to the company will be dependent on the usefulness and usability of the information being used.
Organizations that want to be trusted and known for accurate and responsible AI practices cannot ignore the criticality of knowledge management.
