Most AI investments fail not because of the technology but because of the lack of foundation. A successful AI foundation depends on three often-overlooked pillars: data quality, AI policy, and governance. Without clean, consistent data, AI tools produce flawed outputs. Without clear policies, usage becomes inconsistent and risky. Without governance, neither data nor policy is enforced.
There’s a lot of time, money, and effort being invested in AI. And you already know that because there is a lot of time, money, and effort being invested in writing about AI, talking about it, and speculating about where it will go next. But I’m not going to do that here. Instead, I want to write about three things that can make or break all of those AI investments. Three things that aren’t getting nearly as much attention as they should.
What are they? Data, policy, and governance. Let me explain.
Data
Data is the fuel that powers AI. Without data, tools cannot ‘learn’ about the environment that they operate in, and they cannot generate meaningful recommendations, analysis, solutions, or actions. Yet many organisations lack complete and accurate data. Historic inconsistencies in the use of tools and the application of processes have resulted in disconnected data along with inconsistent (and incompatible) data structures.
Ineffective audit and control functions have resulted in missing and inaccurate data within those databases. And there is no easy way to distinguish between high- and low-quality information. Feed that to an AI tool and the results will be a long way from what is expected – or required.
To ensure that AI investments are capable of delivering meaningful business value, organisations simply must address their data deficiencies as a priority. While it may not be possible to correct all historic data gaps and errors, work must be undertaken to ensure a consistent data model across the enterprise combined with robust processes to ensure that all future data is complete and accurate. Where possible, remedial data cleansing and rebuilding should also be pursued.
Finally, the limitations of current data must be acknowledged, and those limitations must be reflected in the work undertaken to train AI tools. Otherwise, the problem will not only get worse as AI usage expands, the errors and gaps will be lost, leaving individuals, teams, and the entire business unaware that their AI enabled decision making and guidance is based on flawed data.
Policy
AI is disruptive. It is unlike any technology that has gone before, and needs to be managed differently from those other technologies. In addition to all of the standard policies around the acquisition, maintenance, and use of technology, organisations must implement a number of policies specific to AI.
Most importantly, security and privacy of enterprise and customer information is paramount, and in some scenarios, it may be necessary to completely prevent AI solutions from accessing them. Where access is being considered, contracts and agreements may need to be modified to ensure that explicit permission is given for the use of information in AI applications, particularly for customer data.
Policies must also address the need for consistency of application of AI across all business areas and functions, and it must define the acceptable use cases for AI. Such policies must include who is allowed to use the tools, the circumstances where such use is appropriate, and the need for transparency in reporting and recording that use.
Most important of all, such policies must be reviewed, and they must evolve, regularly due to the rapid nature of AI advancement and the continuously evolving regulatory and legislative landscapes.
Governance
Essential though policies are, on their own they are not enough. Organisations must also establish appropriate governance to ensure that those policies are managed and enforced. Such oversight must be the accountability of a senior individual, and cannot itself be solely run through technology.
In the short-term, such governance around AI is likely to be focused more on education than on enforcement, helping functions and departments to leverage AI appropriately and consistently. However, it will need to shift to ensure that the organisation is not only complying with its own policies, but also with appropriate regulations and legislation.
Governance needs to evolve constantly to reflect emerging AI capabilities and their use, both within enterprise operations and for products and services developed by the organisation. It will also need to reflect the rapidly changing regulatory landscape that will occur as legislation catches up to the technology.
The bottom line
AI offers significant opportunity to organisations. It can already help streamline operations, and it offers the promise of enabling better decision making, greater innovation, and improved business value. But, like any other technology, it needs to be used in the right way. And, like any other project, the work to deploy it and utilise it must be done in a planned and structured way.
That includes ensuring that the prerequisite work is carried out to enable AI initiatives to succeed. That’s where data, policy, and governance become critical factors. Businesses who don’t address these areas will never be able to optimise AI.
