When organisations embark on data projects, with AI as the aspired outcome, there are a bunch of preconceived ideas about the challenges they’ll face. You only need to go so far as Googling the failure rates of AI and data projects to understand why Exec teams invest so heavily in de-risking their strategy so they don’t meet the same fate.
This perception that it’s really hard is fed by a lot of consultants and vendors… after all, that’s in their best interest. Equally, with AI tied up in most organisations complex “transformation” plans, these projects meet a lot of fatigue. It’s a new capability to be built - and it requires organisations to be receptive to change. So, naturally, it’s understandable that there’s still a lot of pessimism lingering about the value (and practicalities) of AI.
That being said - as someone who’s worked in data science and deployed successful machine learning products inside large organisations and now working on the other side, I wanted to break down some common myths and traditional thinking which I’m seeing hindering AI projects.
Here’s my take on some thinking to challenge internally if you’re finding your plans to utilise Machine Learning are meeting roadblocks time and time again:
Myth #1 Developing and deploying ML applications is hard and time-consuming
The reality: There’s been a massive shift over the past couple of years, taking ML away from ‘science’ and into ‘software’. Less maths, more productisation! This means that there are products already built and ready to be deployed within days to your environment.
While there’s no question that deploying models into a live environment comes with other challenges - particularly if you’re in a larger/bureaucratic business, it’s no longer a hard thing to do from a technical perspective. There’s no reason that building applications should take months - data capabilities and the technology have evolved past this point.
Myth #2 Competitive AI insights will be captured with your current data
The reality: Most data collection practices were not implemented with the intention of then using the data for AI. Meaning a lot of the time it’s actually not suitable for anything meaningful. Equally, many of the insights shown through common data (such as customer feedback, employee productivity, operational efficiencies) are (generally) not unique across businesses, and therefore, won’t provide a competitive advantage. More just opportunities for business improvement - which needs to be accepted as a desired outcome.
Myth #3 Your current data is valuable, period
The reality? Organisations looking to make fast and impactful gains with AI will acquire new data. The time lost collecting, cleaning and prepping existing datasets for machine learning outweighs the cost of acquiring new data to explore potential AI use-cases, and start laying the runway for future investments. Having valuable current data is the exception rather than the rule.
Myth #4 Your use-case/model requirements are unique
Most Machine Learning models are not unique. Teams don’t need to build bespoke applications. If you’re building a model for common business uses, such as for churn predictions, chances are, there’s an existing solution out there that can be used. Time is being lost thinking that the problems organisations can solve through the use of AI will be unique.
Myth #5 It will be a competitive advantage to develop this capability in-house
There’s a war for talent, the tech is evolving - fast, and internal projects are more likely to be burdened by legacy issues. Organisations pulling ahead are utilising networks and ecosystems of the right talent and technology that they need. They recognise that it isn’t their organisation’s purpose to be a leader in AI. Rather, AI provides the capabilities to be the leader in their actual purpose.
Myth #6 Data models need to be built and experimented on in the lab, not out in the factory
The unfortunate reality is that few data scientists inside large or risk-averse organisations can get their models live/into production to start demonstrating value. But the most meaningful learning happens outside of the lab - with real data from real business interactions. There’s no way to demonstrate a pathway to scale and wider use-cases with experimentation happening behind closed doors. Equally, while things may have looked great in the lab, very few models actually hold up to “reality” when faced with real world data.
Myth #7 Capacity and capability are the bottlenecks
In my experience… organisational resistance to change and technical red tape are the bottlenecks, coupled with a lack of understanding about how to adopt this technology. As mentioned, ML is more accessible than ever before - making the shift away from science to software. A single data scientist can be an AI hero with the right products at their disposal. Citing issues with finding the right people and bandwidth just doesn’t make sense in the current landscape as it did 5 years ago.
Is your organisation letting these myths hold them back from actually getting value out of their investments in AI?
I’d love to hear what other myths you’ve uncovered throughout your career!