< Back to posts
06 Aug 20:46, by Phil Jordan
It must be the techs fault...
A friend shared this article recently from Gartner suggesting that 40% of all Agentic AI projects will be cancelled by 2027 as executives lose faith in their delivery due to rapid cost increases, unclear value delivery or risks that are not robustly controlled.
There's a part of me that feels like this may be true because people aren't starting correctly or getting the fundamentals correct in the approach, the goals, the expectation set by vendors and from a technical standpoint, they think it's a magic box, when really it's a workflow that needs engineering.
After all "agentic AI" is more of a collective noun for multiple technologies than it is just one thing.
Where I see most people struggle with AI delivery is the attitude and skills they start with. Rather than our typical interpretation of evolution where progress is made slowly over time, our attitude needs to change to suggest we are taking evolution into our own hands and re-engineering all of our operational processes with AI and automation as a native technology.
The shift away from what you’ve always known is not a comfortable one. I have worked with a number of people across different industries delivering change, and these are some of the repeatable challenges I have observed;
- Within many organisations, operations and technology are typically separate with each acting as a customer of each other with technology implementation being treated more like a “Technology Led, Operations Consulted” approach rather than “Operationally Led, Tech Enabled”. I have witnessed many AI and automation “projects” that become just that. Short term, tactical and tech led. As a result, the value that's gained is limited. In contrast, I believe those businesses that integrate tech and operations into a product-led function will have a better chance of success.
- Businesses struggle to articulate what the tech will do at each stage, with the correct KPIs to communicate the improvement it will have on time, quality and cost. In order to overcome challenges, doing the hard work of an up front operational assessment will make everything after a lot easier but it's often the step that businesses attempt to shortcut.
- Sponsors struggle to create a clear business case with data down to a low enough level of detail to make information driven decisions. Even if they manage to do this, most businesses don’t anticipate the longer term investment required to drive continuous improvements that incrementally increase benefit.
- Senior stakeholders are uncomfortable with the unknown and something as ethereal as AI can be an intimidating undertaking without a really robust and data driven assessment of value and risk. You are asking decision makers to back a complex operational re-engineering and without all the information and questions being asked and answered, how can they be expected to do so? This might be easier now that there’s an arms race to adopt AI, which has put external pressures on executives and businesses to remain competitive, but we shouldn’t skip the basics.
I'm not undermining these challenges as many businesses are entering into these projects with limited experience of delivery. I was there once too with all the hurdles that came with convincing senior stakeholders, at a time where AI was far less established.
The biggest change businesses need to undertake is that these technology implementations are the starting point of a new operational department, whose role it is to create new methodologies of customer communication and transaction completion which is completely different to the traditional “human led” product delivery. With that comes a new way of thinking and a new approach:
- The team you build and the skills you need: Make sure you have the correct balance of process engineering, operational planning, automation and data / integration expertise. Around this team will be all the support functions that come with any major operational change including risk management, compliance, quality assurance, finance and senior stakeholders representing operations and technology.
- Getting Started: Start in a way that you can test what works and what doesn’t with limited impact on the customer and the operation. When you have started, how do you continually remain focussed on adding and proving value, solve the inevitable early release challenges and prove to stakeholders that you’ve got this under control.
- Monitoring performance and resilience should be at the forefront of your mind with every new release and day to day running. This is the new operation, and like any other you wouldn’t let it run itself without continually understanding its output and quality. Being prepared for the known knowns, known unknowns and unknown unknowns is only done through experience, and understanding the process being built and everything that feeds into it to make it work.
- Upholding self-correcting principles that any operational function needs to be successful. Automation processes are not a “one and done” endeavour. Making sure you are honest and candid about the performance of the process, services that feed it and your own priorities you can build a team culture of continual refinement and improvement. The business will be looking to your team as the experts, but being humble and truthful are the traits that will help you become that.
- Autonomy of decision making and priority. Someone needs to be the central voice and owner for the AI and automation roadmap, who can translate the overlapping priorities of the customer, business and technology. This is not an easy job and it’s also not part time. It requires investment of both money and trust from the business and this needs to be reciprocated with honesty, clear communication and most importantly success from the owner of the roadmap. They can then decide how to balance business case value, process quality and maintenance to make sure everyone gets the outcomes that satisfies them.
Knowing where to start is hard enough, but then you need to convince operations, technology and finance stakeholders that this is a good idea. If you don't want to fall into the 40% that abandons their AI implementation, make sure you have your mindset and structure right before you start.