How to find your best AI Automation opportunities
It is a truth, universally acknowledged, that a single consultant in search of a fortune, must invent a 2x2 strategy matrix.
Here’s mine.
Faced with the question: “What should we be doing with AI?” many teams freeze up. It’s not that there is a lack of ideas – quite the opposite; the challenge is rather knowing which ones to pursue, and how.
In this post, we’ll introduce a simple, 2x2 matrix that we use to answer this question. It will help you to organise your ideas along two dimensions, thereby determining how to approach each one and which ones to pick first.
The MAtrix
The dimensions against which your ideas should be scored are:
Complexity. The expected technical challenge of achieving an AI-based solution to the identified business need. Factors that can contribute to increased complexity include: a lack of data to train AI models, technical dependencies like data pipelines or repositories that need to be developed first, difficult governance and regulatory requirements or complex business processes with many human/machine interface points.
Utility refers to the impact that the AI solution would have on the wider business. Factors that contribute to utility include: the size of the user-base for a new capability, expected efficiency savings, whether the capability unlocks opportunities to scale or new revenue opportunities or whether material improvements to quality and risk management activities are expected.
After assessing an initiative against these two dimensions, it can be placed into a quadrant which dictates the best approach to implementation.
The four quadrants
Here’s what it all means:
Low complexity, low utility: empower the experts.
Businesses often have a large number of opportunities to improve workflows but the majority of these will affect the activities of only a handful of colleagues. The best approach here is to introduce some “no-code” AI tooling into the organisation and empower users to develop their own “apps” to solve their daily needs. Typical choices here include MS 365 CoPilot, Anthropic’s Claude or Google’s Gemini for generative AI and n8n, lindy.ai or crew.ai for agentic applications.
I recommend a three-step process to introduce such tools:
Get the governance right. Ensure that you have the right policies, procedures and guardrails in place to ensure that the AI is used in an appropriate and controlled manner.
Roll the AI tools out slowly, expanding the user groups slowly as you learn how to effectively use it.
Help your early adopters become “power users” of the tools; use training or outside assistance to help them learn how to get the most out of the tools. Then, have colleagues share and learn from each other as you expand the user groups with access.
High complexity, low utility: park and monitor
Sometimes, a task may not well be solved by existing AI technologies and services – or at least not at a cost which makes economic sense for your use case. In these cases, we recommend conducting a periodic “market scan” of vendors to monitor their offerings. In a fast-moving field like AI, it may not be long before a solution is brought to market, or AI models become better at solving more complex tasks.
Low complexity, high utility: demonstrate and iterate
For use cases where no material blockers prevent immediate exploration and the value to the business is material, it makes sense to dive right in and prototype an AI capability. The prototype should be quick and low cost to produce. Aim to have results within 20-30 days. The objective should be to get real users to evaluate how well the AI does at solving a real problem for them – therefore, make sure the prototype can actually be used in day-to-day work. “Shop window” demonstrators do not give you enough insight into the utility of your approach, only actual usage does that.
(If you have an idea that’s ready to go, we’d love to help you build it! Check out our Demonstrate service for more information.)
If the demonstrator is successful, follow good Agile software development practices to iterate your solution, turning it, over time, into a fully-fledged capability.
High complexity, high utility: lay the foundations
Oftentimes a business challenge may be ripe for tackling with AI but there is groundwork to lay first. In such cases, you are best served by developing a “milestone plan”, covering the activities you will need to undertake before you can introduce AI or machine learning. Often this work involves building and populating data stores which – although complex and time-consuming initiatives – can yield broader benefits across the organisation.
What Next?
Now that you’ve categorised your ideas, the next steps are clear.
Pick the lowest complexity, highest utility idea from Demonstrate & Iterate and get building.
Gather the best ideas from Empower the Experts and convene a working group to discuss governance and AI software options.
Start putting together a milestone plan for your highest utility idea in Build the Foundations.
Do a bit of “market scanning” for the most promising vendors for the ideas in Park & Monitor.
You’re no longer paralysed with choice: you have a path to travel.
We at Veratai offer a structured workshop which will help you build out this matrix – among other things – in half a day. Check out the Determine section of this website.