Agentico publishes a six step process for developing an AI strategy at an SME. But In this blog I want to share some common themes from strategy discussions with clients.
“AI is just hype” vs “AI will take over the world”
Social media craves engagement hence dramatic but wildly opposing opinions get attention. If you are in the entertainment industry then that’s good news. It's true that these extremes are relevant to the likes of Google or Microsoft who must build tools not threats.
However, from the perspective of small medium sized enterprises, these scary stories are a simply a distraction from the business opportunity. Moreover, they alienate staff from otherwise exciting strategy conversations.
AI is intelligence on tap, it is clearly much more than hype. However, it is also far from world domination. For SME’s, AI is a tool for growth, it remains a competitive advantage for those ready to grasp it.
The Value Is In Your Unique Data
The AI revolution has been almost unique in history. Everyone has access to the same AI, at the same time, globally and irrespective of employment status or age. This is looking like the commodification of intelligence.
The only thing that makes you truly special compared to competitors and new entrants, is your data. Look after your data. But data is like capital, don't hoard it under the bed, make it go out and work for you.
Many AI products are little more than a wrapper around ChatGPT. Making your data available to the AI adds value. Chatting with that data is a convenient and introductory use case, but so much more is possible. I have discussed those possibilities in this blog “Think Outside the ChatBot”.
Experiment Fast
Beware engaging in large projects which get overtaken by developments. Instead, experiment fast and often, building confidence within the organisation as to where the opportunities really lie.
As ever, your best researchers are your own staff, eager to apply AI to make their life easier. Yes, this can resemble 'black market IT', but if at all possible, resist the temptation to ban all experimentation. Some companies have offered prizes for the best AI ideas, applications also gently alert the business to less wise applications.
As stated earlier, your data is what makes you unique. Even the most potent intelligence is only as useful as the data it has access to. The objective of experiments is to discover which data sources are useful to clients and staff. If you don’t have such data, then start collecting. Each experiment in building or using an AI tool is an opportunity to comb through data sources, tidying and verifying as you go.
Investment in tidying up data is rarely wasted. Having said that, don’t boil the entire data lake, focus on reliable data pipelines for valuable projects. Suitable approaches to tidying the remainder of the data will become apparent through experience.
Got a New Hammer? Everything Looks Like a Nail
AI is an exciting tool and the temptation is to try it on all new opportunities. This is natural and an important part of learning what AI can and cannot do.
Chatbots like ChatGPT are just one way of working with Large Language Models (LLM’s). AI agents are the other method of utilising LLM’s. Agents are arranged to automate tasks by either processing work in a programmed sequence, or work in teams to overcome tasks where the process is not well defined. That’s when the magic really happens.
Remember that many problems yield well to simple tools, straightforward rule based software or ‘statistical learning’. AI agents can then be empowered with access to those tools, just as a human employee would be.
Work Towards Outcomes Not Outputs
AI tools can deliver very impressive outputs, for example, elegant code to solve a problem or beautiful prose. But that is no guarantee it presents the best, or even a useful, outcome. AI outputs should be benchmarked whenever possible, else they give the impression of being useful but add only noise.
AI Needs Best Practices, Just As Employees Do
Data permissions, integration standards, approved frameworks, common cloud services and safety guard rails are all required to govern AI. We touch upon this in the ‘Culture’ stage of the AI Strategy process.
We would expect an employee to be given such guidance, AI is no different. It requires some initial experiments in building AI to know what the meaningful options for the standards are to what degree they should be applied.
And that's it for now. We'll keep this strategy section updated, its fascinating to see the common themes amongst clients.
