AI is not ALWAYS the Answer

Now, regardless of your stance on AI and whether it is a bubble, an opportunity, a savior or our potential self-inflicted doom, the reality is that it is here to stay. What hopefully will change, though, is how we talk about it. Do you remember during the Pandemic how apparently all other diseases just took a break, and we could only get infected with COVID-19? Have a cough? COVID! Have a sneeze? COVID! Have a seat? COV- I mean CHAIR! You get the idea. Well, it seems the same is happening today in the realm of technology, where everything now somehow falls under the umbrella of AI. Which brings us to today’s lesson: AI. It is NOT always the answer.

AI needs careful consideration

Taking an innocent look at the variety of AI-powered offerings today, it becomes immediately clear that this is a buyer’s market. Therefore, buyer beware. AI, by its complex nature, requires significantly more consideration before purchasing. It is not a one-size-fits-all solution. It is not just another SaaS. It is definitely not “plug and play” (if you hear someone say that, hang up, run, avoid at all costs. You are staring down 12 months of billable consulting work to get off the ground and a cool €500,000 minimum investment).

It’s the difference between going to the store to buy groceries versus deciding to start your own vegetable garden in the backyard. It is not just a matter of convenience and short-term need. It is a long-tail decision that requires careful, thoughtful consideration: what will you grow? For private use or commercial? What tools do you need? Do you have anything already in place? How big is your garden? Will your neighbors be bothered by you choring in the yard? Are you equipped to handle any nosy critters looking to pilfer your produce? Put it this way: AI is not a purchase. It’s a commitment.

But say you are indeed inclined to commit to AI. You did the work and now know this is the direction you want to head. Here is how you can get started and protect yourself from bad decisions.

Define a clear purpose for your AI

Before any resources or time flows into this, consider its purpose thoroughly. What are you trying to achieve? Here are some examples to help you start out (and examples of when maybe AI is not exactly what you are looking for):

1. Production: We want to optimize processes within our manufacturing to reduce costs without affecting our product quality.

   
   

   AI may seem like an obvious answer here, but there are also other options to consider (read: cheaper ones): Robotic Process Automation (RPA) has been used over the last decade or so effectively in optimizing and automating processes, particularly in well-defined, repetitive production and operational processes.

   
   

2. People: We want to streamline our recruitment process by implementing filtering systems for CVs and cover letters so only those that fulfil our criteria get through to a human recruiter.

   
   

   While this may in fact not seem like a big deal, definitely not one worth AI consideration, picture this: a large consultancy can actually spend an approximate €25,000 per potential recruit (this includes software costs, labor costs for recruiters, interviewers, travel costs for applicants, and so on). That is A LOT of money for someone who will very likely not get the job. An AI-powered scraper with a large language model to produce reports on each applicant may very well be the way to go here.

   
   

3. Portfolio: We want to constantly optimize our product portfolio based on live sales performance across our categories and market trends to ensure we are investing in products best suited for commercial success both today and tomorrow.

   
   

   Another potentially great implementation for AI, but a highly complex one. It is best to break down such requirements into bite-sized components. For example, start with portfolio strategy (likely best done by a dedicated team of experts, not AI). Then an element of market research (also carried out by experts but supported by AI). Finally, there is an element here of continuous improvement, which has AI written all over it. Then consider how the human and AI “resources” need to collaborate: all work done by your experts needs to be planned out and executed in a clear, structured fashion so that, by the end of it all, their outputs can become inputs for an AI learning model to ensure they no longer need to constantly supervise the portfolio and the market, but a learning AI can do that moving forward.

   
   

I know, this may seem a bit overkill for a blog post, but it is important to understand these nuances. Primarily so people stop asking me “wouldn’t an AI be good to increase our sales”? Sure it would, let me just check my desk drawer in which I keep semi-sentient robots. I am sure one will do. NEXT!

Audit your current architecture and IT environment, and adapt it as needed

Now that you know why you need the AI and you have a fairly good idea of what your vision for this tool will be, it’s time for some introspection: are your current IT architecture and databases suitable for your ambition? An AI needs, above all, a database to learn from (read: a STRUCTURED database). That involves having your data architecture audited and, where needed, improved to fit the needs of your learning model.

While you are at it, also consider WHERE you will have that database sit. In the Cloud? How will you protect your data there? Which service will you use for that? Or will you set up your own servers? Where will you store these? How many will you need? How will you ensure their safety (both physical and in terms of cybersecurity)? Do you have in-house experts to manage these?

Long story short, it all starts and ends with data. If yours is not up to scratch, your AI will not have the environment it needs to get off the ground.

Map out in detail what jobs need to be done

Since your AI is “employed” to do specific tasks, start mapping those out similarly to when you hire for a new role in your company. Think through what that role (this AI) needs to accomplish: their responsibilities, day-to-day jobs to be done, whom they answer to etc. Do this at a very granular level, mapped out as a structured workflow. This helps down the road when programming your AI (or choosing an AI tool) by outlining clearly what it needs to be able to do, step by step.

Develop the model (or customize a purchased AI software)

In case you were keeping track: this is the point in the process where you can say “let’s get an AI” and actually mean it, because the one thing left to do is to go ahead and develop your model. This means either purchasing an AI tool that emulates the workflow you have outlined previously or developing one on your own. The tools are in place, the garden is prepared, the sun is shining, you are wearing your best farming clothes, it is time to get your hands dirty.

This is also the point where you will most likely need some outside support. But fret not, the steps you have taken so far will ensure that your partner in development will have a clear path forward and that you are aligned on what exactly needs to be done. This will hopefully minimize any misunderstandings along the way, manage expectations, and reduce the risk of needing to go back on your work and pivot your development. Hey, you can even go a bit crazy and be Agile in your development! Remember that can of worms?

I do… I can never forget it.