This is the third entry in AI Fluency Corner, a 16-part weekly series that builds one connected mental model of AI in plain language.
Open the Maps app. Two routes are displayed. One is short. The app will push you towards longer. Traffic is already increasing near the off-ramp. I can’t see the calculations. You will be prompted to “Follow this path.”
Fraud analysts take a different view of holding this transaction. The recruiter reviews another candidate. In other words, consider this candidate before that one. There may be other factors in the financing system. It’s a decline.
Last week we looked at data, the recorded evidence that AI can learn from. This week brings us the next layer. Once the data exists, you need to decide what to do with it. That something is an algorithm.
What actually is an algorithm
An algorithm is a step-by-step method for transforming input into output. Recipes are algorithms. Add the ingredients and serve the meal. Loan affordability calculation is an algorithm that takes into account income, expenses, interest rate, term, etc. I no longer have the ability to repay. Call center routing rules are algorithms that take into account complaint type, language, customer value, and agent availability. Cue position out.
AI did not invent algorithms. It made them more powerful, less visible, and harder to interrogate. In typical software, instructions are created by humans. If an invoice is 30 days late, flag it. If inventory drops below a threshold, reorder. All predictable cases are pre-coded. Reliable against known scenarios. As the world changes, it becomes fragile.
AI works differently. Instead of a human writing all the rules, the system is shown many examples (the data we discussed last week) and comes up with its own internal rules. Instructions emerge from data rather than being created manually. This is why navigation apps improve without developers having to rewrite the code every week.
Traditional software: A human writes the complete instructions first, then a computer executes them. AI: The system discovers useful instructions from examples and applies them to new situations. This reversal is worth maintaining.
Three algorithms have already decided things about you
The sorting algorithm determines what appears first. Gmail’s Focused Inbox didn’t have a list of important people programmed into it. I learned from your actions of quickly opening, ignoring, and replying. Importance is not a property of email. It is the output of the algorithm that studied you.
A scoring algorithm determines access. At Absa, Nedbank and Standard Bank, your loan application is processed by a model that takes into account your payment history, usage, account age, and many other variables. You never negotiate that weight. Scores are not judgments. This is a computation, constrained by the data supplied and the priorities encoded in the model.
What you encounter next is determined by the recommendation algorithm. YouTube, Spotify, and TikTok identify people who act like you and reveal their engagement. Your preferences are approximated by your history, filtered through the preferences of others. As a result, we can’t introduce anything that truly deviates from the pattern. This limitation is worth knowing.
Algorithms have priorities
All algorithms optimize something. Maps apps may optimize for the fastest routes, least tolls, or least amount of fuel. Banks may optimize for fraud prevention, convenience, compliance, or loss reduction purposes. These are not technical details. These are business choices and will be determined prior to arrival.
If fraud algorithms are tuned too aggressively, they can frustrate legitimate customers while protecting banks. When recruiting algorithms optimize for candidates who resemble past successful hires, they can quietly repeat old definitions of promise. The algorithm never pauses to ask if the goal was fair, complete, or up-to-date. It pursues goals given or inferred from history.
Algorithms do more than just calculate. Express priorities.
When you doubt your logic
Three situations merit scrutiny:
- When the stakes are high. Algorithms that inform credit decisions, job shortlisting, medical triage, or legal risk are not value-neutral. Before accepting the output, ask what the system was optimizing for and whose results were represented in the training data.
- If the population is different. A model trained on urban salaried workers may misbehave when applied to gig workers and temporary workers. Fraud models built on US transaction data behave differently in Joburg. This is a limitation that should be checked before deployment and is not a flaw in the concept.
- If there is no explanation. Many modern algorithms are unable to explain individual decisions in plain language. If your loan is declined or your profile rank drops, the system may not explain why in a human-readable way. When decisions must be explainable due to regulatory, fairness, or audit requirements, the choice of algorithm is just as important as the choice of data.
What this means for your job
Algorithms are embedded in more business processes than most managers realize, including invoice approval, staff rostering, customer segmentation, demand forecasting, and compliance monitoring. Most of it runs invisibly until something goes wrong.
You don’t need to understand math to speak fluently. To trust the output of an algorithm, you need to ask four questions: What was being optimized? What happens if it’s wrong? And who is responsible?
Algorithms that misprice products lose margin. Misclassifying customers destroys trust. If you make unsubstantiated lending decisions, you may be in violation of the National Credit Act. The results are not technical. They are financial, reputational and legal.
Automated reliability is not the same as accuracy.
our mission this week
Select one automatic decision that occurred this week (Route, Score, Recommendation, Approval). Ask what went in, what purpose it served, what if it was wrong, and how to challenge the results.
• Mafinyani is a senior partner in financial engineering and artificial intelligence at Intellica Analytics, a firm specializing in finance, risk and applied technology. Next week: Machine learning — how systems learn from data. If algorithms are instructions, then machine learning is the process by which AI creates those instructions for itself.
