Imagine rushing your sick aunt to a regional hospital in Bariadi. She is visibly unwell, but a long queue is at registration. You go to the registration desk and politely ask if they can attend to her while the administrative issue is resolved. 

The response is familiar: “The system is down.”

In your desperation, you stop a passing nurse and ask what else you can do. She then directs you to a nearby QR code that links patients to another digital service. But your phone has limited network coverage, the battery is running low, and at that moment your concern is not digital innovation. Your aunt needs medical attention.

The aforementioned example is illustrative, but the underlying question is very real: when technology fails, have we designed an alternative way for the service itself to continue?

The question increasingly occupies my mind when I hear the phrases data-driven, digital transformation, artificial intelligence (AI) and innovation. I strongly believe in all four, but I am becoming equally convinced that the next stage of our technological progress must focus less on what we can build and more on whether we are solving the right problem.

The distinction can have consequences far beyond convenience. As the illustration shows, it can be the difference between life and death.

Building has become easier. Judgement has not.

Until recently, developing software required substantial programming expertise. Today, generative artificial intelligence is rapidly changing that equation. Through natural-language prompts, increasingly capable systems can generate code, analyse data, and help people with limited formal programming backgrounds develop prototypes and applications.

We have even acquired a new phrase for part of this phenomenon: vibe coding, which describes the increasingly accessible practice of using AI instructions to generate and modify software. Personally, I find this development exciting, since it democratises technological creation.

READ MORE: I Stopped Correcting Reports and Taught My Students to Think Instead. Here’s What Happened

However, in my opinion, it does create another serious challenge: as the cost and technical difficulty of producing software decline, we may see an enormous increase in the supply of digital solutions without a corresponding improvement in our understanding of the demand side – the people expected to use them.

Building an application is becoming easier. Understanding whether to build an application in the first place remains complicated. In short, AI is lowering the technical barriers to building software, but it does not reduce the need for judgement about what is worth building, for whom, and to what end.

This is where quantitative professionals should become particularly reflective.

Data is not the destination

Quants often describe data as the new oil, gold or even the most valuable commodity of the twenty-first century. I prefer a more cautious description: data is a strategic asset. Its value depends on its quality, relevance, representativeness, timeliness and, increasingly, how responsibly it is governed.

Data by itself is not wisdom.

Being data-driven should not mean abandoning professional judgement or letting numbers make decisions for us. Senior decision-makers, including those in government and public institutions, have a responsibility to practise evidence-informed decision-making by combining empirical evidence with professional judgement, institutional objectives, uncertainty, and contextual understanding.

They do not necessarily need to reproduce the calculations behind every model. They should, however, understand what the model is intended to measure, the assumptions it relies on, its limitations, and the consequences of getting the decision wrong.

This is because a sophisticated model can still answer the wrong question. A beautiful dashboard can display the wrong metric. An accurate forecast can still lead to an inappropriate intervention.

And a perfectly functioning application can solve a problem that its intended users never considered their most important problem in the first place!

The proper chain should therefore be: data → information → insight → decision → implementation → outcome.

Yet I sometimes worry that we stop somewhere around “information” or “insight” and congratulate ourselves (especially quants). The citizen, customer or policyholder lives at the end of that chain. They experience the outcome.

READ MORE: Tanzania Court Adopts Artificial Intelligence (AI) in its Processes

What does this mean in finance?

Consider a credit-scoring model developed for a financial institution. Statistically, it may be excellent. Its predictive accuracy may satisfy every internal validation test. But suppose it depends heavily on formal transaction histories, regular electronic payments and structured employment records. 

What happens when the potential borrower is a profitable informal trader whose economic life is largely recorded outside those systems, as is the case for many informal workers in Tanzania?

The model may perform exactly as designed and still fail to represent the economic reality of the customer standing in front of it.

Consider an insurer that digitises its claims process. On paper, this is progress: shorter processing times, better fraud controls and lower administrative costs. But suppose every claimant must upload several high-resolution documents through a smartphone application.

For a policyholder with an older handset, limited storage or unreliable connectivity, the claims process has not necessarily become easier. We may have simply shifted part of the insurer’s administrative burden to the customer.

Or consider a Defined Contribution (DC) pension fund with an impressive, regularly updated management dashboard: assets, liabilities, contribution densities, investment returns, actuarial projections, and every key performance indicator updated beautifully. That information is genuinely useful to management, but what does the member understand?

Can a 45-year-old contributor look at their information and answer a much simpler question: “At my present contribution pattern, what might my retirement income actually look like, and what can I do about it now?”

That is where quantitative sophistication meets professional communication. The model’s purpose is not merely to calculate. It is to support better decisions.

Tanzania’s digital reality is encouraging, but currently uneven…

Tanzania has made remarkable progress in digital connectivity and financial inclusion. That progress deserves recognition.

Recent regulatory statistics show tens of millions of internet subscriptions in Tanzania, alongside continued growth in smartphone penetration, with mobile broadband remaining the dominant form of internet connectivity.

But those statistics themselves illustrate why responsible interpretation matters. A subscription is not necessarily the same thing as an individual enjoying reliable, affordable and continuous internet access. One person may hold multiple subscriptions. Network availability does not automatically imply smartphone ownership, sufficient data, digital literacy or confidence using a particular service.

Worse, national averages can hide geographic differences. According to the 2022 Population and Housing Census, a substantial majority of Tanzania’s population lives in rural areas.

This should influence how we design services intended for the entire country. 

READ MORE: No Policy, No Shield: Why Tanzania Is Still Unprepared for the AI Era

The question therefore should not simply be, “Can this service be digitalised?”

It should be, “Under the actual conditions in which our intended users live, what combination of channels provides the most reliable service?”

Sometimes the answer will be a smartphone application. Sometimes it will be a web portal. Sometimes USSD or SMS may be more appropriate.

For essential services, resilience may require an assisted or controlled manual alternative when the primary digital system fails.

That is not resistance to technology. It is simply good systems strategy, also known as the design thinking process.

Access is not the same as usefulness

The financial sector offers another useful lesson. Tanzania has recorded substantial progress in access to and usage of formal financial services. Yet disparities remain among groups including rural residents, smallholder farmers, young people and women.

That distinction between access and usage is important. Now, you may ask, what do you mean?

Access does not necessarily mean understanding the services. Usage does not necessarily mean satisfaction. Most importantly, digitalisation does not necessarily mean inclusion. 

Responsible quantitative analysis asks what lies behind the headline percentage. It asks about the denominator, the distribution, the outliers and the people hidden inside an average.

Our young population should be part of this conversation.

Tanzania is also remarkably young. Census data show that a large majority of the population is children and young adults (62% for ages 0-24), representing extraordinary intellectual and innovative potential.

But young people cannot meaningfully participate in economic conversations if professionals communicate important information almost exclusively through technical dashboards, lengthy institutional reports and unexplained statistical indicators.

Perhaps the problem is not that younger citizens are uninterested in serious issues. Perhaps we have not always made serious issues sufficiently accessible.

Professional communication is therefore not an optional “soft skill” attached to quantitative work. It is part of the work itself.

If I develop a technically excellent model but cannot explain its implications to the decision-maker, customer or citizen affected by it, I have not completed the analytical process.

This is where training institutions and industry must meet.

I believe this has implications for quantitative education in Tanzania. Universities must certainly produce graduates who understand financial mathematics, statistics, actuarial models, machine learning, econometrics and programming. Students should learn Power BI, Tableau, Python, R, SQL and modern analytical tools.

But industry needs something more

We should train quantitative graduates to ask: What decision are we trying to improve? Who will use this analysis? What assumptions drive the result? What happens if those assumptions are wrong? Who bears the cost of that error? Can the intended user realistically use the solution we are proposing?

Those questions require mathematics, domain knowledge, ethics and communication simultaneously.

Industry, in turn, should help universities understand the operational problems that exist outside lecture rooms. The relationship should not begin when companies arrive to recruit final-year students. It should exist throughout the educational process through case studies, guest lectures, internships, research collaborations and genuine problem-solving partnerships. That is how academia–industry linkage becomes more than a conference slogan.

This is another dimension we cannot ignore. Just because data can be collected does not automatically mean every data point should be collected.

READ MORE: Data Protection Commission Issues Final Warning Ahead of April Enforcement Deadline

Responsible innovation must consider privacy, proportionality, accuracy, transparency and security. Tanzania’s personal data protection framework itself emphasises principles including lawful and transparent processing, purpose limitation, data minimisation and accuracy.

For quantitative professionals, responsibility therefore extends beyond the accuracy of our algorithms. We must ask whether the data is appropriate, whether the methodology is fair, whether the assumptions are transparent and whether the resulting decision can be defended.

Back to Bariadi

Let us return to the aunt waiting for treatment. Her problem was never the absence of an application. Her problem was access to a service.

Technology should make that service faster, safer, cheaper or more reliable. Data should help us understand where delays occur, who is affected, how frequently the system fails and which intervention produces the greatest improvement.

If the analysis tells us that an application is the best answer, we should build an excellent one. But if the analysis tells us that the better answer is USSD, stronger network redundancy, better staff processes, an offline mode, or simply a reliable fallback procedure, we should be equally willing to accept that conclusion. That, to me, is what being genuinely data-driven means. 

It is not beginning with the solution we want to build and searching for data to justify it, but beginning with the problem, understanding the people who experience it, examining the evidence honestly and allowing that evidence to lead us towards the most appropriate solution.

Because ultimately, the measure of quantitative sophistication is not how complicated our model is, how colourful our dashboard appears, or how many applications we launch.

It is whether, because of our analysis, someone makes a better decision, receives a better service, carries less risk, or lives a slightly better life.

Agnella Nemuo (PhD) is a quant and academic specialising in financial mathematics and actuarial science. She’s available at agnella.nemuo@gmail.com. The opinions expressed here are the writer’s own and do not necessarily reflect those of The Chanzo. If you are interested in publishing in this space, please contact our editors at editor@thechanzo.com.