As the COVID-19 crisis continues, it is clear that the economy will be heavily impacted . While CTO’s and technical teams currently focus on coping with the operational challenges of the crisis, their attentions are already squarely on prudent IT cost management. In reviewing their cost options, CTO’s need to assess their business’s strategic context to understand where cuts and investment are needed. More than ever it is crucial that technical teams reevaluate their current solutions and opt for a smart strategy.

When developing analytical software based on the wealth of data that companies have at their disposal, the main goal is often to empower users in their day-to-day jobs, and ensure these users have all the information they need to tackle the challenges they encounter.

When working towards this goal, one of the main obstacles faced by software companies, is the users feeling like the software solution is a “black box.” In other words, it’s difficult to see how the software created the output.

In this article, we will explain where this effect comes from and how it can be avoided or mitigated.

black box effect

Introducing intelligence in software

Introducing intelligence in software can be done in different ways:

  •    Using rules
  •    With the help of algorithms
  •    Using artificial intelligence

Each of these approaches has its own benefits and constraints.


The benefit of using rules is that they can be defined in cooperation with the key users and stakeholders. The results of applying these rules is thus easily traceable and explainable, which makes it easier to get the users’ buy-in.

For example, when working with a telecom operator who wants to optimize their mobile network investments, one rule could state that investment candidates should de facto be prioritized if they are located in areas where the operator has less market share than their main competitor.

Although easy to explain and transparent when applied, the downside of using rules is that they are limited to the existing knowledge, experience and assumptions of the people and teams who define them, potentially leading to biased results.


Algorithms are an extension of applying rules. Compared to rules, algorithms allow for more complex and iterative logics to be applied on the available data.

An example is the introduction of a clustering algorithm in an investment prioritization logic. It would still be based on rules (for example, clusters must have a minimum and maximum coverage and a cluster of candidate investments should have at least a certain Net Present Value (NPV)). The algorithm will also make it possible to do optimizations under constraints (for example, once the clustering is done, the final result should be checked against minimum NPV thresholds, and the algorithm should be rerun if the previous result doesn’t meet the threshold requirements). This means that the solution will aim to optimize all the clusters and achieve the best possible overall outcome given the constraints. This inevitably results in slightly less transparency, as the iterative approach of algorithms makes it harder to trace which exact individual steps were taken to achieve the results presented to the users.

Artificial Intelligence

Finally, introducing artificial intelligence takes the process to the next level. It allows software to fully take advantage of the vast amounts of available data to train itself and learn patterns that can then be applied to new situations. This approach is extremely powerful as it allows the software to do things such as anticipate clients’ and prospects’ appetite for a certain product, or predict churn rates for each segment of an operator’s client base, given one or more network investment decisions.

However, the black box effect is greatest with this approach because the insights gathered through artificial intelligence grow “automatically” without human or manual interaction. In light of this, explaining the exact process and steps followed by the software to arrive at the given conclusions, is near impossible, resulting in potential reticence to use these conclusions.

Mitigating the black box effect

Regardless of the chosen solution, it is crucial for software companies to ensure their solutions are accepted by the users, so they feel confident enough to integrate them into their daily workflow.

To achieve this, the following attention points should be kept in mind:

  •   Always make sure your software introduces intelligence to solve a concrete business issue. Users will be more open to accept the insights you give them if it helps them solve a real problem they encounter.
  •   Choose stakeholders and champions within the user base that can help you support, explain, and defend the output to a broader audience. As these champions are also affected by the problem the software aims to solve, and they benefit from the results, they will be more easily considered as peers by the potential detractors. This makes them more believable when explaining the solution, which in turn increases the confidence in the outcomes of your software.
  •   Communicate.

To avoid surprises, ensure your key stakeholders and champions are involved, or at least aligned with any insights and intelligence you implement and add to the software.

Consider preparing  them for the fact that what they see as output might not be aligned with what they expect, and ensure they understand that this doesn’t necessarily mean the solution is wrong. An example is the appetite for TV bundles and the overall ARPU (Average Revenue Per User) of neighborhoods with low average incomes; although seemingly counter-intuitive, clients in these areas are usually more likely to take additional or more expensive TV bundles, increasing the overall ARPU of the neighborhood. If users haven’t been warned of this potential discrepancy between their “gut feeling” and the reality of the data, they might be reluctant to use the results.

  •   Do not forget to document. Introduce in-app “help” functionalities to have it readily accessible to the users when they need it, or provide on-the-side documentation, which goes in as much detail as necessary for your audience. This means that when dealing with business stakeholders, focus should be kept on the key business background and the context aspects needed to understand the results presented to the users. When developing for more technical profiles, who potentially have prior experience with models, algorithms and artificial intelligence, ensure you tackle the aspects that will allow them to embrace your software and use it in their day-to-day workflow.
  •   Finally, make sure your software is embedded in the standard process. Once users see the predictions and outcomes materialize in their day-to-day work (for example, the predicted market share increase actually happens when the suggested network investments are deployed), they will be more receptive to the solutions’ insights.
  •   Do not force the software’s insights on the users. In some situations, regardless of the precautions taken, users will not accept the outcomes you provide. In that case, the users should be able to override the software, to ensure it’s not bypassed in their workflow, lowering your usage and impact, and putting the overall value of your software at risk. To give an example in the context used above, this can be as simple as allowing the users to remove an automatically prioritized investment candidate from the plan to replace it with another candidate.


Software companies should be mindful of the different levels and methods of adding intelligence to their solutions and consciously choose which ones they want to introduce. This choice should be based on the maturity of the software product, the profile and background of their users, and the prior experience users have with the software company’s solutions, as this will greatly influence users’ trust in the software.

Doing the right preparatory analyses will reveal the potential pitfalls and greatly increase the chances of user acceptance and by extent, the successful introduction of the software, and its intelligence, in the users’ day-to-day workflow.

Enjoyed this infographic? You might also enjoy:

About the author

Tem Bertels is the product manager for Smart Capex Tem BertelsProduct Manager for Smart Capex