Skip to content

Strategic Business Leader · Machine learning, AI and robotics

Risks, Limitations and Implementation Challenges of AI for ACCA SBL

Updated 11 October 2026 · Fact-checked

AI risks are the ways AI can fail or harm: poor data, bias, opaque decisions, cyber attack, high cost, missing skills and staff resistance. To answer SBL questions, name each risk, link it to the scenario, state the business consequence, then recommend practical controls and a phased implementation plan.

Understand Risks, Limitations and Implementation Challenges of AI

Artificial intelligence (AI) lets systems do tasks that normally need human judgement. Machine learning is the main method. It finds patterns in past data and uses them to predict or decide. This means the system is only as good as the data and design behind it.

The main risks fall into groups. Data risks: data that is incomplete, out of date, wrongly recorded or unrepresentative gives unreliable output. Bias: if past data reflects unfair human decisions, the model repeats them at scale. This is algorithmic bias. It can lead to discrimination, legal action and reputational damage.

The black box problem means that even the developers may not be able to explain why a complex model reached a decision. This hurts accountability, makes errors hard to find, and conflicts with the right of customers and regulators to an explanation. Related limits: AI lacks common sense, struggles with situations unlike its training data, and has no ethical judgement of its own.

Other risks are cyber and privacy risk (large data sets attract attackers, and models can be manipulated through poisoned data), cost (software, hardware, data preparation, ongoing maintenance and cloud fees), skills gaps (shortage of data scientists and of managers who can challenge outputs), and people risks (fear of job loss, resistance, loss of morale and knowledge).

Implementation is a change project. Success needs a clear business case linked to strategy, good data governance, a pilot before full rollout, human oversight of important decisions, training, open communication with staff, and monitoring after launch. In SBL, always tie your points to the case, to stakeholders and to governance and ethics.

How to solve Risks, Limitations and Implementation Challenges of AI questions

Use this method for any question on AI risks, limits or implementation. It keeps your answer structured and scenario-based, which earns both technical and professional skills marks.

  1. 1Read the requirement and note the verb (identify, evaluate, advise, recommend) and the audience (board, CEO, audit committee).
  2. 2Scan the scenario for clues: type of data used, decisions the AI makes, who is affected, skills present, budget, culture and regulation.
  3. 3List relevant risks under headings: data quality, bias, transparency, cyber and privacy, cost, skills, people and change.
  4. 4Apply each risk to the case with a specific example and state the consequence for the organisation or a stakeholder.
  5. 5Evaluate: say which risks matter most (likelihood and impact) and whether benefits still justify adoption.
  6. 6Recommend controls and an implementation approach: governance, data checks, bias testing, human oversight, pilot, training, communication.
  7. 7Write in the requested format and tone, with a short conclusion that answers the requirement directly.

Quickest way: The D-B-T-C-C-S-P checklist

When to use it: Use when time is short and you need a fast plan for a 10 to 15 mark requirement.

  1. Write the letters D, B, T, C, C, S, P down the page: Data, Bias, Transparency, Cyber, Cost, Skills, People.
  2. Tick the three or four that the scenario clearly points to.
  3. Write one scenario-linked sentence for each ticked item: cause, then consequence.
  4. Add two or three controls (human oversight, pilot, training, governance).
  5. Finish with a one-line recommendation.

Common mistakes in Risks, Limitations and Implementation Challenges of AI

  • Writing a generic list of AI risks with no link to the case.

    Students memorise headings and write them out without reading the scenario for facts.

    Fix: Quote or use a case detail in every point, such as the data source, the decision affected or the staff involved.

  • Confusing bias with poor data quality.

    Both come from data, so they feel the same.

    Fix: Data quality is about accuracy and completeness. Bias is about unfair or unrepresentative patterns that treat groups differently. Name both separately and give a different control for each.

  • Listing risks but giving no recommendations.

    Students treat the question as a knowledge test.

    Fix: When the verb is advise or recommend, add controls and an implementation plan, and justify them.

  • Presenting AI as only negative.

    The topic title focuses on risks.

    Fix: Give a balanced view. Weigh the risks against the benefits and conclude whether and how to proceed.

  • Ignoring people and change issues.

    Students think AI is only a technical topic.

    Fix: Cover resistance, training, job fears and communication. Link to change models and stakeholder management.

Worked examples

Example 1

A bank uses a machine learning model to approve personal loans. It was trained on ten years of past lending decisions. Applicants from some districts are now rejected far more often, and staff cannot explain individual decisions. Evaluate the risks for the bank. (10 marks)

Show the solution
  1. Bias: past decisions may have reflected human prejudice or district-level patterns. The model has learned and repeated them, so some groups are treated unfairly.
  2. Consequence of bias: regulatory action, discrimination claims, reputational damage and lost lending to creditworthy customers.
  3. Transparency: the model is a black box. Staff cannot explain rejections, so customers cannot be given reasons and complaints are hard to resolve.
  4. Consequence of opacity: weak accountability, difficulty showing regulators that decisions are fair, and errors that go unnoticed.
  5. Data quality: ten-year-old data may not reflect current economic conditions, so predictions may be inaccurate.
  6. Evaluation: bias and opacity are the most serious because the bank's reputation and licence depend on fair lending.
  7. Recommendations: test outputs for bias by district, use explainable models for credit decisions, keep human review of rejections and refresh data regularly.

Answer: The main risks are algorithmic bias and lack of transparency, with ageing data adding to them. Together they create legal, reputational and financial exposure. The bank should test for bias, use explainable methods, keep human oversight and update its data.

Example 2

A manufacturing company plans to introduce AI to predict machine failures across all its plants next quarter. It has a small IT team and no data specialists, and plant managers fear job losses. Advise the board on how to implement the AI successfully. (10 marks)

Show the solution
  1. Business case: link the project to strategy, such as lower downtime and cost. Estimate costs (software, sensors, data preparation, support) against savings.
  2. Challenge the timetable: rolling out to all plants next quarter is high risk given limited skills and untested data. Recommend a pilot at one or two plants first.
  3. Data: check that sensor and maintenance records are accurate and complete. Set data governance with clear owners.
  4. Skills gap: hire or contract data specialists, or use a vendor, and train existing engineers to interpret and challenge the model.
  5. People and change: plant managers fear job losses. Communicate early, explain that AI supports engineers, involve them in the pilot and offer retraining.
  6. Governance and security: set oversight by a senior sponsor, protect sensor data and systems from cyber attack, and keep humans responsible for final maintenance decisions.
  7. Review: measure results against agreed targets after the pilot, learn from it, then extend in phases.

Answer: Implement in phases, starting with a pilot backed by a clear business case. Fix data quality, close skills gaps, manage staff fears through involvement and communication, and apply governance and security. Roll out wider only after the pilot meets its targets.

Exam tips

  • Always tie each risk to a fact in the scenario. Generic lists score poorly on application and professional skills.
  • Match your answer to the verb. Identify needs a short list. Evaluate needs a judgement on which risks matter most. Recommend needs actions.
  • Show commercial awareness: mention cost against benefit, and say whether the organisation should proceed, delay or pilot.
  • Link AI to other SBL areas such as governance, ethics, stakeholders, risk and change management to show wider understanding.
  • Write in the format asked, such as a report or briefing note to the board, and keep a clear conclusion.

Practice questions from Machine learning, AI and robotics

Risks, Limitations and Implementation Challenges of AI in other exams

The same ground in other exams, if you are preparing for more than one or want another angle on it.

Risks, Limitations and Implementation Challenges of AI: frequently asked questions

What is the black box problem in AI?

It is when a model's decision process is too complex to explain, even for its developers. This makes it hard to justify decisions to customers or regulators and to find errors. Using explainable models and human review reduces the problem.

How does algorithmic bias arise?

It arises when training data reflects past unfair decisions or does not represent all groups, or when the model is designed poorly. The model then repeats these patterns at scale. Bias testing, diverse data and human oversight help control it.

How should an organisation implement AI?

Start with a business case linked to strategy, then check data quality and set governance. Run a pilot, train staff, communicate with those affected, and keep human oversight. Monitor results after launch and extend in stages.

How do I score professional skills marks on AI questions?

Analyse the scenario rather than listing theory, show scepticism about claimed benefits, and give commercially sensible advice. Communicate clearly in the format requested, with a firm conclusion.