Business Analyst (AI Product Delivery)
Badger SA | ARC AI | On-site | George, Western Cape
Shape ambiguity into products engineers can build.
At ARC AI, we build AI agents that operate inside real businesses, handle real interactions, and deliver measurable outcomes in a regulated insurance environment. This is not theory. We ship.
We are looking for a Business Analyst who can take unclear, high-level thinking and turn it into structured, testable requirements for systems whose outputs are not always predictable. That combination demands something most BA roles do not: comfort with uncertainty at the specification level.
About the Role
You will sit at the centre of product delivery, where business intent becomes working AI systems.
Your stakeholders are business owners who know the outcome they want but cannot always articulate what the system should do to achieve it. Your job is to manage that ambiguity upstream: ask the right questions, challenge assumptions, and consistently distinguish between what a stakeholder asked for and what they actually need the agent to do.
That means going beyond surface-level discovery, understanding how processes really work (not just what stakeholders say happens), and defining what good looks like in a world where AI outputs vary.
What a Typical Day Looks Like
- Map real business processes and design improved, AI-enabled workflows
- Own requirements end-to-end, from discovery through to delivery and acceptance
- Write Product Requirement Documents that specify inputs, outputs, acceptable behaviour ranges, and the human's role in the loop, not just features
- Translate executive vision into structured, build-ready specifications
- Use data to validate assumptions and challenge what you are told
- Define edge cases, failure modes, and escalation paths as first-class requirements, not afterthoughts
- Run structured discovery sessions and facilitate workshops to resolve conflicting stakeholder priorities
This is not a documentation role. It is a thinking role.
About You
You are comfortable operating without a script. You take a vague brief, structure it, and produce something engineers can build from. You actually prefer working this way.
You will need to
- Have 3+ years of experience as a Business or Systems Analyst, owning delivery from discovery to implementation
- Build structured process maps (BPMN, UML, or similar) without guidance
- Have worked in Agile environments: sprint participation, stand-ups, backlog refinement, and retrospectives
- Write user stories, functional specs, or PRDs you would be prepared to defend at interview
- Query and interpret data (SQL, Excel, or BI tools) to validate requirements and challenge assumptions
- Prioritise work using structured frameworks (MoSCoW or equivalent) and maintain traceability across the requirement lifecycle
- Run workshops, challenge stakeholders, and navigate conflicting priorities within a department
You will stand out if you
- Have written PRDs or functional specifications for products where the output was non-deterministic or required human review: customer-facing automation, recommendation features, or decision-support tools all count
- Understand how to design for uncertainty, including specifying what happens when the AI gets it wrong, not only when it gets it right
- Have experience in regulated environments: insurance, finance, or healthcare
Most importantly: you think in systems, not tasks. And you treat failure modes as requirements, not edge cases to figure out later.
Why Join Us
- Work on real AI products in production, not proofs of concept
- High autonomy, high ownership: you will shape how things get built
- Collaborate directly with engineers and business leaders across Sales, Claims, Finance, and Services
- Be part of a small, focused team that prioritises delivery over process theatre
- Solve complex, meaningful problems in an environment that moves fast and takes AI seriously
How We Assess You
The interview process includes a practical task with two components.
First, you will be given a vague executive brief and asked to produce a structured discovery plan: the questions you would ask, the workshops you would run, and the first-draft requirement categories you would define.
Second, you will review a sample AI product specification and identify what is missing, particularly around failure modes, constraint specifications, and requirements that describe features rather than behaviour.
There is no single correct answer to either component. We are assessing clarity of thought, comfort with ambiguity, and how you think about risk.
Ready to Apply?
If you are comfortable owning ambiguity, thinking beyond the happy path, and writing requirements that hold up when the AI gets it wrong, we would like to hear from you.
