Specify the agent, not just the system. A business analyst in retail banking stays employable by turning a process into an agent's brief: what it may do, at what depth, from which sources, and when it must hand over to a person. Add test cases built from real files and fluent reading of decision records.
Why does the analyst's job change when agents arrive?
Traditional requirements describe screens and rules for software that does what it is told. An agent is different. It reads a case, reaches a verdict and acts, so the question is no longer "what should the system do?" but "what is this agent allowed to decide, and how will we know it decides well?"
That question sits squarely with analysts. Someone has to sit with the card disputes team, the account opening team or the complaints team and write down which steps are routine, which need judgement, and which must never be automated. In a bank, that work also has to stand up when a risk or audit function asks why the agent acted as it did.
What will you be specifying?
A governed agent is certified as a whole, to one depth, and each capability needs its own depth. For each capability you help define:
- Role scope: the job, in plain language. Drafting a response to a card dispute is one capability; issuing a provisional credit is another.
- Depth: 1 read, 2 draft for a person, 3 execute routine work within bounds, 4 move money or take hard-to-undo actions. The agent's role sets a ceiling and evidence sets the actual depth; the lower wins.
- Sources: which facts the agent may rely on, and how strong each is. A figure from the core ledger is sealed; a customer's statement is declared; an emailed screenshot is external.
- The bank's own limits: amounts, customer types or products where the agent must hold for a named person.
- Escalation: who receives a held case and what they need to see.
How do you test an agent like a requirement?
You already write acceptance criteria. For agents, the equivalent is a set of real past cases with known right answers. A certification run replays the agent against them and scores every answer against a floor, gate by gate: grounded, refuses, depth, escalates, scope, adversarial and stable.
Take an analyst we will call Marcus at a mid-sized bank. He is asked to scope an agent for card dispute intake. He pulls two hundred closed disputes, tags which were simple refunds and which turned on a merchant's evidence, and flags twelve where a provisional credit was wrongly issued. Those twelve become the cases the agent must hold, not decide. His requirement is now testable, and the test is the bank's own history.
Which skills matter most?
- Writing an agent brief in plain language, with scope, sources and hand-offs.
- Mapping a process to depth, capability by capability, and defending each choice.
- Building case sets from real files, including the awkward ones.
- Reading decision records to see which bound decided a verdict and whether the process or the agent is at fault.
- Explaining refused against withheld. A refused action breaks the bank's own bounds; a withheld one is blocked at the platform floor for every agent. Stakeholders confuse the two.
- Measuring after launch: hold rates, override reasons and drift, read as signals of a process problem.
Analysts who work close to risk teams will find overlap with the AI skills a compliance analyst needs and with the onboarding judgement covered in AI skills for a KYC analyst.
How do you show it on a CV?
Describe outcomes in agent terms: "scoped a dispute intake agent across four capabilities; built a 200-case test set; set hold rules for provisional credits." The AI agent skills CV section for a business analyst has wording to adapt, and rewriting a business analyst CV for AI operations jobs covers the whole document. Titles vary, so search adverts for duties such as "agent design", "process automation" and "controls" as well as "business analyst". The broader set of proofs employers look for is in AI skills for job seekers.
Where the Academy fits
Analysts usually start with Pryme Intelligence Academy's free Foundations track, then take the Operator track to understand review from the inside. The Builder track is the closest fit for analysis work: designing, training and certifying agents, ending in a capstone agent certified to depth 2 or higher. Each certificate can be checked by number or QR code.
Questions people ask
Is a business analyst the same as an agent builder?
Not always. Some banks split the work, with analysts defining requirements and a separate builder configuring the agent. Analysts who can do both are useful because nothing gets lost in hand-over.
Do I need data science skills?
No. You need to understand what an agent may decide and how to test it on real cases. Comfort with basic data, such as counting outcomes and reading samples, is enough for most of the work.
Which banking processes suit agents first?
Usually high-volume, rule-heavy work with clear right answers, such as dispute intake or document checks. Processes where a mistake moves money or affects a customer's credit are typically held for a person for longer.