What AI Business Analysts Do and Why Engineering Teams Need Them Now
A business analyst turns an unclear business need into a specification a delivery team can build, test and measure. The IIBA BABOK Guide defines the practice as enabling change by defining needs and recommending solutions that deliver value to stakeholders. In a software team that means eliciting requirements, mapping the current and target process, writing user stories with acceptance criteria, and checking that the shipped increment solved the original problem.
The requirements gap is still the expensive one. PMI's Pulse of the Profession research, summarised by Volere, found that 37 percent of organisations named inaccurate requirements gathering as the primary cause of project failure in 2014, up from 32 percent the year before.
AI changes what the analyst has to specify. The 2025 Stack Overflow Developer Survey reports that 84 percent of developers use or plan to use AI tools, yet 46 percent distrust the accuracy of AI output and 66 percent say AI solutions are "almost right, but not quite". An AI feature does not behave deterministically. Someone has to define what an acceptable answer is, when the system must hand off to a person, which data the model may see, and how the team will measure the feature after release. The 2025 DORA report found AI adoption at 90 percent, with a positive link to throughput and a negative link to delivery stability.
The role is gaining weight. The IIBA 2025 Global State of Business Analysis report shows 76 percent of respondents see business analysis taking a greater role in strategic decisions, and 74 percent say AI has a positive effect on their careers, up from 63 percent. Scrums.com provides AI-certified business analysts, forward deployed into your systems and managed through the Scrums.com Enterprise AI Platform for Software Engineering. You get a shortlist in 48 hours and delivered work inside three weeks.
Essential Skills to Look For in an AI Business Analyst
A strong analyst combines elicitation craft, modelling discipline and enough technical literacy to talk to engineers as a peer. The AI-era additions sit on top of that base, not instead of it.
Elicitation and facilitation. The analyst runs workshops, interviews and observation sessions, and turns what people say into what they need. IT Jobs Watch data for UK business analyst vacancies in the six months to September 2026 lists Agile in 34.56 percent of postings, user stories in 20.97 percent, stakeholder management in 18.55 percent, requirements gathering in 18.30 percent and workshop facilitation in 15.06 percent. These are the skills the market pays for.
Requirements and process modelling. Expect fluency in user stories with testable acceptance criteria, Gherkin scenarios that QA can automate, use cases, and process models in BPMN 2.0.
Data literacy. Working SQL, comfort with spreadsheets and a BI tool such as Power BI or Tableau, and the ability to profile a dataset before proposing a change that depends on it.
Specifying AI features. The analyst defines intents and non-goals, the input and output contract, confidence thresholds, human-in-the-loop points, prohibited behaviours, and the audit trail.
Evaluating AI use cases. Value, data readiness, feasibility, unit cost per call and regulatory class, scored the same way every time. Familiarity with the NIST AI Risk Management Framework and the risk tiers of the EU AI Act matters in regulated sectors.
Prompt and eval specification. The analyst writes the golden dataset, the scoring rubric and the pass criteria that gate a release, following the pattern in OpenAI's evaluation guide.
Change management. An AI feature that staff do not trust is not adopted. The analyst plans the rollout, the training and the feedback loop, and measures adoption after release. IIBA certifications such as CBAP, CCBA and IIBA-CBDA remain useful signals of method.
Where AI Business Analysts Deliver Measurable ROI
The return on a good analyst is the rework that never happens and the feature that gets adopted.
FinTech and banking. Onboarding, KYC, disputes and regulatory change are process-heavy and rule-dense. The analyst maps the current flow, finds the hand-offs that create drop-off, and specifies the target process with owners and decision points. For AI, the high-value work is document extraction and case summarisation. The analyst defines field-level accuracy targets, the confidence threshold that routes a case to a human, and the evidence a compliance reviewer needs. Measures are time-to-approve, manual-review rate and exceptions per thousand cases.
Insurance. First notice of loss, claims triage and underwriting all depend on rules buried in legacy systems and in the heads of experienced staff. An analyst extracts those rules, decides which to keep, and writes the specification for a triage model that ranks claims for review. Measures are cycle time per claim, leakage on straight-through claims and adjuster hours per case.
SaaS. Product teams lose weeks to stories that engineers interpret differently. The analyst writes acceptance criteria that QA can test and that an AI coding assistant can act on without guessing. The 2025 DORA report notes AI adoption correlates with higher throughput but lower stability; clear, testable requirements are the control that keeps throughput without the instability. For copilots and in-product assistants, the analyst owns the eval set and the escalation rules.
Public sector. Service redesign, case management and procurement carry accessibility, equality and records obligations. The analyst documents the as-is service, writes the target requirements with those obligations traced in, and runs AI use-case assessments that record the risk class and the human oversight controls. Measures are case-handling time, first-contact resolution and audit findings per release.
In every sector the analyst's output is the same: a specification the team can build, test and audit, and a measure that shows whether the change worked.
Business Analyst vs Product Owner vs Systems Analyst vs Data Analyst: Which One Do You Need?
These four titles overlap, and vendors often blur them. The distinction matters because each answers a different question.
Business Analyst: what needs to change, and what does "done" mean? The analyst owns problem definition, process analysis, requirements, acceptance criteria and traceability. They work across business and engineering and do not own the product roadmap. Hire one when requirements are unclear, processes are undocumented, or a regulated release needs evidence. For AI features, the analyst writes the behaviour and evaluation specification.
Product Owner: what should we build next, and why? The product owner owns the backlog priority and the value decision. A product owner without analysis support tends to write thin stories; an analyst without a product owner tends to specify things nobody asked for. In a mature squad the two work as a pair, with the owner deciding and the analyst making the decision buildable.
Systems Analyst: how do the systems have to change to support the requirement? The systems analyst works closer to the technical design: interfaces, data flows between systems, integration constraints, non-functional requirements. Hire one for integration-heavy programmes, platform migrations and legacy modernisation, where the risk is in the plumbing rather than in the business need.
Data Analyst: what does the data say? The data analyst queries, models and visualises data to answer questions. They rarely own requirements or process change. Hire one when the decision needs evidence you do not yet have, or when reporting is the deliverable.
The AI-era test. Ask who writes the eval set for an AI feature. If the answer is "the engineer", the requirement has not been analysed. Scrums.com places business analysts alongside engineers on the same platform, so the specification and the build are managed as one piece of work.
What AI Business Analysts Cost: US, UK and Africa Benchmarks
United States. The US Bureau of Labor Statistics reports a median annual wage of $101,860 for management analysts as of May 2025, with employment projected to grow 10 percent from 2025 to 2035. PayScale puts the average for an IT business analyst at $81,155, with a 10th-to-90th percentile range of $58,000 to $114,000 as of July 2026. The Robert Half 2026 salary guide lists $63,000 at the 25th percentile, $80,250 at the 50th and $95,500 at the 75th.
United Kingdom. IT Jobs Watch reports a median salary of £55,000 for business analyst vacancies in the six months to 11 September 2026, with a range of £39,818 to £80,000 and a London median of £67,750. Contract rates on the same source show a median of £500 a day, from £363 to £668. Robert Half UK lists £36,250, £51,250 and £60,500 at the 25th, 50th and 75th percentiles for 2026.
Africa. PayScale South Africa reports an average of R403,728 a year for an IT business analyst as of July 2026, with a range of R147,000 to R742,000. PayScale Kenya reports an average of KES 480,000 as of June 2026. At current exchange rates a senior analyst in Johannesburg or Nairobi costs a fraction of a London or New York equivalent.
The total cost of a direct hire. Salary is the visible part. Add recruiter fees, benefits, the weeks a role stays open, onboarding time, and the cost of a bad hire found at month four.
The managed alternative. Scrums.com supplies AI-certified analysts from a pool of more than 10,000 pre-vetted engineers and specialists across the US, UK and Africa. You receive a shortlist within 48 hours and the analyst is delivering inside three weeks. Onboarding, management and quality control run through the Scrums.com Enterprise AI Platform for Software Engineering. Start a conversation to get a scoped proposal.
How AI Business Analysts Work Inside a Forward-Deployed, Platform-Managed Team
A Scrums.com analyst is not an offsite document writer. They work inside your tools, your meetings and your codebase context, with delivery managed through the platform.
Discovery sprint first. The first two to three weeks produce an as-is process model, a stakeholder map, a prioritised problem list and a first backlog with acceptance criteria. For AI initiatives the sprint adds a use-case assessment: value, data readiness, feasibility, risk class and unit cost.
Embedded in the squad. The analyst sits in refinement, planning and review with the engineers, the product owner and QA. Stories are refined to the point where an engineer, or an AI coding assistant working under an engineer, can act on them without interpretation. Acceptance criteria are written so QA can automate them.
The AI feature loop. For any feature with a model in it, the analyst runs a fixed loop: behaviour specification, golden dataset and rubric, build, evaluation run, release gate, post-release monitoring. The eval set lives in the repository next to the code and runs on every prompt or model change. The analyst owns the dataset and the pass criteria; engineering owns the pipeline. This is how the team keeps the throughput gain without the stability loss that the 2025 DORA report associates with AI adoption.
Traceability by default. Requirements, stories, tests and evidence are linked in Jira or Azure DevOps from day one. For regulated releases the analyst maintains the trace to the obligation and includes evaluation results and human-review controls for AI components.
Platform management. The Scrums.com Enterprise AI Platform for Software Engineering tracks the analyst's delivery, surfaces blockers, and gives you a single view across analysts, engineers and QA on the engagement.
Change management runs alongside delivery. The analyst plans the rollout, prepares the training and collects adoption data after release.
Evaluating AI Business Analyst Talent: Interview Signals, Take-Home Tasks and Red Flags
Interview signals. Ask the candidate to walk through a requirement they got wrong, how it was discovered and what they changed in their method afterwards. Strong candidates describe a specific ambiguity, the rework it caused and the technique they adopted. Ask them to explain the difference between a user story and an acceptance criterion, and to write one of each on the spot for a simple feature. Ask how they measured whether a change they specified actually worked after release. For AI, ask them to describe an eval set they built: how they chose the cases, how they scored them, and what pass rate gated the release.
Take-home task. Give a two-page messy brief for an AI document-summarisation feature in a claims or onboarding process. Ask for a one-page process model, five user stories with testable acceptance criteria, a list of open questions for stakeholders, and a short evaluation specification: sample cases, rubric, threshold for human review, and the release gate.
Red flags to watch for:
- Acceptance criteria written as restatements of the story rather than testable conditions
- Cannot write or read a basic SQL query to check a claim about the data
- Describes AI features only in terms of the model, never in terms of behaviour, fallback and audit
- Has never measured a change after release
- Cannot explain why a product owner and an analyst are different jobs
Practical questions. How would you specify when an AI assistant must hand off to a human? What would you put in the trace pack for a regulated release that includes a model? How do you handle a stakeholder who wants a feature the data does not support?
Scrums.com screens every analyst on method, data literacy and AI specification skills before they reach a shortlist. Start a conversation to review profiles.