I stop listening when a candidate draws a perfect issue tree and never names the trial that would kill the idea. At PA Consulting, that is not a polish problem. It is the job.
PA sits scientists, designers, engineers, and consultants on the same client problem. The interviewer is not grading whether you sound like a strategy partner. They are grading whether you can hand three different people the same next step: a prototype metric, a certification path, and a decision rule. Treat the case like soft McKinsey and you will look fluent while scoring soft on the sheet.
After this piece you should be able to run a PA-shaped case from clarification through a 90-day trial plan, with numbers a procurement officer would not laugh at.
What PA is selling when they hire you
PA describes itself as the firm that can innovate and transform because it fields strategists, innovators, designers, consultants, digital experts, scientists, engineers, and technologists in one team. Its own About page puts the scale at roughly 4,000 people across 19 locations in the UK, US, Ireland, the Nordics, and the Netherlands, across eight industries from defence and government to health, life sciences, and transport.
That mix matters in the room. A recommendation that only speaks MBA language fails the scientist. A recommendation that only speaks the lab fails the client. Public-sector and regulated work also shows up often: procurement cycles, safety cases, security clearance, and political time, not a private-equity hold period.
On the commercial side, PA reported fee income of £753 million for the year ending December 2024 in its Ingenuity Review announcement. Jacobs took a majority stake under a deal announced in late 2020 that valued PA at £1.825 billion, with continued employee ownership as a stated feature, per PA's own partnership announcement. None of that changes the interview ask: can you move from insight to something buildable?
What the interviewer is actually scoring
I have sat enough cases to know where points leak. At a firm like PA, the scoreboard is closer to this than to a classic profitability tree:
| Dimension | What earns the mark | What loses it |
|---|---|---|
| Problem definition | Success metric a client can measure in 90 days | Global TAM with no kill criteria |
| Structure | Drivers that include tech risk, certification, ops | Generic 3C / 4P without a build path |
| Math | Unit economics tied to trial size | Beautiful market size that never funds a pilot |
| Judgment | Build, partner, license, or stop — with a reason | "Full potential transformation" as the close |
| Communication | One recommendation three stakeholders can use | Jargon that only one tribe understands |
If your close is "enter the market," you are not done. If your close is "run a 40-unit outdoor trial against a pre-registered false-positive rate, then license to a vendor that already holds the certificate," you are speaking PA.
A McKinsey-trained candidate often sizes the global market for crane AI, staffs a transformation office, and never mentions false alarms. A PA interviewer has sat with engineers who know that a 2% lab false-alarm rate can become a 15% shift-killing rate in spray and night. They hire people who keep the scientist in the room without losing the client decision.
The early-careers loop PA actually publishes
Do not invent the process. For early careers, PA states a three-stage path on its Early careers FAQs:
- Application form and task-based online assessments (not timed as a whole, though some tasks need quick keyboard responses).
- A role-dependent second stage: business review, technical assessment, or a prepared presentation.
- A final assessment event and interview. Formats mix virtual and face-to-face. Typical ingredients include one-to-one interviews, group exercises, presentations, and case studies. Some roles add coding or technical interviews.
Your invitation letter is the format. Campus UK paths lean assessment-centre; some US experienced-hire paths look more like a consulting Super Day. Science, engineering, and digital tracks can add a technical deep-dive. That is stated by the firm, not folklore.
Two more official points candidates miss. First, many early-careers roles require security clearance; long overseas residence or travel can affect eligibility, and PA asks for honesty on the form. Second, PA says candidates must not use AI tools (including note-takers) during interview sessions. Using them can affect the outcome. Prep with tools beforehand if you want; do not run them in the room.
Experienced-hire loops are less standardised in public materials. Confirm with recruiting. Pay too.
Worked example: PortSight on 14 quays
Prompt (original). Northsea Ports Authority runs 14 container quays. A university spinout sold them a lab prototype: computer vision that flags unsafe crane boom angles. In controlled video it catches 94% of flagged events with a 2% false-alarm rate. The authority wants a go / partner / kill recommendation in 30 minutes. Capex appetite is £4.2m over two years. Union rules require human override. You are PA.
Clarify. Success is not "AI leadership." Success is a certified, operable system that cuts reportable near-misses without drowning supervisors in false alarms. Horizon: 18 months to a scalable decision. You may recommend a joint venture with a camera OEM.
Structure (innovation + ops, not 3C).
- Performance in weather, night, and spray — not lab video
- False-alarm budget the union and supervisors will tolerate
- Certification and cyber for networked cameras on critical infrastructure
- Integration with existing crane PLCs and radio procedures
- Commercial path: build with PA + OEM, license the model, or stop
Math you can reproduce. Last year the authority logged 56 reportable near-misses across 14 quays. Fully loaded cost per near-miss (investigation, downtime, insurance load) averages £48,000. Implied annual cost: 56 × £48,000 = £2.69m.
Suppose a field system that cuts near-misses by 35% after 12 months saves about 0.35 × £2.69m ≈ £0.94m/year. A £4.2m programme needs a multi-year view or a narrower pilot. Pilot four quays first.
Pilot cost sketch: 4 quays × 6 cameras × £18,000 installed = £432,000 hardware and install. Model hardening, edge boxes, and PA/OEM labour: £520,000. Training and change: £90,000. Pilot total ≈ £1.04m.
Kill criteria before you start: false-alarm rate above 5% of shifts, or missed critical events above 1 in 200 crane-hours in a pre-registered outdoor sample. If the pilot clears both and near-misses fall at least 20% on the four quays, expand. If not, stop or narrow to one crane type.
Licensing check. Suppose an OEM pays a 5% royalty on £2,400 per camera-year of monitoring fees across 84 cameras (14 quays × 6) in year three: 84 × £2,400 × 0.05 ≈ £10,000 that year. Licensing alone is not a venture payday. It may still be the only path that survives certification and union process. In-house productisation with no OEM partner usually fails the same way: you rediscover that ports buy from vendors with field service fleets.
Recommendation. Do not kill on lab performance alone; the value case is real. Do not build a PA-owned product company. Run the four-quay outdoor trial with a camera OEM that already sells into ports, keep human override, and license or co-own the model if the false-alarm threshold holds. Next step is a trial protocol, not a national AI roadmap.
That close is what I am listening for: numbers, a kill rule, and a partner who already lives in the certification world.
Ned's rule. If you cannot write the 90-day trial on one page — sample, metric, environment, kill criteria, decision owner — you are not ready to recommend build, partner, or buy. Strategy without a prototype plan is theatre.
Fit questions that actually land
"Why PA?" answers that stop at "innovation" are wallpaper. Stronger: a story where you built or tested something small, took it to a skeptical stakeholder, and changed the plan when the data said so. Government and defence tracks should sound comfortable with procurement and clearance reality. Science tracks should explain research in one minute to a non-scientist, then name the decision it enables.
In a typical assessment-centre final, I would rather hear one concrete experiment you ran — even a weekend prototype or a lab method you redesigned — than three buzzwords about digital transformation. The interviewer is checking whether you can sit between tribes without translating everything into strategy-speak.
PA's applied-science practice publicly frames work as moving from first principles through proof-of-concept experimentation to market-ready solutions (Applied science). Mirror that sequence in how you talk about yourself: define the science question, show what you tested, then state the commercial or operational decision it unlocked.
Practice this today
Write a one-page trial protocol for PortSight: four quays, night and spray conditions, false-alarm metric, sample-size honesty, and a named decision owner. Then run one structure and one math drill so the protocol is not just prose.
If you want a timed voice rep before the assessment centre, use the live interview (UrbanBrew, about 18 minutes, seven-score debrief). For a five-minute typed first pass with no account, use /start. Cases and drills are open for seven days with no card; then $120 for a recruiting season or $49 a month.
Frequently asked questions
What is the PA Consulting interview process for graduates?
For early careers, PA publishes application and online assessments, a role-dependent second stage, then a final assessment event with interviews, group exercises, presentations, and case studies. Some roles add technical tests. Confirm the exact mix from your invitation.
Are PA Consulting cases more implementation than MBB?
Often yes. A recommendation without a build, partner, or license path is incomplete. Expect more emphasis on feasibility, certification, and trial design than on pure strategy elegance. Format still varies by role and office.
Does PA Consulting use group exercises?
Yes for many early-careers final assessments. PA lists group exercises among typical final-stage ingredients. Listen, structure on a shared board, and do not dominate. That behaviour is still the job.
Can I use ChatGPT in a PA interview?
PA says not during interview sessions, including AI note-takers and meeting assistants. Using them can influence the outcome. Using generative tools to refine thinking before you apply is a different question; the firm still wants authentic answers grounded in your experience.
Is CoachNed affiliated with PA Consulting?
No. CoachNed is independent and unaffiliated with PA Consulting, Jacobs, McKinsey, BCG, Bain, or other firms named for interview-style practice. Confirm process and compensation with PA recruiting.
Sources
- About PA | PA Consulting — headcount, locations, industries, positioning. Checked 2026-09-24.
- Early careers FAQs | PA Consulting — official assessment stages, AI policy, security-clearance notes. Checked 2026-09-24.
- PA Consulting publishes annual Ingenuity Review — £753m fee income for year ending December 2024. Checked 2026-09-24.
- PA Consulting and Jacobs announce agreement on future investment — £1.825bn valuation and employee ownership feature. Checked 2026-09-24.
- Applied science | PA Consulting — first-principles to proof-of-concept framing. Checked 2026-09-24.