When the interviewer asks, “What do you think?”, answer the question. A list of considerations is useful preparation for a judgment, but it is not yet a judgment.
Say what you would do now, give the strongest reason and name the condition that could change the choice. That can be a decision to expand, narrow the scope, stop or run a particular test. The action should be clear enough that a client could respond to it.
You can do this confidently without pretending the whole case is settled.
Choose a position at the right scope
Consider a fictional AI assistant trial. The client wants to reduce support costs while keeping at least 95% of completed responses acceptable under its review standard.
The case supplies two task groups:
| Measure | Routine tasks | Complex tasks |
|---|---|---|
| Monthly task volume | 6,000 | 4,000 |
| Current cost per task | £4 | £8 |
| AI-assisted cost per task, including review and rework | £2 | £12 |
| Acceptance rate | 97% | 90% |
There is also an £8,000 monthly platform charge for the proposed routine-task deployment. Assume these results represent the workload the client plans to serve.
A clear position is:
“I would deploy the assistant for routine tasks and keep complex tasks on the current process. Routine work clears the quality requirement and reduces cost; complex work fails both tests.”
This is a complete answer to “What do you think so far?” You can then explain the economics and the next check.
Put the best evidence immediately behind the position
Routine work saves £4 − £2 = £2 per task. At 6,000 tasks, that is £12,000 monthly savings before the platform charge, or £4,000 after it.
Complex work costs £4 more per task and falls five percentage points below the acceptance requirement. There is no reason to include it merely to make the rollout larger.
Your explanation can therefore be short:
“The routine segment saves £4,000 a month after the platform charge and exceeds the acceptance requirement by two points. I would preserve the existing route for complex work.”
The numbers support the choice. You do not need a long preamble about AI being transformative, promising or potentially disruptive.
Make the reversal condition useful
“We need to monitor performance” is too broad. Identify the condition that would change the decision.
At £2 savings per routine task and an £8,000 fixed monthly charge, the deployment needs:
£8,000 ÷ £2 = 4,000 routine tasks per month to break even.
If only 3,000 tasks use the system, savings fall to £6,000 before the charge and the deployment loses £2,000 a month on this basis. The 6,000-task opportunity therefore needs enough actual adoption, not just enough eligible work.
A useful condition is:
“I would proceed if we can route more than 4,000 routine tasks a month through the system while maintaining the acceptance standard. Below that volume, the fixed charge consumes the saving.”
That gives the implementation team something specific to validate.
Quick Case Math Drill
A dental group is deciding whether to lease self-service check-in kiosks for its clinics. You want a precise condition for proceeding, not a vague promise to monitor usage.
- Saving per kiosk check-in
- £1.50 of reception staff time
- Kiosk lease cost
- £3,600 per month
- Monthly patient visits
- 6,000 across the clinics
What minimum share of visits must use the kiosks for the lease to break even?
Distinguish a decision from a hypothesis
A hypothesis explains what may be causing an outcome. A point of view says what you currently judge or recommend.
For example:
| Statement | Its job |
|---|---|
| “Complex-task review requirements may explain the higher cost.” | Working explanation to test |
| “Keep complex tasks on the current process for now.” | Decision supported by current quality and cost results |
| “Measure review time by task type.” | Next analysis to sharpen the explanation |
You can have a clear decision before fully explaining every cause. The client does not need to understand exactly why complex tasks cost more to avoid expanding a workflow that currently performs worse on both stated criteria.
When the cause matters for the next improvement, name the competing explanations and request evidence that distinguishes them.
Give a real answer when evidence is incomplete
Sometimes the interviewer asks for your view before you have cost data. Then choose an action appropriate to the information available:
“I would hold the full rollout and calculate cost after review first. The quality result supports routine work, but we have not yet established the saving.”
That is still a position. It identifies the commitment you would defer and the exact work that would unlock it.
Compare that with “There are pros and cons, and I would need more information.” The latter leaves the interviewer to decide what matters and what happens next.
Use your current position to organise the next analysis, then revise it as the case develops. An early answer is useful when it helps you choose what to test.
Change the affected part when new information arrives
Suppose the interviewer reveals that a required data-access permission is unavailable for routine work. That changes feasibility, so pause deployment even though its measured economics remain attractive.
If the new fact concerns only complex tasks, keep the routine-task recommendation unless there is a reason it is affected too. Updating well means tracing the new information to the part of the decision it changes.
You can say:
“The routine economics still work, but the access restriction blocks implementation. I would resolve that permission before committing to the platform charge.”
That sounds more considered than either defending the old answer unchanged or discarding every conclusion because one fact moved.
Quick Business Judgment Drill
You recommended a furniture retailer launch next-day delivery next month. City zones would use its own vans; rural zones would use a contracted courier.
- Measured economics
- Profitable in both zones at quoted rates
- New information
- The courier cannot start rural deliveries until April
How should you update your recommendation after the courier news?
Practise giving the answer in the first sentence
Take a case midway through and answer “What would you do now?” Begin with a verb: launch, retain, pause, renegotiate, investigate or stop. Then support it with the strongest evidence.
Listen back for whether the recommendation appears early enough to follow. Remove introductory phrases that delay it. If your position includes five conditions, choose the one or two that actually control the decision.
Try Cybersecurity Product Expansion. Give a position, a proof point and a specific reason you would change your mind. The pushback guide covers how to respond when the interviewer challenges that view.
Point of view drill
Would your argument persuade an MBB interviewer?
Cybersecurity Product Expansion
Assess a cybersecurity expansion with Casey, Case Room’s AI interviewer. Give a provisional view, its proof point and the condition that would reverse it.
Frequently asked questions
Forming a clear judgement
When the interviewer asks ‘What do you think?’, is listing considerations enough?
A list of considerations is not enough. It prepares a judgement but is not one. Say what you would do now, give the strongest reason and name the condition that could change your choice. The action might be to expand, narrow the scope, stop or run a particular test. It should be clear enough for a client to respond to.
Should I roll the AI assistant out to all tasks to maximise impact?
Choose the scope the evidence supports instead. In the example, routine tasks save money. Complex tasks cost £4 more per task and fall five percentage points below the 95% acceptance requirement. Including complex work merely to make the rollout larger would add cost and breach the quality standard.
What is the difference between a hypothesis and a point of view?
A hypothesis explains what may be causing an outcome. A point of view says what you currently judge or recommend. ‘Complex-task review requirements may explain the higher cost’ is something to test. ‘Keep complex tasks on the current process for now’ is a decision. You can hold both at once.
Supporting and qualifying the view
What evidence should come straight after my recommendation?
Lead with your single strongest piece of evidence, stated briefly. For routine tasks, the saving is £2 per task across 6,000 tasks. That comes to £12,000 a month before the platform charge and £4,000 after it. This one calculation justifies the routine rollout, so state it before working through any other measure.
Why isn't ‘we need to monitor performance’ a good enough caveat?
It does not say what would change the decision. Name the specific condition instead. With a £2 saving per routine task and an £8,000 fixed monthly charge, the deployment needs 4,000 routine tasks a month to break even. That threshold tells the client exactly what to watch.
What happens to the recommendation if fewer tasks use the system?
The deployment can start losing money. If only 3,000 routine tasks go through the assistant, savings fall to £6,000 before the charge, so it loses £2,000 a month. Usage below the break-even volume is therefore the condition that would justify reconsidering the routine rollout.
What should I say if I'm asked for a view before I have cost data?
Give an action that fits the evidence you have. For example, say you would hold the full rollout and first calculate cost after review, because quality supports routine work but the saving is not yet established. That is still a position. It names the commitment you would defer and the work that would resolve it.
Updating and delivering the answer
When the interviewer adds new information, should I rebuild my recommendation?
Change only the part the new information affects. Trace the new fact to the element of the decision it changes. If it concerns only complex tasks, keep the routine-task recommendation unless there is a reason that is affected too. Rebuilding everything suggests you cannot tell which evidence supports which part of your view.
Should good economics outweigh a feasibility problem?
Feasibility comes first. If a required data-access permission is unavailable for routine work, pause that deployment, even though its measured economics remain attractive. Savings only matter if the solution can actually run. Treat feasibility as a separate condition, and say clearly which part of the plan the new constraint stops.
How can I practise putting my answer first?
Stop midway through a practice case and answer ‘What would you do now?’ starting with a verb such as launch, retain, pause, renegotiate, investigate or stop. Follow it with your strongest evidence. Listen back to check the recommendation arrives early enough to follow, and cut introductory phrases that delay it.
My position depends on several conditions. Should I mention them all?
Usually not. If your position includes five conditions, choose the one or two that actually control the decision. In the AI assistant example, routine-task volume against the break-even level is the condition that decides whether the rollout pays. Naming only the controlling conditions keeps the answer easy to follow.