Policy Advisor Interview Questions

I start by identifying the one or two decisions the minister actually needs to make, and I work backwards from there. Every piece of analysis gets filtered through that question: does this change what they should decide, or is it background context? I write the recommendation first, in plain language, then build the supporting evidence underneath it. I also think hard about what the minister cares about politically, not just technically: what will they be asked by journalists, what will opposition parties say, what are the implementation risks? A recommendation that ignores those dimensions is not actually useful. I try to keep the main brief to one page and put the technical detail in annexes, so the decision-maker can engage at whatever level of depth they have time for on that day.

I present the evidence clearly and honestly, and I make sure the decision-maker understands what the trade-offs are. It is not my job to make the political decision for them, but it is my job to make sure they are making it with full information. In practice that means being explicit: "The evidence suggests X, but I understand the political priority here is Y, so the options are A, B, or C, with these consequences." I keep a clear record of the advice I gave and the evidence behind it. If I am asked to misrepresent evidence in official documents, that is a different matter and I would escalate. But operating in an environment where evidence and politics are in tension is just the reality of policy work, and pretending otherwise in an interview would suggest I have not actually done the job.

I would start by identifying the key experts, both inside government and externally, and spending the first few days listening rather than analysing. I want to understand the landscape of views before I form a position of my own. At the same time I would map the policy history: what has been tried before, what worked, what did not, and why. I would look at what comparable countries have done and whether the evidence transfers to this context. Once I have a grounding, I build a simple framework for the decision: what are the objectives, what are the options, what are the trade-offs? I am careful to distinguish what the evidence tells us from what is genuinely uncertain, because that distinction matters enormously when briefing someone who will be accountable for the decision.

I work with a clear triage system. At the start of each week I map out the deadlines, the political moment for each piece of work, and where I can add the most value versus where the work will happen without me. Some things are urgent but not genuinely important, and some are important but have a longer runway. I have learned to be honest with stakeholders early when timelines are at risk rather than trying to absorb pressure silently and then missing a deadline. I also build in time to think, not just produce: policy work done purely in reactive mode tends to be lower quality. When workloads genuinely exceed capacity, I escalate clearly with a proposed solution, not just a problem. I find that framing capacity issues as a resourcing question rather than a personal failure makes those conversations easier.

I was working on a review of a long-standing grant scheme that most people in the team assumed was performing well. The evaluation data was thin but the programme had strong internal advocates. I dug into the original objectives of the scheme and compared them with what the data actually showed about outcomes. The gap was significant: the scheme was reaching the wrong population and most of the evidence of impact was anecdotal. Rather than presenting this as a critique of the team, I framed it as a question: given these objectives, what would we need to see in the data to be confident this is working? Once the conversation shifted to that frame, it became easier for people to acknowledge the weaknesses honestly. The eventual recommendation was to restructure the scheme significantly. The key was making it about the evidence, not about who had supported the scheme previously.

We had 48 hours to produce a briefing on a policy issue that had suddenly become politically live. My first instinct was to try to research everything, but I quickly recognised that was not possible in the time available. I made an explicit decision about what the minister actually needed to know to make the decision at hand, which narrowed the scope considerably. I focused on the three strongest sources of evidence, flagged the uncertainties clearly, and built in an hour at the end to have a colleague with fresh eyes review the draft. The briefing went in on time and was used in a meeting that afternoon. Looking back, the time pressure actually produced a sharper document than it would have been if I had had a week: I could not hide behind thoroughness, so I was forced to make clear editorial choices about what mattered.

I had conducted an internal review that identified significant gaps in the implementation of a programme the department had publicly committed to. The findings were uncomfortable because they reflected poorly on decisions made by senior colleagues. I spoke to my line manager first to make sure the findings were accurate and that I had not missed context that would change the picture. I then prepared a briefing that presented the evidence clearly without editorialising, and proposed three concrete options for how the department could respond. I made sure the briefing reached the right senior official directly rather than being softened through layers of sign-off. The response was constructive: the gaps were acknowledged and an action plan was agreed. The lesson for me was that senior people generally prefer an honest account of a problem over a sanitised one, as long as it comes with proposed solutions.

I apply a hierarchy that starts with experimental or quasi-experimental evidence, then observational studies, then qualitative research, then expert opinion. But the hierarchy is not mechanical: a well-designed qualitative study in a closely analogous context may be more useful than a poorly specified randomised trial in a very different setting. I always ask: does this evidence actually answer the question I am trying to answer, or does it answer a related but different question? I also look at the direction of evidence in aggregate rather than cherry-picking single studies, and I check whether results replicate across different populations and contexts. Where evidence is weak or absent, I say so explicitly rather than papering over uncertainty. A recommendation built on overstated evidence is worse than an honest acknowledgement that we are operating with limited information.

I start by defining the counterfactual clearly, because a CBA is only meaningful relative to what would have happened otherwise. I then map all the relevant costs and benefits, including those that are harder to monetise, and I am explicit about which ones I am including and which I am excluding and why. For interventions with long time horizons, I am careful about discount rates because small changes in the discount rate can reverse the sign of the analysis. I run sensitivity analysis on the key assumptions, particularly where the evidence base is thin, and I present a range rather than a single figure. I also think about distributional effects separately from aggregate efficiency: an intervention that is positive in net present value terms can still be regressive, and that matters politically and ethically. The final CBA should tell decision-makers not just whether the intervention is worth it on average, but who gains and who bears the costs.

The most common failure modes I have seen are: the problem is not what the policy assumes it is, delivery capacity is overestimated, incentive structures create unintended behaviours, and political context shifts before the policy has time to work. I try to address these at design stage by testing the logic model explicitly: what is the theory of change, what does it assume about behaviour, and what would falsify it? I also do a pre-mortem: if this policy has failed in three years, what is the most likely reason? That exercise tends to surface implementation risks that optimism bias otherwise hides. For delivery capacity I push for realistic assessments from the people doing the implementation, not just from the policy team. And I build in review points that are specified in advance, so evaluation is not optional when the political mood changes.