Fish Food 656: Using AI for simulation and scenario planning in strategy

A workflow for using AI in scenario planning, the decline in conscientiousness, when data and anecdote conflict, and investing in slow growth

Aug 15, 2025

One of the most eye-opening ways in which AI engines have opened up new possibilities in my work has been through the ability to use them in simulation and scenario planning. It’s enabled me to do things that were previously not an option for me without using expensive systems or tools. As I’ve said before, the best way to think about using AI tools in this context is as a thought partner and to integrate it into a workflow that combines the best from how a human strategist thinks with the efficiency, speed and unprecedented scope that an AI engine can bring.

Again, like the posts that I’ve written on using synthetic personas and research, using AI as a thought partner, and integrating AI into the strategy workflow (and using project spaces), this is not a definitive guide. It’s merely the way that I’ve found to be most useful. And I recognise that there are many different applications for simulation and scenario planning, but what I’ve tried to do here is to set out an approach which can be applied in multiple contexts. For clarity, I’ve laid it out as a workflow including some simple example prompts which can be adapted to your particular situation.

Step 1: Objectives and strategic context

A way to think about this is that scenario planning isn’t just about imagining the future, it’s imagining relevant futures that will help make better choices now. It can be useful to use a project space to run the scenario mapping (this is what I do) but you don’t have to. Either way, front-loading the context and being clear around these three parameters helps the GPT to stay focused:

Define your purpose and objectives: Why are you doing the scenario planning? It sounds obvious but if your objectives lack clarity the GPT can drift and your outputs will be sub-optimal. There are lots of different objectives that using AI engines in this way can fulfil (for example exploring future customer needs, or scenario planning for a market change, or stress‑testing a brand campaign) so be clear about your need and intent. If it helps, you can ask the GPT to frame relevant questions for you. Example prompt: ‘What strategic questions should we aim to answer if our goal is to future-proof our loyalty programme?’
Outline scope and time horizon: You need to decide the horizon (short, medium, long-term but be specific) and ensure it matches the decisions you need to make. You can get it to help you think through the implications of multiple timelines (time-horizon testing), or suggest why certain time horizons might be more or less relevant for your sector. Example prompt: ‘What key events or industry changes should we anticipate over the next 3 years in [industry] that could impact marketing strategy?’
Set success criteria and decision levers: This is about specifying how you will judge the success of strategies across scenarios (e.g. brand sentiment, market share, sales) and which variables or levers you can actually influence (channel mix, pricing strategy, product innovation). If it helps at this point, the GPT can list likely decision levers, or show trade offs. Example prompt: ‘Help me create a table with three columns: Decision Lever, Possible Adjustment, Likely Impact on [specific KPI]’.

Of-course, you can iterate from the first set of ideas, test for relevance (‘Which of these most directly help us answer our strategic question?’) and bring in likely constraints if needed.

2. Environment scanning and mapping drivers

Scenarios are only as good as the raw material that they’re built from. Scanning the environment helps to identify the forces that may shape the future (trends, uncertainties, and shifts that could create threats or opportunities). Mapping the key drivers that arise from this can then organise these into a structure that helps you spot the most impactful and/or uncertain factors to build scenarios around.

Collect insights: gather as much relevant insight material as you can (market reports, customer data or research, trend forecasts). It can really help to use a Project Space, or Google Drive or Sharepoint folder that you can connect to the GPT to do that since it provides a consistent source (more on using Project Spaces here).
Identify likely drivers of change: Then use the GPT to identify the drivers (internal or external) that can significantly influence your strategic environment. You can categorise the GPT’s focus into traditional models like PESTLE (Political, Economic, Social, Technological, Legal, Environmental) or let it classify them into relevant groupings. Example prompt: ‘Based on these inputs, list 10 potential drivers of change in the [industry] sector over the next 5 years. Categorise by PESTLE’. Remember to always validate (GPTs don’t always source accurately).
Ranking by impact and uncertainty: Not all drivers are created equal of course. The ones that are likely to matter most for scenario planning are high-impact and high-uncertainty. So it can help to ask the GPT to do a first-pass at suggesting how drivers might score on an impact vs. uncertainty scale (or your own alternative criteria), based on the data and context that you’ve provided. This can help prioritise and narrow the list to the to the 4 or 5 key drivers that are worth exploring further. You might even go as far as asking it to map potential knock on effects of specific drivers in the form of ‘if X happens, then Y, then Z’, to understand impact better (an impact chain). Example prompt: ‘Given these drivers and our objective to future-proof our marketing strategy, rate each driver from 1–5 for impact and uncertainty, with short justifications.’

3. Scenario definition and framing

Having shortlisted your key drivers, this is where you can use the drivers of change that you have identified to generate coherent plausible or possible futures. The goal here is not to predict the future precisely, but to build a set of alternative futures that are relevant, internally consistent, and different enough to test your strategy against.

Choosing scenario axes and generating scenarios: You can ask the GPT to generate scenarios straight from the shortlist of drivers, but I’ve found it much more useful to combine drivers by selecting the two drivers that you want to examine or asking the GPT to recommend which driver combinations create the richest divergence. You can then ask it to create a classic 2×2 using two critical uncertainties as the axes, creating four distinct quadrants or scenarios. Example prompt: ‘From these drivers, which pairs could form the most divergent and strategically useful 2×2 scenario framework? Explain your reasoning.’
Sketching scenario logic: Having defined a few scenarios, it can then be a good idea to ask for some scenario logic for each quadrant/scenario, or the storyline of cause-and-effect that explains how the world got from today to this possible future. The idea here is that a strong logic helps avoid contradictions and shows plausible ‘from now to then’ sequences. You can then better connect drivers to outcomes. Example prompt: ‘For a ‘High Regulation / Low Trust’ scenario, create a plausible timeline of events from 2025 to 2030 showing how this world emerged’.
Creating a narrative: Once the logics are in place, you can then use the GPT to name each scenario, and turn each into a short, compelling story that makes it tangible. For example, ask it to write a 200–400 word vignette for each scenario, in the voice/tone suited to your audience (analytical, imaginative, brand-specific), and reframe the same scenario for different audiences (CMO, Head of Innovation, Creative Director) if necessary. It can help when doing this to ask the GPT to avoid overused tropes or cliché and push for relevant details that help it to feel richer. Example prompt: ‘Write a 300-word narrative for our ‘High Regulation / Low Trust’ scenario from the perspective of a CMO of a global beverage brand’, or ‘Incorporate two current weak signals into the narrative for Scenario 3: the rise of closed community commerce and the growth of subscription-based media ecosystems’.

4: Strategy testing and simulation

This stage connects the imaginative futures to concrete strategic moves (like product launches, campaign strategies, innovation bets) and explores their resilience across different worlds. This is where you move from describing futures to testing strategy options against them, where can you see how a strategy holds up under stress and where it needs adaptation, and where the role of the GPT switches to simulation partner.

Selecting strategic moves to test: Select (or ask the GPT to recommend) specific actions, investments, or campaigns that it would be good to evaluate, aligned to your strategic objectives (e.g. expanding into a new customer segment, launching a new subscription model, building an in-house creative AI capability). It can sometimes help to ask the GPT to group moves into themes (growth, efficiency, innovation, risk mitigation), or identify opportunities not currently in play that might perform well in multiple scenarios. Example prompt: ‘Categorise these 12 proposed moves into growth, innovation, and risk mitigation buckets.’
Role-playing the world’s response: For each scenario, you then use the GPT to ‘drop’ a strategic move into that world and see how it plays out. Ideas here include consequence mapping (map out first order and second order effects), persona simulation or asking the GPT to act as a stakeholder in that future (a consumer, regulator, competitor) reacting to your move, or identifying weaknesses in the strategy that could mean it will fail. Example prompt: ‘For each of these strategic moves, list the first- and second-order effects in the ‘Decentralised Media / Low Regulation’ scenario.’
Resilience and fit analysis: I’ve found it quite useful to see which strategies are robust across most scenarios, and which are effectively bets that work only if one specific future plays out. To do this you can ask the GPT to score each move across scenarios for feasibility, potential ROI, and risk, to identify moves that score consistently well across all scenarios, and even to suggest adaptations that can modify vulnerable moves so that they work in more worlds. Example prompt: ‘Create a table scoring these 6 moves for feasibility, ROI potential, and risk across all four scenarios.’
Testing combinations and portfolios: Ultimately, the best answer may not be about picking one move, but assembling a portfolio of strategies that balance resilience with bold bets. Here the GPT can recommend combinations of moves that create balance, or identify moves that strengthen each other when implemented together (synergy spotting), or build an action plan or playbook for what to do if a specific scenario plays out. Example prompt: ‘From these 10 moves, propose a portfolio with 3 robust plays and 2 high-risk/high-reward bets, and justify the mix’ or ‘For each scenario, create a 3-step playbook for executing the most suitable moves’.

5: Indicators, early warnings and monitoring

This is about turning scenarios into an ongoing radar for change, and using the GPT to help set up intelligent early-warning systems for teams. A way to think about this is turning scenario planning from a one-off exercise into a living system where you can start tracking which scenario the world seems to be moving toward. The idea is to get early warning on which scenario is becoming more likely, to trigger pre-defined strategic moves, and avoid being blindsided by change.

Defining signposts and leading indicators: The GPT can suggest measurable, observable signs for each scenario, help you to challenge vague indicators and make them more concrete, and suggest potential data sources or platforms to track each indicator. Signposts may be big shifts that tell you a scenario might be emerging, and leading indicators could be smaller, earlier shifts that precede a bigger change. Example prompt: ‘For each of our four scenarios, suggest 5 leading indicators that this future is becoming more likely. Ensure they’re specific and measurable.’
Prioritising indicators: Some signs are high-signal, low-noise (very reliable predictors), others may be just background noise, so ask the AI to rate each indicator for predictive value and ease of monitoring, to suggest ways to separate meaningful changes from random fluctuations (noise filtering), and help to pick 3 - 5 indicators per scenario to track closely. Example prompt: ‘Rate these 15 indicators for predictive value (1–5) and ease of tracking (1–5), and recommend the top 5 to monitor.’
Setting monitoring rhythms and actions: Of course, indicators are only useful if checked consistently and actioned. So use the GPT to suggest checking frequency (depending on volatility), notify specific people or assign ownership, and if you can, define thresholds for when to act. Example prompt: ‘For these indicators, suggest an optimal monitoring cadence and trigger thresholds for action’ or ‘Draft a one-page “Scenario Watch” dashboard layout showing the top 5 indicators and their current status’.

6: Iteration and organisational buy-in

It strikes me that scenario planning works best when the organisation treats it as a continuous capability rather than a one-off event. And AI is now enabling any organisation or strategy team to do this. This means creating a shared language about the future and keeping strategy adaptable.

Reviewing and updating scenarios: Over time, new data, market shifts, and unexpected events will require updates. AI can compare previous assumptions with new inputs and flag where scenarios may need revision, suggest what’s missing based on recent developments, and update or reframe scenario narratives to reflect new realities without losing their original logic. You can even get it to suggest how to re-weight budgets or focus areas based on evolving scenario likelihoods. Example prompt: ‘Here’s our 2024 scenario outline and a summary of market changes since then. Which parts of the scenario logic are now outdated?’
Socialising scenarios in the organisation: People are more likely to use scenarios if they understand them and see relevance to their own work. So you can use AI to reframe scenarios for different teams (creative teams, planners, product managers), suggest storytelling formats to bring scenarios to life, or draft short, digestible summaries for internal comms.
Creating a continuous feedback loop: Creating a rhythm of review and refinement helps turn scenario planning into a habit. AI can recommend how often to refresh scenarios, build simple checklists or scorecards to track scenario relevance, and suggest what events should automatically trigger a scenario refresh. Example prompt: ‘Suggest a quarterly review process for keeping scenarios up to date, including data sources and discussion prompts.’

So there it is. A six stage workflow that starts with a clear purpose (Part 1), builds a robust foundation of drivers (Part 2), turns them into vivid futures (Part 3), tests strategies against them (Part 4), monitors real-world signals (Part 5), and iterates and embeds them organisationally (Section 6).

Scenario planning is genuinely one of those areas where I feel that AI can be like a superpower for me (and for other strategists and consultants). It enables you to do things that were simply not possible before, at least not without significant time and resource. It’s a wonderful example of AI superagency for strategists and how the process of strategy is fundamentally changing in the era of AI.

Rewind and catch up:

Transforming systems and thinking differently

What is AI still not good at?

Separating fads from trends, and the second and third order effects of AI

Using synthetic personas and research to explore ideas

Image source: Mike Baxter, inspired by Joseph Voros

If you do one thing this week…

Literally a few minutes after I sent out last week’s update OpenAI launched ChatGPT 5 into the world (LOL). There’s been plenty of reviews (Ethan Mollick amongst the best). Having played around with it for a week my overwhelming impression is that it’s an evolution rather than a revolution. It’s super quick and, as Tyler Cowen mentions it has an uncanny sense of what you might want to ask next (I found myself typing ‘yes please’ in prompt chains a lot more than I did before). It’s also a lot better at doing stuff.

One of the big changes is that if it’s in Auto mode it selects models for you and so it can sometimes default to a simpler model for the sake of speed. So if you’re really trying to use it as a thought partner, always use the ‘Thinking’ model which is very good at reasoning. It also supposedly has a much lower rate of hallucination, and longer context and better memory (meaning that it can maintain better coherence over extended conversations).

The visual above is taken from the Latent Space’s (more technical) review which talks about how the model is able to do things in parallel much better, and how (especially for things like coding) you need to think about it as prompting an agent rather than a model. But more useful for most of us, Will Francis talked through the key features in a short video (including study and learn which looks amazing for learning new topics, Agent Mode, and connecting it to your calendar and email), and he also did a handy short video on prompting GPT 5.

Links of the week

Some shocking stats about the decline in conscientiousness and extroversion in this well-researched FT (£) article by John Burn-Murdoch. It rather stopped me in my tracks this week.
‘Overall, total organic click volume from Google Search to websites has been relatively stable year-over-year... in contrast to third-party reports that inaccurately suggest dramatic declines’. Google counters the prevailing view that AI Overviews are dramatically impacting search traffic and clicks. I’m left wondering how long this will be the case, and also how the type of links that people are clicking on is changing.
An interesting POV on that WPP paper on the AI-empowered agency from a former Publicis CMO
A good post from Sangeet Paul Chowdray on why the polarised debate around AI misses the point about how it will rewire systems. Pair that with Helen Toner writing about the unresolved debates about the future of AI
‘The best approach is surely to pit hard and soft evidence against each other; to use data to interrogate anecdote and anecdote to interrogate data. In fact this is something like a principle of good judgement. Intuition and abstract reasoning are competing but complementary sources of insight.’ More good writing from Ian Leslie on when data and anecdote conflict (sub reqd for full post).
A bit late to this, but I liked Seth Godin’s 65 ‘notes’ to himself. I particularly liked ‘Invest in slow growth’.

And finally…

A nice idea for parents - Gemini Storybook enables you to create a unique 10-page online storybook using your own files and pictures.

Weeknotes

This week I’ve been overseas with work, running a series of leadership sessions focusing on change and agility, and I’ve also been authoring a course on GenAI in marketing for a long-term work partner organisation. It’s been a busy old time but next week I’m taking a few days off so at this point I’m not certain that there will be an update next Friday. Either way normal service will resume the week after so I hope you all enjoy the remainder of a sunny August in the meantime.

Thanks for subscribing to and reading Only Dead Fish. It means a lot. This newsletter is 100% free to read so if you liked this episode please do like, share and pass it on.

If you’d like more from me my blog is over here and my personal site is here, and do get in touch if you’d like me to give a talk to your team or talk about working together.

My favourite quote captures what I try to do every day, and it’s from renowned Creative Director Paul Arden: ‘Do not covet your ideas. Give away all you know, and more will come back to you’.

And remember - only dead fish go with the flow.