Introducing the Model Assistant: from ‘what if’ to modelled insight
When markets shift, policies change or geopolitical risks emerge, the first question is usually “what if?” What if oil prices spike again? What if a trade war escalates? What if bond yields keep rising? Answering those questions rigorously means building a macroeconomic scenario, a job that has traditionally required specialist modelling skills most teams don’t have on hand. Our new AskOEAI Model Assistant changes that.
The problem: a gap between the question and the answer
Scenario analysis sits at the heart of good planning and risk management. Risk managers use it to stress-test portfolios, strategy teams to prepare for contingencies, and investment analysts to judge how exposed their views are to a change in the outlook. But turning a “what if” into a credible, quantified scenario has historically been challenging and time consuming.
First, you need to know which levers to pull. Our Global Economic Model (GEM) contains thousands of variables, but a subset of these are the key ones to generate complete economic scenarios with full transmissions. Second, you need to know how big the shock should be and how long it should last, which depends on research and judgement about how similar events have played out. Third, you need to know how to set up the inputs, solve the model and interpret the results.
For many of the people who rely on scenarios, that has made timely analysis hard to come by. Even experienced modellers can spend much of their time identifying the right indicators and building inputs before any analysis begins, which limits how many scenarios they can explore.
“Organisations need to explore how different economic shocks could affect their decisions, particularly when uncertainty is high. The challenge is turning those questions into scenarios that can be modelled and benchmarked quickly and easily. The Model Assistant brings our economic research and modelling expertise into a conversational workflow, helping clients move more quickly from “what if” to a rigorous, quantified assessment of the impact.” — Neil Walker, Head of Macro Modelling, Oxford Economics
What the Model Assistant does
The Model Assistant is the world’s first AI assistant for scenario modelling. It lets you build and run macroeconomic scenarios in the GEM through conversation, with no specialist modelling knowledge required. The workflow has four steps:
- Describe. Set out the scenario you want to test in your own words. Prompts can be broad, such as “Create a severe and sustained energy price scenario”, or specific, setting the size and timing of shocks, linking indicators or drawing on historical episodes.
- Research. The assistant finds the most relevant Oxford Economics research, supplemented by verified external sources, and uses it to propose a set of shocks and their timing, each with a short rationale.
- Review. You preview every input against our baseline and can change, add or remove anything before running the scenario. Once approved, a single click or message runs the scenario through the GEM.
- Analyse and refine. Ask about the indicators you care about and get interactive charts with an explanation of what is driving the results. Adjust your assumptions and re-solve as often as needed. Scenarios are saved in your chat history and online model workspace.
Why it is more than a chatbot on top of a model
General-purpose AI tools can talk fluently about economics. Ask one what an escalation in the Middle East might mean for bond yields and you will get a plausible-sounding answer. But the numbers are not produced by a validated economic model and the assumptions are hard to trace. Our economists and AI experts built the Model Assistant to close that gap, with the work focusing in four key areas：
- Grounded in our research. When you describe a scenario, the assistant reads the most relevant Oxford Economics research and uses it to calibrate the shocks. If our economists have published a downside case for a prolonged conflict or a trade war, that becomes the starting point for the size and timing of the inputs, so each scenario reflects our collective expertise rather than a generic view. We have invested heavily in testing how the assistant selects and reads our research, so it utilises the most pertinent and timely reports and translates it into modelling inputs.
- Trained on how the GEM works. Our modellers have built their knowledge of the GEM into the assistant: which indicators are the right levers, duration and scale for a given scenario. If you try to shock a variable that does not meaningfully feed through the model, it steers you towards a better option.
- Connected to the model itself. Critically the assistant does not generate the results, it sends structured inputs to the GEM and runs a full simulation, using the same model our economists use to produce our own forecasts and scenarios, and relied on by central banks, governments, financial institutions and multinationals worldwide. Results are consistent across economies and benchmarked against our latest baseline, giving you a clear reference point for comparing scenarios.
- Transparent and in your control. Every input comes with a rationale and a preview chart, so you can see exactly what is being assumed and why. You can change any assumption through conversation, and ask how shocks transmit through trade, prices, financial markets. When complete, view the results or open the full scenario in the GEM for deeper analysis.
Table 1: What sets the Model Assistant apart
| General-purpose AI tool | Model Assistant | |
| Assumptions | Drawn from general training data and open-web search | Calibrated from Oxford Economics research and expertise, plus verified external sources |
| Model inputs | Little knowledge of which model variables matter or how to set them | Built-in knowledge of the GEM’s key inputs and transmission channels |
| Results | Narrative or rough estimates, not a solved model | Full GEM simulation, consistent across economies |
| Reference point | None | Benchmarked against our latest baseline forecast |
| Transparency | Reasoning hard to trace or reproduce | Every input shown with a rationale; downloadable, editable and saved |
Source: Oxford Economics
Who it is for
The Model Assistant is designed for anyone who needs scenario analysis but do not have technical model expertise: risk managers, investment analysts, portfolio teams, strategy and planning teams, and consultants. They know the question they want to answer, and the assistant bridges the gap between that idea and model-ready inputs.
The second is existing GEM users, from economists to model specialists, who want to speed up routine scenario setup. For them, the assistant produces a first-draft scenario in minutes, making it easy to test multiple shocks and freeing up time for interpretation and discussion with stakeholders.
The assistant is built for high-level narrative scenarios and for testing many “what ifs” quickly. Complex bespoke exercises, such as regulatory stress tests or detailed geopolitical scenarios that need client-specific calibration, still benefit from tailored work, and our economists and consultants are always available to help.
Next: the Model Assistant in action
In part 2 of this series, we put the Model Assistant to work on a question many of our clients are asking right now: what would a renewed escalation of the war in Iran mean for the global economy and for government bond yields? We walk through each step, from a single prompt to a solved scenario, and show how we refined it through conversation. To read on, find it here.
The Model Assistant is available as an add-on to the Global Economic Model. To see it in action, contact your account manager or request a demo.
