Decision forecasting for domain experts

Model the business. Simulate. Decide.

Foretex lets the expert work directly with a model of their own business's future. No consultants in the middle, no six-month project. The first iteration takes 30 to 40 minutes.

One model, 420 runs Today → +12 months
Most likely future: plan for this Rare surges: use these Breaks and slow declines: prepare for these
The problem

Some decisions can't be made without a forecast

Demand and sales

Demand forecasting, assortment planning

Production and capacity

S&OP, line utilisation, output planning

Inventory and supply chain

Safety stock, lead times, networks

Liquidity and cash flow

Cash gaps, payment calendar

Credit risk and receivables

Probability of default, arrears, provisions

Pricing

Elasticity, revenue management

Macro and stress scenarios

Rates, FX, commodities and their impact on the business

Investments and new products

Payback, launching what has never existed

Software and services have been sold against each of these for decades. The problem is still not solved.

Why it is still unsolved

Forecasting fails on the data, and the alternatives are slow and expensive

Too little data

Annual indicators give you 5 to 10 observations, and over those years everything changed. The numbers aren't even comparable with one another.

Plenty of data, about one thing

Terabytes from the plant describe one part of the process over and over. They say nothing about what happens outside the regimes you've already seen.

No data at all

A new product. A new plant. A scenario that has never happened.

What works today: two poles, and most problems fall in between

6–12monthson practically any serious problem
$0.5–1.5Mper projectfor a typical supply chain optimisation engagement
R&Dno guaranteesyou find out whether it works only after you pay to try. An in-house team changes nothing: the same expensive people, the same iterations.

Missing data doesn't excuse you from making the decision.

How it works

Three stages, two steps in each

The methods and the data science already existed. What was missing was a system that speaks the expert's language. An LLM is the first tool that speaks both.

Stage 1 · Framing

Interview and framing

The expert explains how the business works and what has to be decided. About twenty questions; "don't know" counts as an answer.

back and forth from the start

Data discovery

What exists, in what shape, and how much of it can be trusted. Every column gets a passport; defects are stated.

First iteration: 30–40 minutes
Stage 2 · Model

Model generation and review

The system assembles the model. The expert checks the links and assumptions and corrects them.

in parallel

Calibration against data

Parameters and distributions are fitted to history wherever history exists.

Stage 3 · Decision

Scenario simulation

The expert sees what happens to the business in every scenario.

together

Optimisation

The system finds the best set of decisions within the constraints.

Then a new iteration, by reframing the problem.

Stage 2 in detail · The model

Four kinds of building blocks. Statistics is usually only 10–20% of them.

Known exactly → formula

Revenue, cash flow, profit. Accounting identities, not guesses.

Unpredictable → scenario

Policy rate, FX, commodity prices. Nobody forecasts these; you decide which worlds to test.

History exists → statistics

Demand, deliveries. Fitted to your data where there is data.usually 10–20% of the model

Behaviour → agent simulator

Customers, dealers, borrowers. Thousands of agents acting on rules the expert can read.

What you control → decision

Price, stock, capacity. The levers the optimiser searches over.

Stage 3 in detail · Optimisation

Thousands of decisions interact. The system finds the best allowed set.

max profit

or free cash flow, output, service level: one metric is the objective, the rest become constraints

CAPEX ≤ budgetloss risk ≤ 5%capacity ≤ availablestock ≥ minimum
Expected profit, low → high Breaks a constraint Best allowed combination
Thousands of futures

We don't forecast one future. The result is a distribution.

A scenario is a world, not a forecast. Pick one and see how the cost of funding for the quarter spreads across 1,000 runs, and how often it stays within target.

Median result
0mn
Cost of funding for the quarter
Lower outcome p100
Median p500
Upper outcome p900
Runs within target (≤ 150 mn)0%

Each dot is 5 runs. Dots right of the target line breach the limit. Illustrative figures on hypersynthetic data.

Trust, not faith

Don't trust the forecast. Check the model.

Every number is labelled: computed from the project's tables, taken as an assumption, carried over from a finding, or flagged as an open question.

ParameterValueSource
Inflow rate elasticityinflow_elasticity
Set to 0 until an estimate from inflow and rate-grid history is available.
0 1/pp
range 0 … 0.15
assumption
Monthly inflow volatilityinflow_sigma
Month-to-month scatter of new-money volume; to be recomputed from monthly openings.
0.25 σ (log)assumption
Current rate gridbase_rates
The grid in force before the committee meeting, read from the latest effective rows.
91: 12.8 · 181: 13.2 · 367: 12.4from data
Pass-through of a rate cutfollow_share
Neither in the data nor in the expert's words yet. Goes on the list of questions.
?open question
from data 3 parametersassumption 9open question 9
How Foretex compares

Where Foretex sits among the alternatives

Each option below solves part of the problem. Foretex is built for the expensive, uncertain decisions that fall between them.

BI with AI forecastingPlanning platformsSimulation toolsConsultants or in-house data scienceForetex
Time to first resultMinutesMonths of implementationWeeks6–12 months30–40 minutes
CostLicenceLicence plus implementation partnerLicence plus specialist time$0.5–1.5M per projectStaged: demo, PoC, production
Works with little or no dataNeeds long historyPartlyYes, if a modeller builds itYesYes: scenarios and simulators fill the gaps
Who builds the modelThe tool fits a trendImplementation partnerSpecialist modellerConsultants and data scientistsThe domain expert, with the system
Expert can check the logicBlack boxPartlyOnly modellers can read itThrough reportsEvery link and number is labelled
Scenarios and tail riskA confidence band on a trendWhat-if on the planYesPer projectThousands of runs, probability of breaching each limit
Optimisation under constraintsNoIn specific modulesAdd-onsCustom-builtBuilt in
Your dataIn the vendor's platformIn the vendor's platformOn your machinesShared with the teamDemo on hypersynthetic data. PoC inside your perimeter
Strong fitPartialWeak fit
Industries

Built for decisions with a lot at stake

Banking and finance

Funding cost, deposit pricing, credit risk, liquidity

Use case coming soon

Manufacturing

Capacity, S&OP, output planning, new plants

Use case coming soon

Retail and distribution

Demand, assortment, pricing, safety stock

Use case coming soon

Energy and commodities

Price scenarios, production, investment payback

Use case coming soon

Logistics and supply chain

Lead times, network design, inventory

Use case coming soon

Your industry

If the decision is expensive and the future is uncertain, it fits

Tell us about it
Where it fits

Seven signs your problem belongs here

The more of these that apply, the more a model beats a forecast.

  • The decision is expensive to get wrong
  • Many factors interact, and pulling one lever moves the others
  • The uncertainty is material, not a rounding error
  • History is short, incomplete or from a different regime
  • Expert knowledge matters as much as the data
  • You need to compare levers, not just predict a number
  • The downside and the extreme scenarios matter

If an ordinary forecast solves the problem well, use an ordinary forecast.

Foretex is for the rest: the decisions no boxed product covers and no consulting budget reaches. Bring the problem, not the data. The first demo runs on hypersynthetic data and needs no access to your systems.

Check your problem with us
The payoff

A direct effect on business metrics

In BI the effect is indirect. Here every decision taken on the model moves a specific number, and you see it straight away.

↑

More profit

Better pricing and assortment. More output from the same capacity.

↓

Lower operating costs

Less inventory without losing service levels. Lower capital requirements.

↓

Lower critical risks

Ready for rare scenarios. A smaller catastrophic downside.

Pricing

Start small. Pay as the value is proven.

Pricing follows the three stages. Details are coming soon; contact us for a quote.

Stage 1

Custom demo

Price on request

  • Your business problem
  • Hypersynthetic data
  • No access to your systems
Request a demo
Stage 2

Proof of concept

Price on request

  • Inside your perimeter
  • A real problem, real data if needed
  • Your data as it is: Excel, documents, PostgreSQL, ClickHouse, MS SQL, Snowflake, Databricks
  • Built together with your expert
Discuss a PoC
Stage 3

Production

Price on request

  • A working decision tool
  • For your experts and committees
  • Cloud or on-premise GPU
Contact us

Hypersynthetic data is a generated dataset that matches your real data in structure, scale, dependencies and business context, but holds none of your confidential information.

The next step

Give us one specific business problem.

Book a custom demo

We use your details only to reply about this request.
Received

Thank you. We'll come back within two business days.

We read every problem ourselves. If it is a fit, the next message is a proposal for the custom demo.