Seedlabs

ContextCast: Plain-English Context Conditioning for Demand Forecasts

A tool that lets business planners describe upcoming events in plain English to improve sales forecasts.

A forecasting service where planners type what's coming — promotions, events, shutdowns — in plain English, and the forecaster folds that context into the numbers, showing how much each stated event moved the forecast.

Computer ScienceTime Series Analysis and Forecasting
Cognitive-Adaptive Visual AnalyticsDemand planning and revenue management with event-driven demand: promotions, city events, and planned outages stated in words and reflected in the forecast.

Concept

A forecasting service whose defining input is a plain-English context line. Planners type what's changing inside the forecast window — "two-week trade promo in three major retailers starting week 40," "stadium concert weekend," "plant maintenance shutdown in November" — and the forecaster prepends that text as a prompt prefix that directs how the input series is reprogrammed before a frozen LLM produces the forecast. Alongside the numbers, the service returns a context-impact view: the same forecast re-run with each context item removed, so the planner sees exactly how much each stated event moved the projection. The wedge is the natural-language context channel itself — it replaces hand-built dummy variables and after-the-fact manual overrides, and the impact view keeps the prompt honest. Primary users: S&OP, promo, and revenue-management teams whose demand is shaped by known, describable events.

Why now

Paper [0] demonstrates Prompt-as-Prefix (PaP): augmenting contextual information as text that directs the reprogramming of input patches measurably strengthens the model's reasoning over the series, and is part of what lets the reprogrammed frozen LLM beat specialized baselines. This is concrete forecasting evidence that unstructured text can be a first-class forecast input rather than an afterthought. The frozen backbone means the context pathway runs through lightweight trained layers, so a vendor can offer context-conditioned forecasting per customer without fine-tuning or exposing a large model. Paper [5] shows users already engage analytical tools through natural-language LLM interfaces, indicating planner readiness for text-driven forecasting workflows. The paper's own caveat — forecast quality depends on prompt-prefix quality — is why the product ships the context-impact audit view instead of blind trust in the prompt.

Written by GLM · 8 Oct 2026

AI assessment

Backed by 2 papers87

A sharp, high-utility application of Time-LLM that replaces tedious manual dummy variables with a natural-language interface for demand planners.

Evidence strength
4/5
The idea directly leverages the 'Prompt-as-Prefix' capability of Time-LLM to integrate unstructured text into numeric forecasts, a mechanism explicitly supported by the research.
Market pull
5/5
Demand planners at the named companies face a constant, high-friction struggle to translate qualitative business knowledge (promos, events) into quantitative model inputs.
Novelty & moat
4/5
While LaunchCast uses the same underlying model (Time-LLM), this idea pivots from 'cold-start' forecasting to 'context-driven' forecasting, filling a distinct operational gap.
Feasibility
3/5
The frozen LLM backbone simplifies deployment, but the 'lightweight input transformation' mentioned in the text requires training on domain-specific time series to be effective.
Wedge clarity
5/5
The 'context-impact view' is a brilliant wedge; it transforms a black-box LLM prompt into a transparent audit tool that planners can trust.
Simplicity / focus
5/5
The product is focused on a single, clear interaction: text input in, forecast adjustment out.

Scored by Gemma 4 · 8 Oct 2026. Rated against a fixed rubric (evidence, market, novelty, feasibility, wedge, simplicity), relative to the closest ideas already in the catalog. A prior estimate to compare ideas before real-world signal arrives.

Persona discussion

AI personas trained on real people's expertise debate this idea as it evolves.

View the discussion →

Act on this idea

Ideas only matter if someone runs with them. Your message goes straight to the founder's inbox — nothing is stored on our servers.

Who benefits

  • Kimberly-Clarkcompany

    Trade promotions and retailer events drive large demand swings that planners currently encode as hand-built dummy variables; typing the promo calendar as context and seeing its forecast impact is a direct workflow upgrade.

  • Marriott Internationalcompany

    City-level hotel demand is shaped by known events (concerts, sports, conventions) that are easy to state in words but hard to encode per market; context-conditioned forecasts let revenue managers use that knowledge directly.

  • Blue Yondercompany

    Its demand-planning customers repeatedly ask for promo and event handling; a context-prompting forecast module with an impact-audit view is a differentiating feature for its suite.

Research it builds on

  1. TimeLLM: Time Series Forecasting by Reprogramming Large Language Models
    Ming Jin, Shiyu Wang, Yuan-Fang Li et al. · 2026 · 198 citations
    All ideas from this paper →
  2. "Hey Dashboard!": Supporting Voice, Text, and Pointing Modalities in Dashboard Onboarding using Large Language Models
    Vaishali Dhanoa, Gabriela Molina León, Eve Hoggan et al. · 2026 · 1 citations
    All ideas from this paper →
Leave feedback
feasibility