← → arrow keys
$1.25M Pre-Seed
Forecast next week.
Explain why it moved.
Decide what to do.
WeatherVane is a weekly demand decision system for retail brands. It helps operators forecast demand, understand what caused the move, and decide whether to push, pause, stock up, or wait.
SOUNDBITE
Train on pooled retail demand across brands, not just one brand’s history.
+16%
held-out accuracy improvement on M5 from pooled retail and context data
thagorus.com
March 2026 · Confidential
Nathaniel Schmiedehaus
nate@thagorus.com
nate@thagorus.com
Company Purpose
WeatherVane helps retail brands make better weekly demand decisions.
It gives one operator-facing workflow for forecasting next week’s demand, explaining what caused the move, and deciding whether to push, pause, stock up, or wait.
WHAT IT DOES
Forecast next week
Weekly SKU-level demand with confidence ranges before inventory and media decisions are locked.
WHAT IT EXPLAINS
Show the cause
Separate weather, promotion, competition, and baseline so teams stop misreading demand shifts.
WHAT IT CHANGES
Tell the operator what to do
Know when to push, pause, wait, stock up, or avoid wasting budget into non-responsive demand.
WHO BUYS FIRST
Weather-sensitive brands
Where timing mistakes are frequent, expensive, and obvious to the buyer.
Operator owner
Usually the marketing, demand, or inventory lead making the weekly call.
Weekly cadence
Not a quarterly planning suite. A decision workflow teams feel every week.
UNIQUE INSIGHT
No single brand has enough history to separate demand drivers cleanly on its own.
Pooled learning across brands is what turns forecasting into attribution and attribution into action. That is the whole company.
The Problem
450,000 brands sitting on demand data.
Each one guessing alone.
None of them can see the full picture.
NorSari
Blanket sales spike 40% in a cold snap. They think it's their Instagram ad. They double the ad budget. Weather changes. Sales drop. They never learn why.
Homesick Candles
Demand surges every October. Is it the campaign? The gifting season? The first frost? They can't tell. So they overstock, overspend, and repeat. Every year.
×
450,000 BRANDS
Every other brand
Same story, different product. Each sitting on years of demand data. But no brand alone has enough data to see the full picture. Together, they would.
The recurring failure is not bad effort. It is bad causal visibility. Teams cannot tell whether demand changed because of weather, promotion, competition, or baseline demand.
stock wrong
spend wrong
learn nothing
learn nothing
The Solution
One product. One user. One weekly job to be done.
WeatherVane is not a planning suite and not a generic AI dashboard. It is a weekly decision workflow for the operator who has to decide what demand is going to do next, why it changed, and what action to take.
1. FORECAST
Know next week
Weekly SKU-level demand forecasts with uncertainty before inventory and media decisions are locked in.
This is the planning layer.
2. EXPLAIN
Know why
Separate weather, promotion, competition, and baseline so teams stop misreading demand moves.
This is the causal layer.
3. ACT
Know what to do
Tell the operator when to push spend, pause, wait for weather, stock up, or avoid forcing demand that is not there.
This is the decision layer.
EXAMPLE OUTPUT
“Pause ads this week. Demand is being pulled by weather, not paid media. Resume after the cold front.” That is much closer to the buying moment than “another analytics platform.”
Progress
The proof is no longer theoretical.
We have benchmark lift, partner data,
and a weekly operating system.
This deck does not depend on a future research miracle. It depends on three things that already exist: pooled-learning benchmark proof, real brand history from live design partners, and a product that translates prediction into a weekly operating decision.
WHAT IS TRUE TODAY
Progress: benchmark proof, two design partners, working product surface, and a specific 18-month ask.
ONE SENTENCE
Training across brands improves prediction quality, and that extra signal can be turned into clearer weekly decisions for each brand.
EVIDENCE LEDGER
+16%
Held-out accuracy improvement on M5 from pooled retail and context data.
2
Design partners with long-run demand history: NorSari and Homesick.
8+ yrs
Historical brand demand data already informing the system.
Live
Forecast, attribution, and timing outputs already expressed as product behavior.
Benchmark context: M5 is the largest public retail forecasting benchmark, with roughly 100,000 Walmart time series. The measured lift is about 4x what is typically treated as meaningful in that literature.
WHY INVESTORS SHOULD BELIEVE THIS
Not a single experiment: the result is benchmarked, paired with real partner data, and attached to a product workflow.
Not just forecasting: the product value is the decision layer built on top of better prediction.
Not “AI for retail” vapor: this has a narrow buyer, a weekly use case, and a measurable proof claim.
WHAT THE PROOF MEANS IN MONEY
Better inventory and media timing.
For a $10M brand, a cleaner weekly read on demand can mean materially less overbuying, fewer wasted ad pushes, and better capital allocation into weather- or season-driven demand windows.
This is why the first buyer is an operator, not a researcher.
THIS HAS BEEN DONE BEFORE:
IMS Health → IQVIA Pooled pharmacy data across rivals. Now worth $50B.
Renaissance Technologies Pooled diverse data (weather, satellite, economic). 66% avg annual returns.
Bloomberg Aggregated financial data no firm would share. $70B private company.
Verisk Pooled 32B insurance records across competitors. $35B market cap.
Nielsen Built retail measurement by pooling scanner data across brands.
Why Now
This company makes sense now because three conditions finally line up.
Five years ago this would have looked early. Today the modeling approach is credible, the data plumbing is real, and the buyer pain is sharper than ever.
1. MODELING CAUGHT UP
Multi-source forecasting and transfer learning are now established enough that pooled demand learning no longer sounds like science fiction.
2. DATA IS REACHABLE
Commerce, weather, media, and macro signals can now be assembled into one usable weekly operating surface without custom enterprise projects.
3. BUYER PAIN GOT WORSE
Operators still need to make weekly demand calls, but post-iOS attribution and single-brand tools still do not answer the causal question cleanly.
THEN
Too early.
Cross-brand forecasting sounded speculative
Data integration was bespoke and painful
The pitch would have been mostly theory
NOW
Commercially plausible.
Benchmark proof exists
Signals are ingestible
The buyer already feels the pain weekly
WHY THIS MATTERS TO INVESTORS
“Why now” should make the company feel timely, not trendy. The timing case here is specific: model capability, data access, and buyer pain finally overlap.
2026
the window feels open now
Weekly
the pain shows up every week
Competition
Why this is not already owned by incumbents, APIs, or foundation models.
The honest investor question is why a large incumbent or model company does not automatically win here. The answer is that they solve adjacent problems, not this exact one.
Unique insight: the valuable asset is not “an AI model for forecasting.” It is a commercially assembled cross-brand retail corpus attached to a weekly operator workflow.
1
Legacy planning tools
Good at internal workflow and planning depth. Not built on pooled cross-brand learning or day-one attribution value.
2
Forecast APIs and foundation models
Good model ingredients. Not a complete weekly operating product, and they do not automatically own the retail training corpus.
3
WeatherVane
A weekly operator workflow built on pooled demand data, where attribution and timing are the product rather than an add-on.
Where WeatherVane Wins
Others are good at ingredients
Legacy systems: planning workflow depth.
Foundation models: generic forecasting horsepower.
Data vendors: narrow but useful inputs.
WeatherVane is built for the combined job
Cross-brand learningCore
Attribution + timingCore
Weekly operator workflowCore
Why OpenAI does not automatically win
General model companies can supply model capability. They do not automatically own the partner relationships, the weekly operator workflow, or the pooled retail corpus.
The Market
Start with one buyer, one painful weekly decision, and one credible path to paid value.
The market story should read like a seed company, not a spreadsheet. Beachhead customers are weather-sensitive retail brands whose operators already feel the cost of bad demand calls in inventory and media every single week.
INITIAL BUYER
Marketing / demand / inventory lead
The person already making the weekly call on spend, stock, and timing with incomplete causal visibility.
BEACHHEAD SEGMENT
Weather-sensitive mid-market retail
$5M-$500M revenue brands where timing mistakes are expensive, recurring, and easy to feel in margin.
PAID WEDGE
Weekly operating workflow
Forecast + explanation + timing recommendation sold as a recurring operating product, not an annual planning suite.
BOTTOM-UP LOGIC
$100K ACV target: justified if the product prevents even a modest amount of wasted inventory or mistimed media.
First 10 paid brands: enough to prove budget ownership, repeat usage, and a compounding corpus.
Then expand: adjacent categories, more operators inside each brand, and later data/API monetization.
GO-TO-MARKET
1. Sign design partners in categories where weather and seasonality visibly move demand.
2. Turn weekly usage into a repeatable operator workflow with proofs the buyer can point to.
3. Convert the strongest users to paid software accounts before selling any broader vision.
MARKET READ
$3.8B beachhead SAM.
Large enough to matter, small enough to sound believable, and directly connected to a clear first buyer. TAM still exists, but it should not be doing the selling here.
Business Model
The first sale is simple.
The long-term business gets larger as the corpus compounds.
Start with a focused software product for weather-sensitive retail operators. Monetize the weekly workflow first. Let the data asset deepen the moat and open larger monetization later.
PHASE 1
Free design partners
Brands exchange data and feedback for research access and weekly decision support.
Purpose: prove workflow value and build the starting corpus.
PHASE 2
Software subscription
Annual subscription for forecast, attribution, and timing. Sell into a weekly operating problem, not a research budget.
Initial target: 3-10 paid accounts at meaningful ACV.
PHASE 3
Data + API access
Once the corpus is large enough, sell differentiated demand intelligence into broader systems and workflows.
This is upside, not the seed-stage dependency.
WHY THE MODEL IS CREDIBLE
One model first: software subscription tied to a weekly demand workflow.
No dependency on platform dreams: the company can become real before any later data/API business exists.
Pricing logic: charge materially less than the budget wasted by bad inventory and media timing.
18-MONTH ECONOMIC STORY
$0
Intentional year-one revenue while partner value and corpus quality are established.
3-10
Paid accounts as the first commercial proof point after live weekly usage.
Investor read: this is not a business model slide about optionality. It is a business model slide about discipline. One buyer, one workflow, one clear path to paid value.
The Team
Nathaniel Schmiedehaus
Founder & CEO, Thagorus PBC
Founder-market fit is the strongest part of this company.
The obvious risk is that it is still a lean team.
BUILT & EXITED
Homesick Candles
Zero to 8-figure revenue. Sold to WIN Brands Group. Gives WeatherVane a real demand-history proving ground, not just a theory deck.
OWNER & OPERATOR
NorSari
Owns NorSari and runs its marketing. Design partner since 2017. 44% CAGR over 8 years. This is the closest thing to customer-zero.
BUILT THE PROOF
+16% accuracy, 4x M5 threshold
Designed the causal inference + forecasting architecture and produced the benchmark proof. The technical core already exists.
OPERATOR ADVANTAGE
Warm intros to dozens of brands
10 years in DTC. Knows the buyer, the pain, and the weekly cadence. First two hires are already obvious: ML/data and commercial operator.
What investors will care about: solo-founder concentration risk is real. The defense is unusually strong founder-market fit and a very clear first-hire plan, not pretending the risk does not exist.
Ask & Vision
$1.25M
to turn proof into a real
software business,
then earn the right to a bigger vision.
Raise
$1,250,000
Instrument
Post-money SAFE
Valuation cap
$8M post-money
Runway
18 months
Contact
nate@thagorus.com
Use of Funds
Product + engineering55%
Partner onboarding25%
Infrastructure (compute, storage)15%
Legal & ops5%
18-Month Milestones
1
Sign 10 design partners across clear beachhead categories (months 1-12)
Outdoor apparel, home fragrance, seasonal consumables. Start where timing mistakes are frequent and expensive.
Outdoor apparel, home fragrance, seasonal consumables. Start where timing mistakes are frequent and expensive.
2
Run weekly decision support in production (months 6-15)
Forecasts, attribution, and timing signals used on a real operating cadence with repeat usage, not demo usage.
Forecasts, attribution, and timing signals used on a real operating cadence with repeat usage, not demo usage.
3
Convert first paid accounts (months 12-18)
Show that this is software customers will pay for before any larger data business is required.
Show that this is software customers will pay for before any larger data business is required.
What this round answers: Can WeatherVane become a repeatable software company with real weekly usage, paid accounts, and a growing corpus? If yes, the long-term company can become much larger than the initial workflow.
Appendix
Honest Answers to Hard Questions
OpenAI / Google could build this
They could build a general time-series model — and several have (Chronos, TimesFM). What they can't build is the cross-brand training corpus of retail x weather x macro data, assembled through commercial relationships with real brands. OpenAI doesn't have NorSari's SKU-level transaction history. The moat is the data.
What about TimeGPT, Chronos, TimesFM?
Foundation model forecasters do zero-shot prediction on individual time series. They predict each series independently — no cross-brand learning, no causal attribution, no retail-specific training corpus. We're building the domain-specific data network they can't.
What is the +0.405 nats claim exactly?
The difference in negative log-likelihood (NLL — a standard measure of prediction quality; nats are natural units of information) between a retail-only model and one trained on diverse economic data. N=100K tokens, honest first-order result. 0.1 nats on M5 is considered significant. Ours is 4x that. Currently expanding.
Zero revenue, two design partners — too early?
NorSari has been running on the WeatherVane model since 2017 — 44% CAGR over 8 years. Homesick provides 8 years of demand data. The math works. This raise funds formalizing the next 10 partner relationships and converting weekly usage into a repeatable business.
Appendix
Financial Model & Long-Term Vision
YEAR 1-2
Prove paid workflow value
Free design partners become live weekly users. The first commercial goal is not scale theater; it is software customers who keep coming back.
Primary output: usage, conversions, and a better corpus.
YEAR 2-4
Expand within retail
More brands, more adjacent categories, more operator seats, stronger onboarding, and a visibly compounding demand corpus.
Primary output: credible ARR and a moat investors can see.
YEAR 4+
Demand intelligence layer
If the workflow wins first, the long-term company can supply a broader demand intelligence layer into other systems and decisions.
Primary output: a much larger company built on top of the first wedge.
SEED STORY
Narrow
One buyer, one workflow, one budget owner.
SERIES A STORY
Repeatable
Paid accounts, weekly usage, and a corpus that compounds.
LONG-TERM STORY
Layer
A broader demand intelligence layer once the initial workflow is proven.
Framing discipline: the seed deck should sell the narrow software company first. The larger vision only matters because the narrow wedge can plausibly create a durable data advantage over time.