The opportunity

• The model + full methodology, the ensemble, the feature engineering, the sentiment scoring, the consensus filter. The how behind the 62.3% walk-forward hit rate.
• The curated practitioner-region map, the project's primary defensible IP. A season-plus of curation; not extractable from the public site.
• The accumulated historical sentiment corpus, scored UK arable practitioner data that compounds in value daily.
• The public forward track record, timestamped calls made ahead of harvest. The artefact that de-risks every claim on the site.
• The codebase + daily pipeline, ingest, scoring, render, publish, email, weekly video. Running in production today.
• The brand, domain, indexed footprint, and subscriber list.

What transfers is derived, proprietary data and the methodology that produces it: the curated practitioner-to-region map, the scored sentiment features and rolling aggregates, the compound-stress series, and the forward-call log, none of it republished third-party text. The weather, satellite, and price inputs are standard open or licensed sources (ECMWF ERA5 via Open-Meteo, Sentinel-2, AHDB Corn Returns); the practitioner-sentiment layer is built by observing public UK arable discussion under a clearly-identifying, publisher-respecting ingest, and is stored as scores and features, not as a raw corpus for redistribution.

Full data lineage, source terms, and exactly what conveys at completion are documented in the diligence pack, so a buyer's data and legal teams get a clear provenance map under NDA before anything is signed.

CropIntel is engineered as a UK arable platform, not a single-crop tool. The expensive, slow-to-build parts, the weather pipeline, the curated practitioner corpus (already multi-crop, the same growers discuss wheat, barley and oilseed rape), the scoring engine and the brand, are shared across crops. The marginal crop is cheap; the buyers are the same (merchants, insurers and desks trade the whole rotation).

In internal feasibility testing the same engine has been shown to extend cleanly to oilseed rape and barley, with correct agronomy matching wheat-grade performance on a second crop and delivering strong bad-year detection across crops. A costed, validated, ready-to-execute build plan for OSR and barley, with the full results, is in the diligence pack.

So an acquirer isn't buying a wheat forecaster; they're buying a proven UK arable engine with a shovel-ready expansion across the combinable-crop complex, included in the acquisition. (Wheat is the live, proven product today; the additional crops are demonstrated capability on the roadmap, not yet live.)

The right yardstick is value to you, not our build cost. Price this against the value it unlocks, which is a multiple of any build cost, and the build can't be compressed in calendar time anyway. Illustratively, per segment: a data publisher, around £3m of new recurring revenue from a forward tier across an existing base; a grain merchant, £1-2m a year of better-timed cover on a 500,000-tonne book; an insurer, a seven-figure swing on a single anticipated bad year; a commodity desk, a seven-figure seasonal PnL contribution. The per-segment workings are on the use-case pages, value at stake, illustrative, not guaranteed returns.

Revenue upside, not just a capability. For a data publisher, a forward-yield tier lifts ARPU across an existing subscriber base: the acquisition can pay back on cross-sell before a single new logo. For an operator, the value is in the decisions: a directional edge on a book worth tens of millions, or one early bad-year call that lets an insurer reserve a quarter ahead. The per-segment sizing is on the use-case pages.

Cheaper than building, and far faster. Starting from scratch means standing up the full pipeline, calibrating a model on two decades of regional data, and hand-curating a UK arable practitioner corpus, and only then can a public track record begin to accrue, one harvest at a time. That last part can't be bought off the shelf or accelerated with headcount: a credible multi-year record is calendar time, not engineering time. Acquiring gives you the working system and a head start measured in years.

High margin to operate. The system runs autonomously on commodity cloud infrastructure with negligible marginal cost to serve. Any revenue an owner layers on top drops almost entirely to the bottom line: a high-gross-margin asset, not a headcount-heavy service.

Optionality. UK wheat today; the methodology generalises to other cereals and geographies. The clearest near-term roadmap item is an earlier-lead bad-year detector on the accumulating sentiment layer, kept separate from the validated ensemble, pushing the call further ahead of harvest as the forward data matures. The roadmap is yours to direct post-acquisition, you're buying a platform, not a single number.

Public data is free. The entire site, every call, the track record. No login, no paywall: it is the diligence artefact.

Acquisition (POA). The whole asset: model, methodology, corpus, track record, code, brand. An asset sale with full IP assignment and a transition period, for the strategic buyer who wants to own and extend it.

CropIntel runs autonomously in production on a single cloud instance (AWS, London region). An automated daily pipeline ingests data, scores sentiment, runs the ensemble, renders the site, pings search indexes, emails subscribers, and on Sundays renders the weekly video, finishing by 07:25 UTC. The database is backed up daily, with an append-only off-host copy a compromised server cannot erase. No manual step is required for the daily call to publish.

It is deliberately simple to operate and to transfer: standard Python, standard cloud primitives, no bespoke infrastructure, one documented runbook. The third-party inputs (weather, satellite, price, and the LLM scoring layer) are mainstream and substitutable, none is a single point of lock-in.