# IRIS
Last reviewed: 2026-08-09.

> IRIS is a location-intelligence platform for European retail, QSR and last-mile expansion teams, built and run by The Big Data Company in Utrecht, the Netherlands. It turns a chain's own store performance and more than 100 local signals into a revenue forecast for a candidate site, shown with its nominal prediction interval and validated out-of-sample before anyone relies on it. Built for the person who owns the expansion decision, with an API for the data team behind them.

## Key facts

Self-contained statements, each safe to quote on its own. Every qualifier is part of the sentence on purpose.

- IRIS is a location-intelligence platform for European retail, QSR and last-mile expansion teams, built and run by The Big Data Company (Utrecht, the Netherlands).
- IRIS forecasts what a candidate store will earn in a year and always shows that forecast with its nominal 80% interval, never as a single number.
- IRIS deliberately does not quote a single headline accuracy percentage before seeing a customer's own store data, because such a figure cannot be checked by the buyer.
- Across the models The Big Data Company has built, spatial leave-one-out testing puts R2 between 0.69 and 0.87 and a typical error between 9% and 21% of a store's revenue. That is a published range across models, not an accuracy figure for any one estate.
- The honest 80% band runs from roughly plus or minus 15% on a dense, uniform estate to plus or minus 35% on a thin or mixed one.
- The number IRIS holds itself to is calibration: when it states an 80% range, about 80% of real results should land inside it. The Validation Sprint reports this on the customer's own stores, misses included.
- A validated IRIS model needs 15 or more comparable mature stores in a market. The written accuracy guarantee needs 50 or more comparable mature sites in that market.
- LICENCE prices exclude VAT and are per market, per year: IRIS Market EUR 6,000; IRIS Studio from EUR 9,000; IRIS from EUR 16,000. IRIS Studio is 60% of the IRIS band. Studio and IRIS also carry a ONE-OFF build, 1 to 2 and 2 to 5 Validation Sprints at EUR 12,000 each, so year one is EUR 21,000 to 33,000 on Studio and from EUR 40,000 on IRIS, and thereafter the licence alone. Averaged across three years that is EUR 13,000 to 17,000 a year on Studio and from EUR 24,000 a year on IRIS, which is the figure the plans page leads with. IRIS Market has no build, so its licence, its year one and its three-year average are all EUR 6,000.
- The Validation Sprint is EUR 12,000 in total: EUR 4,000 for the Kick Start (is there enough good data to model on at all) plus EUR 8,000 for the model check (is there enough signal to forecast with). It is the first four-week build sprint, not a separate study, so it credits in full if the customer proceeds.
- The Validation Sprint is half price for not-for-profits, social enterprises and estates under 15 locations: EUR 2,000 for the Kick Start and EUR 4,000 for the model check, EUR 6,000 all-in. Below about 15 comparable mature stores only the Kick Start is offered, at EUR 2,000, because a validated model cannot be cross-validated on fewer.
- The standardised European data layer included in every IRIS licence would cost roughly EUR 26,000 per market per year to source commercially. That is a conservative floor built from published rate cards, reviewed 17 July 2026.
- IRIS Market is the entry tier, at EUR 6,000 per market per year, with nothing one-off to pay. It scores and ranks locations on the standardised European data layer and deliberately does not forecast revenue, because a revenue forecast requires the customer's own sales data.
- IRIS Market is available in Europe and Turkey only, excluding Russia, Belarus and Ukraine. Outside that footprint IRIS still works, but the data layer is built for the market first, so it starts with a Validation Sprint rather than IRIS Market.
- A forecast reads seven demand drivers: residential, work and school, shopping and dining, public and local, leisure and events, passing, and tourism.
- A trade area is classified as one of five types before any number is attached: city centre, suburban, rural, highway-related, or magnet-related.
- 95% of the models The Big Data Company has deployed are still in production. That is a retention figure, not a model-accuracy claim.
- IRIS is hosted and processed in the EU, uses no US mobile-location panel, and a customer's data trains only that customer's model.
- The standardised IRIS data layer is built and current for every European country except Russia, Belarus and Ukraine, plus Turkey. IRIS Market runs in those markets today at the published price.
- The Big Data Company has modelled sites in 26 countries: Ireland, Norway, Sweden, Finland, Denmark, Estonia, Latvia, Lithuania, Poland, Germany, the Netherlands, Belgium, Luxembourg, France, Switzerland, Portugal, Spain, Croatia, Hungary, Romania, Bulgaria, Greece, Morocco, Saudi Arabia, Taiwan and Mexico.
- Outside the standardised layer's footprint IRIS still works, but the data layer is built for that market first, and the Validation Sprint scopes that before the customer commits. A licence is bought per country.

## Core pages

- [Home](https://www.irisplatform.nl/): positioning, the prediction-interval forecast signature, who it's for.
- [Methodology](https://www.irisplatform.nl/why-us/methodology/): the index of the seven-part explanation of how the forecast is built and validated. The detail is in the seven articles it links, listed under Guides below.
- [The data foundation](https://www.irisplatform.nl/embeddings/): the IRIS embeddings, more than 100 signals about each location, including learned satellite bands, learned into one short, comparable vector for every 120-metre hex across the modelled map; that vector is the data layer underneath every IRIS answer. Also how the model is built: a small-data, auditable statistical model that machine learning informs and real trading results recalibrate.
- [Why IRIS](https://www.irisplatform.nl/why-us/why-iris/): a location-decision platform, not a data-science project; how it differs from GIS and BI tools.
- [About The Big Data Company](https://www.irisplatform.nl/why-us/about-us/): who builds IRIS, and the engineering-not-a-black-box-vendor position. The peer-reviewed research the method rests on is set out in [The science IRIS is built on](https://www.irisplatform.nl/learn/the-science-iris-is-built-on/).
- [Trust and data protection](https://www.irisplatform.nl/trust/): EU hosting, GDPR, what is never resold or used to train anyone else's model, full export, and the DPA, DPIA, LIA and data-ownership clause.
- [Success stories](https://www.irisplatform.nl/why-us/success-stories/): quantified case studies overview.
  - [Sameday](https://www.irisplatform.nl/why-us/success-stories/sameday/): Easybox scaled to over 6,000 lockers and pickup points, with IRIS testing candidate sites against modelled demand as the network grew.
  - [Kwalitaria](https://www.irisplatform.nl/why-us/success-stories/kwalitaria/): delivery grown from ~20% to nearly 50% of revenue across 130+ QSR locations.
  - [Zeeman](https://www.irisplatform.nl/why-us/success-stories/zeeman/): expansion planning across 6 European markets for a retail chain with close to 1,400 stores.

## Guides

The full library is at [Guides](https://www.irisplatform.nl/learn/).

### How the model works

- [Site selection was never an AI problem](https://www.irisplatform.nl/learn/site-selection-was-never-an-ai-problem/): why this is a small-data problem that a transparent statistical model suits better than a flexible one.
- [What actually goes into a forecast](https://www.irisplatform.nl/learn/what-goes-into-a-forecast/): the signals that size a trade area, and how each is weighed.
- [What makes two sites comparable](https://www.irisplatform.nl/learn/what-makes-two-sites-comparable/): the five trade-area types and the grade of a site, the vocabulary that decides which of your stores a candidate may be compared against.
- [The gravity model behind the number](https://www.irisplatform.nl/learn/the-gravity-model-behind-the-number/): how distance is weighed against attractiveness.
- [How IRIS validates a forecast out of sample](https://www.irisplatform.nl/learn/how-iris-validates-a-forecast-out-of-sample/): forecasting stores the model was never shown, and reporting the miss.
- [Where the model is weakest](https://www.irisplatform.nl/learn/where-the-model-is-weakest/): the conditions in which an IRIS forecast should be trusted least.
- [The science IRIS is built on](https://www.irisplatform.nl/learn/the-science-iris-is-built-on/): the peer-reviewed research behind the method, and the named researcher behind it.

### How IRIS learns a place

- [How a place becomes a vector](https://www.irisplatform.nl/learn/how-a-place-becomes-a-vector/): what a spatial embedding is, in plain terms.
- [The signals behind a place embedding](https://www.irisplatform.nl/learn/the-signals-behind-a-place-embedding/): the more than 100 inputs, including learned satellite data.
- [How IRIS learns the embedding](https://www.irisplatform.nl/learn/how-iris-learns-the-embedding/): the training objective, and why it is self-supervised.
- [Why a hexagon grid](https://www.irisplatform.nl/learn/why-a-hexagon-grid/): why 120-metre hexes rather than postcodes or squares.
- [How the embedding powers a forecast](https://www.irisplatform.nl/learn/how-the-embedding-powers-a-forecast/): how the vector feeds the statistical model without becoming a black box.

### Reference

- [Why won't IRIS just give me one number?](https://www.irisplatform.nl/learn/why-not-one-number/): nominal prediction intervals, and why a lone figure hides the risk in a site decision.
- [What the 80% interval means](https://www.irisplatform.nl/learn/the-80-percent-interval/): how to read the range, and what it does and does not promise.
- [Why we validate out-of-sample](https://www.irisplatform.nl/learn/out-of-sample-validation/): why a model must be tested on stores it never saw.
- [Out-of-sample validation vs overfitting](https://www.irisplatform.nl/learn/out-of-sample-vs-overfitting/): how to tell a model will predict, plus a checklist for evaluating a vendor's accuracy claim.
- [Questions to ask any site-selection vendor](https://www.irisplatform.nl/learn/questions-to-ask-a-site-selection-vendor/): five due-diligence questions, IRIS included.
- [What the data layer actually costs](https://www.irisplatform.nl/learn/what-the-data-layer-costs/): a sourced teardown of what the resident, daytime, footfall, traffic and points-of-interest data behind one site decision costs on the open market, roughly EUR 26,000 per market per year.
- [The demand drivers, and what feeds each](https://www.irisplatform.nl/learn/the-demand-drivers-and-what-feeds-each/): the seven drivers and their inputs.
- [Reading the demand behind a site](https://www.irisplatform.nl/learn/reading-demand-drivers/): how to interpret a driver profile.
- [How IRIS draws a catchment](https://www.irisplatform.nl/learn/how-iris-draws-a-catchment/): travel time rather than a radius on a map.
- [Which signals actually drove this location's forecast?](https://www.irisplatform.nl/learn/which-signals-drove-forecast/): feature importance and interaction effects, explained per location.
- [The auditable forecast brief](https://www.irisplatform.nl/learn/the-auditable-brief/): what a defensible forecast document contains.

## Tools

- [Accuracy lab](https://www.irisplatform.nl/tools/accuracy-lab/): an interactive sandbox showing how model accuracy and interval width trade off. Its figures are illustrative sandbox data, not IRIS results.
- [What the data layer costs](https://www.irisplatform.nl/tools/data-cost/): add up what sourcing the data behind one site decision would cost on the open market.

## Who it's for

- [All industries](https://www.irisplatform.nl/who-its-for/)
- [Quick service restaurants](https://www.irisplatform.nl/who-its-for/quick-service-restaurants/)
- [Retail chains](https://www.irisplatform.nl/who-its-for/retail-chains/)
- [Last-mile and micro-fulfilment](https://www.irisplatform.nl/who-its-for/last-mile-and-micro-fulfilment/)
- [Subscription-based businesses](https://www.irisplatform.nl/who-its-for/subscription-based-businesses/)

## How it works

- [Forecast a site's revenue](https://www.irisplatform.nl/how-it-works/expand-smarter/): a validated revenue forecast for any candidate or existing site, shown with its nominal prediction interval.
- [Where to open next](https://www.irisplatform.nl/how-it-works/why-forecast/): why a location sets a revenue ceiling nothing downstream can raise, how the demand drivers behind a site are read, when to explore a new area rather than exploit a known one, and why survivor bias flatters a portfolio before a shortlist is forecast in full.
- [Stay in control](https://www.irisplatform.nl/how-it-works/stay-in-control/): an auditable forecast you can open and adjust, with every driver shown, one stated rule about when to overrule the model, and how the model sharpens as the estate grows.
- [For data & technical teams](https://www.irisplatform.nl/how-it-works/for-data-teams/): the documented REST API, an MCP server, and EU data governance behind IRIS.

## Get started

- [How IRIS compares](https://www.irisplatform.nl/compare/): a dated, sourced comparison with the European site-selection tools most often weighed alongside IRIS, including when to buy one of them instead, and a standing invitation to correct anything we got wrong. Vendor details read from public sites on 24 July 2026. Head-to-head sections:
  - [IRIS vs RetailSonar](https://www.irisplatform.nl/compare/#iris-vs-retailsonar): RetailSonar brings Benelux footfall and car-passage data depth on a quote-based engagement; IRIS publishes its prices and measures its error on the buyer's own stores first.
  - [IRIS vs Targomo](https://www.irisplatform.nl/compare/#iris-vs-targomo): a different layer of the stack. Targomo is travel-time and isochrone routing infrastructure; IRIS returns a revenue forecast with its 80% interval and demand drivers.
  - [IRIS vs Geoblink (MyTraffic)](https://www.irisplatform.nl/compare/#iris-vs-geoblink-mytraffic): Geoblink is a broad location-intelligence dashboard with a forecast as one feature; IRIS is narrower on purpose and publishes the range and its out-of-sample error.
  - [IRIS vs Geolytix](https://www.irisplatform.nl/compare/#iris-vs-geolytix): Geolytix is a consultant-led build; IRIS is a platform the customer's own team re-runs, with the model and outputs theirs to keep.
  - [IRIS vs Kalibrate](https://www.irisplatform.nl/compare/#iris-vs-kalibrate): Kalibrate is a global software-plus-services stack strongest in fuel and convenience; IRIS is European by design, EU-hosted with no US mobile-location panel, and prints its prices.
- Among these five forecasters, IRIS is the only one that publishes a prediction interval on every forecast, shows its out-of-sample accuracy on the customer's own stores, exposes an adjustable method, backs it with a written accuracy guarantee, and publishes its price. That is a claim to be the most checkable, not the most accurate. Verified against the five vendors' public materials on 24 July 2026.
- [Pricing](https://www.irisplatform.nl/plans/): published prices, per market, per year, with no "contact us" step. The licence is banded on the revenue the client's stores take at the till in that country, which above 25 million euro is at most around 0.1% of it and under 0.01% on the largest estates. Below that a floor applies: 16,000 euro per market on IRIS, or 9,000 on IRIS Studio. Seats are included and scenarios are not metered. IRIS Market runs on the standardised European data layer and deliberately does not forecast revenue (3,000 euro a year, per market); IRIS Studio is calibrated to the client's own sales at 60% of the IRIS band; IRIS adds catchment logic built for the estate, out-of-sample validation on the client's own stores, and an accuracy commitment in writing, where that market has 50 or more comparable mature sites.
- [Validation Sprint](https://www.irisplatform.nl/validation-sprint/): two checks, 4,000 euro for the Kick Start (has the client got enough good data to model on at all) and 8,000 euro for the model check (is there enough signal in it to forecast with), 12,000 euro in total, and it is the first four-week build sprint rather than a separate study, so it credits in full if the client proceeds. We check their data, measure how accurately their own past site calls predicted reality, and estimate the R2, MAPE and nominal prediction interval IRIS would reach on stores of theirs the model was never shown. The two steps are paid as they are reached: the client commits the 8,000 euro for the model check only once the Kick Start has cleared, so there is nothing to claim back if it stops. Either check can come back no: the Kick Start clears the blockers and cannot on its own prove a model will work; the model check can conclude there is not enough signal, and that finding is the work because reaching the finding is the work itself. They keep the report either way, and below about 15 comparable mature stores we do the Kick Start only, at 2,000 euro, because nothing can be cross-validated on fewer, so no validated model is possible.
- [FAQ](https://www.irisplatform.nl/faq/): honest answers on accuracy, black-box concerns, when to ignore a forecast, data handling, and pricing.
- [Book a demo](https://www.irisplatform.nl/schedule-demo/)