Geospatial Data Quality at Scale: A Framework from Detection to Decision

Executive Summary

Most geospatial data quality workflows answer one question: what is wrong? They rarely answer what to do about it, in which order, and why.

We built a framework that does, and applied it to approximately 330,000 geometries from several providers across 20 countries covering several EUDR commodities.

The results are clear: 35.6% of geometries carry quality issues, but less than 1% require field intervention. The majority can be fixed through automation. More importantly, the dominant errors are systemic rather than individual, pointing to pipeline and integration failures that no amount of field retraining will solve.

Data quality, managed this way, becomes a strategic lever rather than one-off cleaning task.

The Reality of Geospatial Data Quality

Each geometry was scanned for quality issues, scored along two dimensions: impact on analysis (severity) and likelihood of correction (fixability), and placed into a priority matrix that maps directly to a remediation strategy.

Our findings:

• Most geometries carry 0 to 2 flags (quality issues), with a rapidly decreasing tail, but a non-negligible share accumulates multiple compounding issues.

• Severity is skewed toward low to moderate values, meaning most problems do not fully block analysis but still introduce bias. Fixability, however, is heavily concentrated at high values: a large proportion of detected issues can be corrected through automated or semi-automated processes.

• The dominant error types are cross-feature and structural, not geometric or topological. It means that inconsistencies at the dataset level (duplicates, overlaps, encoding problems) are more prevalent than individual shape errors. The root cause is systemic data management, not isolated digitization mistakes. Improving data pipelines and integration processes will yield higher returns than focusing solely on field practices.

Near duplicate polygons are geometries from same supplier that overlap with >80% IoU and fake multipolygons are polygons with type MULTIPOLYGON that are in reality POLYGON

• Mapped into a priority matrix, the flagged polygons (representing 35.6% of the total dataset) resolve into three actionable segments: a dominant quick wins group (easily fixable at scale), a small critical issues group (concentrating most of the risk), and a high-value automation group (where fixing yields the strongest analytical gains). This reduces a complex distribution of errors to a small number of operational decisions.

Provider-level error profiles and where to focus effort

Aggregating signals by provider shows that data quality issues are not uniform and require tailored solutions rather than a single generic approach.

One provider’s dataset, for example, is dominated by cross-feature inconsistencies, indicating duplication or data integration issues, as well as structural anomalies suggesting problems in data encoding or export pipelines. Targeted interventions for this provider would include deduplication at ingestion, stricter data integration rules, and validation of export formats to enforce consistent geometry structures.

This kind of profiling provides direct guidance for improving data collection and processing: it identifies whether issues originate from field practices or data pipelines, targets training or tooling where error types concentrate, and enables monitoring of improvements over time using consistent metrics.

The Framework that Generated These Results

Error Categories

Geospatial data quality issues can be regrouped into four categories, each capturing a distinct type of failure, from the internal validity of a single geometry to its consistency within a dataset :

From Categories to Checks

We implemented a set of checks across the four categories to detect signals indicative of quality issues:

When a check fails, a flag is assigned to the geometry.

A Bivariate Scoring System

Detecting errors is insufficient. The objective is to decide what to fix, how, and in which order. Each detected signal is therefore translated into two complementary dimensions.

Severity quantifies how much a given issue affects downstream analysis, ranging from 0 (no impact) to 5 (analysis not possible). Topological errors are blocking as invalid geometries cannot be used in most operations. Geometric and structural errors introduce varying levels of bias, distorting metrics or affecting interpretation. Severity is computed hierarchically: invalid topology immediately sets the maximum score, other errors are aggregated by category, and only the most severe issue within each category is retained. This avoids over-penalizing geometries with multiple correlated issues.

Fixability measures how likely it is to correct a geometry while preserving its meaning, ranging from 0 (not fixable) to 5 (fully fixable). Some errors are purely technical and can be fixed deterministically. Others require interpretation and may introduce uncertainty. Some require recollection because no reliable correction can be applied. The aggregation follows a bottleneck logic: the least fixable issue dominates, because one blocking problem is enough to invalidate automated correction.

The Priority Matrix

Severity and fixability define a two-dimensional decision space. Each geometry is positioned in a matrix that maps directly to a remediation strategy:

From One-Off Diagnosis to Continuous Quality Monitoring

Most data quality workflows stop at detection : issues are identified after collection, but not prevented or systematically prioritized. This framework addresses that gap by integrating decision-making directly into the data pipeline.

The ultimate goal of this framework is not to diagnose problems long after collection, but to control data quality as it is produced. In practice, this means embedding quality checks into ingestion pipelines, evaluating geometries as they are collected, and surfacing critical issues immediately while remediation is still feasible.

This shift enables operational feedback loops. Field teams receive timely signals on critical errors, allowing remapping before they move to new regions. Data quality becomes a constraint of collection rather than a downstream concern.

At scale, this transforms the system: heterogeneous practices converge toward standardized processes, reactive cleaning is replaced by proactive control, and unstructured errors become measurable performance indicators. The transition relies on a simple structure: a limited set of error categories, interpretable signals derived from checks, and a scoring system that prioritizes actions.

The insights and thresholds presented here are derived from the dataset analyzed and therefore reflect its specific characteristics. As new datasets and error patterns emerge, both the checks and the scoring logic will be refined. The objective is not to define static rules, but to build a system that continuously adapts and improves as new data is collected.

L’article Geospatial Data Quality at Scale: A Framework from Detection to Decision est apparu en premier sur Sustaain.

Understanding Traceability and Traceability Risk

Traceability requires producing, collecting, and storing documents that prove the origin of products and their chain of custody. Traceability risk is the risk of mixing or substitution with products of unknown origin along the chain of custody.

The EUDR and other regulations have requirements on traceability: due diligence must be performed to determine that the risk of circumvention by mixing with products of unknown origin is negligible.

The Complexity of Supply Chains and Chain of Custody Documentation

The material reality of supply chains is infinitely complex. A full description is intractable, but some supply chains are more complex than others in their numbers of steps, intermediaries, and branches.

Depending on the specifics of the supply chain and the availability of documentation, the amount of documentary evidence available about chains of custody can range from incomplete to overwhelming.

Operators performing the due diligence, especially small and medium actors, face multiple challenges:

• They have limited access to documentary evidence;
• Their leverage to request it when it is not readily available is limited;
• Their capacity to process the documentary evidence is also limited by resources or competencies.

Modelling the Chain of Custody for Traceability Compliance

How to deal with this complexity? The chain of custody in traceability due diligence is a model of the material reality of supply chains; it is a simplification. This model is built using the available documentary evidence. How to make sure — with limited resources — that the model is good enough (i.e. close enough to the reality it describes, with the appropriate level of detail) to conclude that the risk of circumvention is negligible, and reach compliance with traceability requirements?

This requires identifying what are the non-negotiable parts of regulations, and where are the moving parts, and finding a workable compromise.

A non-negotiable requirement of the regulation is that all entities (businesses and persons) along the chain of custody must be documented.

Among the “moving parts” are the legal value of documentation (a simple declaration vs. a contract/certification), and the level of detail of documentation (a general description of a connection vs. a description of all volumes transferred between entities). These two dimensions together define the “evidentiary level” of documentation

How to set the cursor for these “moving parts”? The answer lies in the context. Traceability risk does not exist in isolation; it is one among other dimensions of compliance risk. The country risk and the supplier risk define a “background risk” level.

With the assumption that traceability risk is correlated with the background risk, the required evidentiary level of traceability documentation can be modulated based on background risk. In a low-background risk context, the evidentiary threshold of traceability documentation can be lowered.

Designing a Traceability System for EUDR Compliance

Based on these principle, we have designed a traceability risk methodology to guide operators through the collection of traceability documentation, and a traceability system that can capture relevant data from the traceability documents, store it in a structured and safe system that is audit-ready, flag traceability issues and advise users on mitigation strategies.

Want to go further on traceability and EUDR compliance?

Watch our webinar series “Data Clarity in a Post-Truth EUDR.” In the first episode, with Sopex, we share a pragmatic approach to supply chain modelling and traceability.

L’article Managing Traceability Risk in EUDR Compliance est apparu en premier sur Sustaain.

2026: The year nothing happens, yet everything shifts.

When politics stalls, trust migrates, from institutions to systems. The advantage will belong to those who can measure, prove, and explain reality when narratives start to wobble.

In Davos, the future is being reassembled (again) out of well-tailored phrases. On our screens, young people film burning roofs, as if documenting catastrophe were the last form of agency left. In France, there is the familiar theatre of leadership – a head of state auditioning for an action film while the plot is written by constraints. And then a small, revealing detail: a proper winter, yet no proper cold snap. Everything is there. Not a rupture, but a drift.

I wonder: how do we protect ourselves from the present age—reflective, over-informed, briefly enthusiastic, then sliding back into inertia? The answer is not more commentary. It is a better stance: widen the lens, narrow the priorities, and ship.

The age of drift

America heads into midterms with institutions primed for trench warfare; France approaches its presidential horizon with policy increasingly performed rather than built. The net effect is not paralysis, exactly, but a thinning of ambition. Decisions that require patience, trade-offs, and political capital are postponed. Meanwhile, the world’s old “rule-based” comfort continues to fray. Not replaced by a new doctrine, but by a patchwork of coalitions, bargains, and temporary alignments.

Europe, in particular, has a choice to make. It is often described as a “power of measure”, but that should be read in the older sense too—measure as moderation. Aristotle would have recognised the instinct: not regulation for its own sake, but the disciplined insistence that rules apply, contracts hold, and standards are predictable. In a world drifting away from the rule-based order, Europe’s advantage is not necessarily to add layers of bureaucracy, but to double down on what it can uniquely offer: the rule of law as infrastructure.

The machinery of accountability keeps moving. In years when politics slows, measurement accelerates. Standards are refined. Reporting expectations harden. Due diligence becomes less a slogan than a checklist. The centre of gravity shifts away from speeches and towards systems that can withstand scrutiny. That is why Europe may find itself unusually well-positioned. It is not the loudest player, nor the most forceful. But it can be the author of the grammar: what counts, what can be compared, what can be verified. In a world of drift, proof becomes a form of power.

Food refuses to negotiate

If 2026 has a single, unavoidable reality, it is food. Agriculture sits at the junction of climate risk, trade friction, and social stability, where the costs of wishful thinking are immediate. This is not a sector that tolerates long debates about values; it rewards operational competence and punishes fragility. Supply chains that feed the world are being asked to deliver more resilience, more traceability, and more sustainability, often while margins remain tight and volatility remains the norm.

Soft commodities are the stress test in plain sight. Their markets oscillate between fatigue and sudden spikes; their supply chains run through regions where data is sparse and verification is hard; their stakeholders face simultaneous pressure from climate variability, regulatory scrutiny, and shifting consumer expectations. In such conditions, vague commitments are not merely insufficient, they are dangerous. “Sustainability” is becoming operational not because the world has become more virtuous, but because the times are less forgiving.

The questions are no longer philosophical. They are practical, sometimes brutal: Where did it come from? Under what conditions? Can you prove it? The decade is turning from declarations to verification, from narrative to evidence.

Sustaain 2026: boring, on purpose

If 2026 is an interlude for politics, it is not an interlude for us. It is a year of construction, quietly ambitious, deliberately unglamorous. We have chosen discipline over spectacle. While hype cycles tempt companies to promise shortcuts, we are building what endures: the infrastructure that makes proof possible when incentives shift and scrutiny tightens.

Our work sits where complexity is real: between geospatial observation, documentary evidence, and field data. The aim is not to produce prettier reporting. It is to create systems that hold under pressure, systems capable of making sense of fragmented realities, detecting inconsistencies, and delivering defensible answers in the places where the world is hardest to model. That is what it means to de-risk technology for real food systems: not rushing to the loudest product, but hardening the foundation.

We are unapologetically focused on soft commodities and the sustainability of food systems, because this is where the future will be decided. Not in abstract ESG rhetoric, but in supply chains that feed billions, where verification is costly, and resilience is no longer optional.

A year in which “nothing happens” is often the year in which foundations shift. The loud events come later. The quiet rearrangements come first. 2026 will reward those who do not confuse motion with progress. Those who understand that, in a fragmented world, proof is not a constraint. It is the advantage.

L’article 2026: The year nothing happens, yet everything shifts. est apparu en premier sur Sustaain.

Coffee Suitability Maps: Mapping Where Coffee Can Thrive in Uganda

Finding coffee plots from satellite imagery requires knowing where coffee can exist in the first place. Suitability mapping provides fast, actionable spatial constraints grounded in agronomic knowledge.

What is Suitability Mapping and Why It Matters

Coffee is a plant, and like any plant it grows only within specific ecophysiological limits: temperature, rainfall, soil chemistry, terrain, and land use all define whether coffee can survive, let alone be productive.

Decades of agronomic research have focused on identifying these optimal growing conditions for soft commodities. This knowledge underpins critical decisions: where to grow crops, what yields can be expected, and how production zones may shift under climate change (DaMatta et al., 2018; Pham et al., 2019; Kath et al., 2020).

For coffee, suitability is driven by four main groups of factors:

• Climatological factors such as temperature and rainfall
• Edaphological factors related to soil properties
• Physiographic factors such as elevation and slope
• Socioeconomic factors including land use and legal constraints

Scientific literature provides concrete thresholds for many of these variables. For example, studies such as Salas et al. (2020) and Nigussie et al. (2024) define clear suitability ranges for Arabica coffee across climate, soil, terrain, and land-use dimensions :

Sub-criteria suitability threhsolds (requirements) for Arabica growing from Salas et al. (2020) and Nigussie et al. (2024)

Suitability mapping consists of combining all these constraints into a single spatial representation. This is inherently multi-factorial and non-trivial. Each variable may be suitable in isolation, yet unsuitable once combined with others. Elevation may be ideal, but temperature too high. Rainfall may be optimal, but land use prohibitive…

Traditional approaches address this complexity through expert-based weighting systems (Salas et al., 2020; Nigussie et al., 2024). While powerful, these methods are slow to implement, difficult to reproduce, and sensitive to expert choice and regional context. Different experts often assign different importance to the same factors, leading to inconsistent results.

Suitability mapping therefore sits at the intersection of agronomy, data integration, and decision-making. Done well, it converts fragmented environmental knowledge into a coherent spatial filter that supports planning, investment, and downstream analytics such as remote sensing and machine-learning models.

Our Approach

The objective was to produce suitability maps for both Arabica and Robusta coffee across Uganda, using a method that is fast to deploy, transparent in its assumptions, and technically defensible.

Our methodology :

1. The starting point is the scientific literature. For each coffee species, suitability thresholds are extracted for key variables, defining optimal, sub-optimal, and unsuitable conditions (see the table above for Arabica coffee).
2. We then assemble a set of open, global datasets covering the main drivers of coffee suitability. Climatic variables, elevation, soil properties, land cover, and protected areas are collected from established reference sources and harmonized onto a common spatial grid covering Uganda at 100 m resolution. Coarser datasets are interpolated to this grid to enable pixel-level comparison across all variables.

1. Each variable is assigned a suitability score based on literature thresholds.
   • Continuous variables such as rainfall, temperature, elevation, pH, and CEC are classified into discrete suitability classes.
   • Categorical variables such as land cover are mapped directly to suitability scores or exclusion masks. Protected areas and water bodies are treated as hard constraints and fully excluded from suitability.
2. Suitability scores are combined through simple addition across factors. No expert-based weighting is applied.
   • Elevation is enforced as a hard threshold due to its dominant role in defining coffee agro-ecological zones in Uganda (UCDA, 2019). This choice prioritizes interpretability and reproducibility over fine-tuned optimization.
3. The resulting composite scores are reclassified into three final categories: not suitable, sub-optimal, and optimal. The full workflow is applied independently to Arabica and Robusta to reflect their distinct ecological requirements.

This approach deliberately favors clarity over complexity. Every assumption is explicit, every transformation traceable, and every output directly linked to agronomic knowledge and open data.

What We Found

Single factor suitability maps

Each factor produces a distinct spatial signal.

• Climatological suitability highlights broad regional patterns driven by temperature and rainfall gradients. Large areas of Uganda fall within acceptable climatic ranges, but marginal and unsuitable zones emerge at the extremes, particularly in hotter lowlands or cooler high-altitude regions.
• Physiographic suitability, dominated by elevation and slope, acts as a strong spatial constraint. Elevation sharply differentiates Arabica and Robusta zones, confirming its role as a primary limiting factor.
• Edaphological suitability is more spatially homogeneous at national scale. Soil constraints eliminate fewer areas than climate or elevation, but they refine suitability locally, especially at the margins of otherwise favorable regions.
• Socioeconomic factors introduce fragmentation. Land use and protected areas break otherwise continuous suitable zones into patchy landscapes, reflecting real-world constraints on cultivation rather than biological limits.

Taken individually, no single factor explains suitability. Their value lies in combination.

Composite suitability maps

Once factors are combined, coherent agro-ecological patterns emerge.

Robusta suitability dominates warm, humid, lowland regions, forming large contiguous areas aligned with known production basins. Arabica suitability is far more restricted, concentrated in cooler, higher-elevation zones, primarily along mountainous and highland regions.

Sub-optimal zones appear as transition belts where one or more variables fall outside optimal ranges. Unsuitable areas are driven by elevation limits, climatic extremes, or hard land-use exclusions rather than gradual degradation.

The resulting maps align closely with Uganda’s known coffee-producing regions (see the map in the work of Kyalo et al. 2023). This agreement is not a calibration outcome but an emergent property of combining independent, literature-based constraints.

These maps do not predict yields. They define where coffee cultivation is ecologically and legally plausible. As such, they function as a spatial filter: reducing uncertainty, constraining downstream analysis, and establishing a defensible baseline for field operations, monitoring, and modeling.

How does it compares with real data ?

The suitability maps were evaluated against real-world data: approximately 39,000 coffee farm polygons collected from multiple traders across Uganda. For each polygon, suitability was extracted from the Arabica map, the Robusta map, and a combined map defined as the maximum suitability between the two species.

The comparison reveals clear structure. When evaluated against the Arabica suitability map, more than half of the observed farms fall into areas classified as unsuitable. This is not a failure of the model but a reflection of reality: Arabica occupies a narrower ecological niche and is not the dominant variety nationwide.

For Robusta, a majority of farms fall within optimal or sub-optimal Robusta zones, consistent with Uganda’s production profile. Sub-optimal classifications dominate, highlighting that much of production occurs in marginal but viable conditions rather than textbook optima.

The absence of variety-level information at farm scale limits interpretation. Some farms classified as unsuitable for Arabica may in fact grow Robusta, and vice versa. The combined metric compensates for this limitation and is therefore the most relevant comparison. Over 99.7% of observed coffee plots fall within areas classified as suitable for at least one of the two species. This indicates that the rule-based approach effectively captures the broad ecological envelope of coffee cultivation without being explicitly calibrated on farm locations.

Only a negligible fraction of farms fall into areas classified as unsuitable for both species. Manual verification shows these cases correspond primarily to geometrical overlaps with protected natural areas rather than true ecological mismatches.

This validation step confirms the role of the maps as a spatial filter. They are not yield predictors and do not aim for pixel-level accuracy. Their strength lies in excluding implausible regions while retaining virtually all observed production, which is exactly the behavior required for downstream monitoring, targeting, and modeling workflows.

Limitations and Next Steps

The model assumes equal importance across all factors and ignores interactions between variables. This simplification is deliberate but restrictive. Elevation, temperature, and rainfall are not independent, and soil constraints often modulate climatic effects rather than acting in isolation. Introducing weighted scoring and validating suitability classes against observed yields would improve the mapping.

Beyond incremental refinements, the search space can be constrained further. Satellite-derived signals or representation-learning approaches based on embeddings trained with ground truth data could capture complex environmental patterns (Brown et al., 2025). These methods would not replace agronomic reasoning but tighten it, reducing false positives while preserving coverage.

Conclusion

Suitability mapping turns environmental uncertainty into clear spatial insight. It provides a fast, interpretable baseline grounded in agronomy, showing where coffee can realistically grow. The resulting maps can help focus field surveys, support policy decisions, and constrain AI workflows for coffee detection and monitoring to ecologically credible areas, directing effort where it matters most.

The approach is intentionally simple and therefore imperfect. There is substantial room for refinement, from better factor interactions to satellite-driven representations and yield-aware validation.

The framework is not specific to Uganda or coffee. It is transferable to other countries and other commodities wherever agronomic knowledge and open data are available.

Interested in discovering the suitability for another countries or commodities to improve your decision making ? Don’t hesitate to contact us !

Sources :

• Brown, C. F., Kazmierski, M. R., Pasquarella, V. J., Rucklidge, W. J., Samsikova, M., Zhang, C., … & Kohli, P. (2025). Alphaearth foundations: An embedding field model for accurate and efficient global mapping from sparse label data. arXiv preprint arXiv:2507.22291.
• DaMatta, F. M., Avila, R. T., Cardoso, A. A., Martins, S. C., & Ramalho, J. C. (2018). Physiological and agronomic performance of the coffee crop in the context of climate change and global warming: a review. Journal of agricultural and food chemistry, 66(21), 5264-5274.
• Kath, J., Byrareddy, V. M., Craparo, A., Nguyen‐Huy, T., Mushtaq, S., Cao, L., & Bossolasco, L. (2020). Not so robust: Robusta coffee production is highly sensitive to temperature. Global Change Biology, 26(6), 3677-3688.
• Kyalo, G., Apunyo, P. C., Mwanjalolo, M., Luswata, C. K., Kawooya, R., & Niyibigira, E. I. (2023). Characterisation and Mapping of Soils in Major Coffee Growing Regions of Uganda. Journal o
• Pham, Y., Reardon-Smith, K., Mushtaq, S., & Cockfield, G. (2019). The impact of climate change and variability on coffee production: a systematic review. Climatic Change, 156(4), 609-630.
• Nigussie, W., Al-Najjar, H., Zhang, W., Yirsaw, E., Nega, W., Zhang, Z., & Kalantar, B. (2024). Enhancing coffee agroforestry systems suitability using geospatial analysis and sentinel satellite data in gedeo zone, Ethiopia. Sensors (Basel, Switzerland), 24(19), 6287.
• Salas López, R., Gómez Fernández, D., Silva López, J. O., Rojas Briceño, N. B., Oliva, M., Terrones Murga, R. E., … & Barrena Gurbillón, M. Á. (2020). Land suitability for coffee (coffea arabica) growing in Amazonas, Peru: Integrated use of AHP, GIS and RS. ISPRS International Journal of Geo-Information, 9(11), 673.
• Uganda Coffee Development Authority (2019). Robusta Coffee Handbook.

L’article Coffee Suitability Maps est apparu en premier sur Sustaain.

I used to be a Product Manager for Trading Softwares. In that life, the rules were simple: if a feature wasn’t in the spec, it didn’t ship. If the data didn’t validate the user story, you killed the feature. You defined the product, you built it, and you shipped it.

On the very first day of my career, I was tasked with building an Interest Rate Curve. Looking at the soft commodity markets today I not only see curves where financial theory does not apply (hello absence of arbitrage opportunities), I see a massive, industry-wide product failure.

For a century, the “product” was simple: a physical bean that met a sensory profile (Grade 1, Screen 18, FAQ). The market dynamics traders are currently wrestling with – the basis blowout, the liquidity squeeze, the backwardation – are not just signs of physical scarcity. They are symptoms of a redefinition of the product itself.

The market knows the price is wrong. What the C-suite needs to accept is that the spec has changed. And right now, the global supply chain is struggling to ship the new version.

1. The “Breaking Change”: The Bean is No Longer Enough

You know the market context: Cocoa is structurally broken, and Robusta is following suit. You don’t need me to tell you that high prices and volatility are painful.

But consider why the physical premium has disconnected from the futures terminal. It’s because the “futures bean” and the “physical bean” are no longer the same asset.

• The Futures Bean: A generic financial contract.
• The Physical Bean (Post-2025): A physical asset + a traceability proof + a polygon + a deforestation check + due diligence statement.

In product terms, the regulator (the EU) is shipping a mandatory update (EUDR) with a hard deadline (well, not so hard it seems…). The industry is trying to patch a legacy backend (analog supply chains) to support a modern frontend (digital traceability).

2. Traceability is Not Compliance; It’s Inventory Control

Many stakeholders in the soft commodities space still view traceability as a “sustainability tax”. For them, it’s a compliance cost to be minimized. This is a strategic error.

In a world of structural deficits (which we are likely entering for cocoa, coffee and a wide variety of agriculture products), data is your only inventory control. If you cannot see the farm, you cannot predict the yield. If you cannot predict the yield, you cannot hedge the risk.

The traders winning in this environment are not the ones with the biggest warehouses; they are the ones who converted their supply chain into a data lake years ago. They know which cooperatives are replanting, which ones are digitally mapped, and which ones will survive the “compliance cut.” If you don’t have the data, you don’t have the volume. The “invisible hand” can no longer find invisible commodities.

3. What to Expect Next: The “Identity Preserved” Economy

So, where does this settle? As we move past the initial shock of implementation, expect two structural shifts. To understand these, we must reconcile how the London “Premium” of 2024 evolved into the London “Quality Discount” of 2025 and 2026.

A. The Great Bifurcation (Two-Tier Pricing)

We are seeing a permanent split in the price structure between Tier 1 (Compliant) and Tier 2 (Legacy) materials. Trading strategies must now account for a widening spread between these tiers, a reality reflected in the diverging roles of ICE London and ICE New York.

1. The Scramble for Nearby Beans (Backwardation) Throughout 2024 and early 2025, the cocoa market was in a state of extreme backwardation, where near-term prices traded at a significant premium over future contracts.

• The Scarcity Premium: Because London is the primary benchmark for West African origins (Côte d’Ivoire and Ghana), the failure of those crops hit London inventories first and hardest.
• Factory Needs: European processors needed beans immediately to fulfill existing contracts. They were willing to pay a massive absolute premium for whatever was already in London warehouses, regardless of whether it met the new 2026 “digital specs”.
• The Inversion: In late 2024, extreme supply tightness drove the market into a deep inverse, with nearby contracts reflecting urgent competition for depleted terminal stocks.

2. The Absolute Price vs. the “Basis” Premium It is important to distinguish between the absolute futures price and the commercial basis.

• London (The Physical Surcharge): Its higher price in mid-2024 reflected a “here and now” emergency. By June 2024, London cocoa reached a high of $11,530 per tonne as European stocks declined by 47%.
• New York (The Quality Premium): While NY’s absolute price eventually hit records (peaking at $12,565 in Dec 2024), it was increasingly viewed as the “Premium Basis” for future-dated contracts. Manufacturers began shifting long-term buying to New York because they knew NY inventory, enforcing stricter bagged-delivery standards, was far more likely to be EUDR-compliant by the 2026 deadline.

3. The 2025 Pivot: Why NY is Now Higher The “Compliance & Quality” logic became the dominant driver of the absolute price once the physical panic subsided.

• Improved Supply: As West African yields recovered in late 2025, the “emergency surcharge” in London vanished. By late 2025, cocoa port arrivals in the Ivory Coast were up significantly compared to the previous disastrous season.
• The EUDR Deadline: As the implementation date approached, formally set for December 30, 2025 for large companies, the market “dumped” legacy London beans. London prices collapsed nearly 60% through 2025, falling faster than New York’s, leading to the NY Premium observed as we enter 2026.

Current Snapshot: January 6, 2026

The “Identity Preserved” era is now reality. The market has fully transitioned from a Physical Crisis (where London was higher due to scarcity) to a Compliance Crisis (where New York is higher due to data integrity). With cocoa re-entering the Bloomberg Commodity Index (BCOM) this month, the “Tier 1” compliant bean has solidified its place as a distinct, premium financial asset.

B. Integration and Refactoring of Origins

The era of spot-buying “anonymous” coffee is ending for major brands. To guarantee the data, you must guarantee the relationship. Expect a wave of Mergers &Acquisitions or deep Joint Ventures where traders effectively become “remote procurement departments” for Fast Moving Consumer Goods, managing the digital twin of the farm as closely as the physical logistics.

Just as a PM refactors code to make it efficient, the market will refactor origins. Sourcing from millions of unmapped smallholders is becoming operationally expensive. We may see a shift toward larger, organized estates or cooperatives that are “API-ready”—capable of pushing data as easily as they push bags.

Shipping the New Version

The volatility of the 2020s is not a cycle; it’s a migration. We are migrating from a trade based on volume and opacity to a trade based on value and visibility.

As a former PM, I’d say the “user requirements” have changed permanently. The consumers (and regulators) demand a product that doesn’t cost the Earth. The market is frantically trying to build the features to deliver it.

The C-suite leaders who treat this as a data challenge (rather than just a sourcing headache) will be the ones who successfully ship the product in 2025 and beyond.

L’article The Bug in the Bean: Why Data, Not Weather, is the New Supply Shock est apparu en premier sur Sustaain.

Sustaain, a data and artificial intelligence platform dedicated to improving data for sustainable supply chains in soft commodities, has announced the expansion of its partnership with Volcafe, one of the world’s leading green coffee merchants.

Following a year of collaboration focused on optimising data flows, Sustaain and Volcafe have entered a new phase with the signing of a multi-year agreement. This partnership will address structural priorities such as data quality, integration, and consolidation to enhance reporting, transparency, and traceability across the coffee supply chain. These efforts aim to establish robust foundations for long-term sustainability and traceability systems at scale.

By investing in advanced data strategies, Volcafe is reinforcing its commitment to operational excellence and compliance, in a sector in which interoperability is often rare.

Raphaëlle Peinado, Sustainability Director, Volcafe

“Strengthening our collaboration with Sustaain helps reinforce the practical foundations of reporting, transparency and traceability across our supply chains. By consolidating and standardising operational data, we’re improving the reliability of the information our teams use day to day and for reporting – ensuring that our roaster partners can access clear, credible data to support their own sustainability commitments.”

Clément Collignon, CEO of Sustaain

“Volcafe is among the companies demonstrating that an ambitious data strategy can profoundly transform the way supply chains are managed. We are delighted to support their teams in deploying robust data-management tools rooted in the operational realities of tropical commodity supply chains.”

This strengthened partnership marks an important step towards more coherent and operational traceability, setting a benchmark for data-driven sustainability in the coffee sector.

Founded to drive sustainability in global trade, Sustaain provides advanced data and AI solutions to build transparent and responsible soft-commodity supply chains. We support EU importers of tropical products and Food & Beverage brands with scalable traceability, compliance and impact measurement tools.

Established in 1851, Volcafe is one of the world’s largest coffee merchants, specialising in worldwide green coffee procurement, processing in origin countries and distribution to roaster partners. Volcafe provides access to all the major coffee producing origins and supplies beans for over 66 billion cups a year. For more on Volcafe, visit www.volcafe.com

Contact:

• Sustaain: Louise Baudry, Growth Product Marketer: louise@sustaain.org
• Volcafe: Marie Renou-Ullrich, Head of Marketing & Communications: marie.renou@volcafe.com

L’article Sustaain strengthens its partnership with Volcafe est apparu en premier sur Sustaain.

Let’s talk today about a topic that moves calmly in the eye of the Gen AI hurricane: document information extraction. While some flashier topics steal the scene, information extraction silently benefits from the exponential progress of LLMs, multimodal models and agentic workflows.

Why document processing is still a tough task

In order to be on the same page, let’s clarify what we call document information extraction. It is a sub-domain of the more general task of information extraction, with a goal to automate the extraction of pre-defined key-value pairs from digital or scanned documents. For example, one may want to automatically extract and store in a database the name of customers and amounts payed from invoices.

Why do we and our clients think that such capabilities matter? While good ol’ documents are losing more and more territory to digital forms and the underlying databases, they still host a big chunk of the world’s data. And right now, we’re sitting at the junction between the paper and digital worlds. More and more data that used to be confined to documents is being dematerialized and structured into machine-readable databases. While this kind of necessary work used to be the burden of humans, we can all agree that reading and manually entering data from thousands of documents into a database is one hell of a time-consuming, brain-rotting task that we humans abhor.

How AI does the heavy lifting for Document Extraction

Such repetitive jobs definitely seem better off in the hands of an algorithm. However until recently, such task used to be too complex to be handled by “classical” algorithms. When you decompose it, the job consists in a mix of computer vision and natural language processing. Computer vision in the sense of Optical Character Recognition (OCR), whether the characters are printed or handwritten, and document layout comprehension such as tables, graphs. And natural language processing in order to understand the semantic content of the information captured by computer vision in order to extract the targeted data. This is why until recently, such tasks were quite burdensome to automate except for some big players that developed solutions on small, unflexible scopes. With the advent of LLMS, computer vision and VLMs, this has changed and made document information extraction much more accessible to everyone with sufficient tech background. Moreover, it made it possible to build tools with a much greater flexibility on extraction tasks, without having to start over every time and build swarms of expensive training/testing datasets.

Overcoming accuracy and consistency challenges

Nowadays, multimodal models, and in particular Visual Language Models (VLMs) allow everyone to just upload a document and asks questions about it. So it it enough to automate the question-answer process in order to automatically extract targeted document information from a collection? Not at all, and here is the reason: LLMs/VLMs answers widely vary in accuracy and completeness.

Even with newer, more specialized models or dedicated wrappers, testing extraction tasks often leads to disappointing results due to inconsistent accuracy and output quality. You will observe large quality disparities when asking the same question across different documents, or different questions of seemingly equal difficulty over the same document. In addition to that, no model can fully grasp the technical terminology or business-specific rules you intuitively apply when reading a document.

If you go one step further and search for more specialist tools, you can find as many “we solved document extraction” tools as there are containers in the world. However, in addition to the extremely variable quality of extractions, their usage is also at best difficult, or quite expensive. Defining specific extraction tasks, modifying them, adjusting details and infusing domain knowledge through prompt engineering is an extra workload where every change might break extraction performance over something unrelated.

Sustaain’s targeted solution to Bills of Lading and other transport documents

Here at Sustaain, we rode the LLM wave and quickly understood that simply waiting for better models would not make the cut. So we focused on one target: extracting fields from Bills of Lading (BLs) in order to bring the best of AI models, agentic workflows, domain knowledge and solve that one problem, but do it well.

For the non-connoisseur, BLs are used by carriers, mostly international container shipping lines, to acknowledge receipt of cargo for shipment. Such documents are at the core of international cargo transport operations, and some of their data often end up being entered manually into many proprietary softwares and databases. Commodity traders and operators typically waste hours manually writing such information into some software interface: transportation dates, ports of loading/discharge, BL and container numbers, detail and quantities of the shipment…

No miracle

When we started developing extraction solutions from available LLMs, we faced that wall of wrongness. Not only do LLMs read things wrong all the time (in particular serial numbers that have no semantic meaning), they often straight up hallucinate answers or do not find them at all.

Add to this scans of terrible quality where even a human being struggles to read information, and your VLM will fail to provide any interesting insight. Trying all the trendy “miracle” extraction softwares that are not much more than some VLM wrapped up with a restrictive prompt and some basic retry mechanism did not help any further.

Intelligence with guardrails

Basically, the rule of thumb is to think of LLMs/VLMs as very distracted interns that do not know the domain specifics yet, and try to read the document way too fast. In order to have those do a proper job, you need someone to micromanage them:

• cut the entire extraction task into separate atomic tasks, one field at a time
• explain the context clearly: what is a BL, what information does it contain; give specific instructions for specific templates with different layouts and vocabulary
• define the task even more clearly: describe exactly what information you want, in what format, where it can be found, provide examples that reflect the diversity of your sources
• enforce strict validation rules: if the output is a date or a container number, it should follow a specific format; if you are searching for the carrier company, provide the list of existing companies and make sure the answer is one of them; apply some consistency checks that allow to detect mistakes from business logic like the weight of the cargo being suspiciously low, or the date being too old
• when one extraction or method fails, do not despair and try another way; different models may perform better or worse on the same document, classical OCR libraries may sometimes shine where VLMs fail, twisting task prompts may change a failure into a success

Packaged and ready to ship

While that would be a lot of work for a human, it can be automated into a streamlined workflow that executes all those tasks without any human intervention. And that is exactly what we developed at Sustaain for BLs and other supply chain documents. This intelligence layer, built on top of high-performance GenAI models, allows us to overcome the shortcomings of those models: errors, hallucinations, lack of business understanding, lack of multi-page capability… Such a control layer allows use to provide extractions that are reliable enough to be used in production contexts with low human-in-the-loop requirements.

Moreover, in order to fit the infinite diversity of needs in information extraction tasks, we developed a nifty task definition interface that allows humans to define their own custom extraction tasks. This interface allows us to define any extraction task with a fine-grained control on task definition, output formatting and quality checks, all in one structured, human-readable text file that will be parsed by our system.

We also developed a large, diverse dataset of annotated document sources that we use to benchmark any new development in terms of global and specific performance metrics (machine learning specialists know that without such global performance-tracking processes, only quickly reaches a threshold where any targeted improvement may lead to catastrophic performance drops in other areas).

Document extraction API for full automation or graphic interface for human-in-the-loop validation

Because no system is perfect, we also developed an API that not only delivers the extraction contents, but also some analysis of the extraction quality. For all extracted content that did not pass a check, was not found in the document, or has a low confidence score, a relevant message will be added to the response so that the recipient may react accordingly.

You may want to use our own user interface to upload documents, visualize quality warnings, and download extraction results. You can also connect your own systems to our API in order to do all those things yourself and fuel your own platforms with the extracted data batches.

Perfection is an ideal but that does not mean that we stop reaching for it. We keep refining our extraction workflows by enriching the control layer with smarter decision-making agents, testing newer models, and adding more business-relevant checks and rules. So if you are tired to copy some document data into your proprietary systems, feel free to contact us!

L’article Automate Document Extraction with AI – Bills of Lading and More est apparu en premier sur Sustaain.

2025 feels like Thriller on repeat: everyone’s dancing, no one’s sleeping, and somewhere in the background, venture capital just hit another high note. AI startups have soaked up over half of global VC flows since January. Valuations moonwalk upward; due diligence quietly exits the room. Founders talk about “planetary intelligence,” though many still struggle to make payroll.

We’ve seen this choreography before. It starts with rhythm and ends with rigor mortis.

At Sustaain, we prefer a slower groove — one with bass, breath, and balance. We still believe in planetary intelligence too, but of the tangible kind: the one that pays invoices and fixes supply chains. In an era where funding behaves like fog, we prefer liquidity to illusion.

Cash, after all, keeps the beat.

1. Discipline over drama

The AI bubble is our era’s pop anthem: catchy, loud, and destined to date. Every cycle needs its chorus of overconfidence. But in business, as in music, silence often says more.

When your only amplifier is your own cash flow, you learn the discipline of timing and the economy of silence. You build fewer features, but you play them right. You listen harder. You find rhythm in constraint.

That’s where product-market fit lives: not in vision decks but in contracts signed. Every paying client is a note of truth. The market doesn’t care for crescendos. It rewards persistence. The founders who learn to breathe between beats — to manage burn, to delay gratification — are the ones who still stand when the lights come back on. Michael Jackson knew it: mastery is repetition without fatigue.

Capital efficiency is the new growth. And the next great exits won’t come from hyper-funded rockets but from well-oiled engines that keep their rhythm.

2. Precision in motion

Liquidity is not constraint; it’s choreography. It’s how you move through uncertainty without losing tempo.

We watch cash flow like Richter watched color — in layers, transparencies, and contradictions. His paintings are never just abstraction; they’re control disguised as chaos. That’s how we treat liquidity — not as static reserve but as living pigment.

We run operations like a cockpit: fuel, altitude, visibility.

R&D happens only when two conditions are met:

1. It solves a client’s real pain,
2. It aligns with the megatrend.

Otherwise, it waits its turn. It’s not timidity; it’s choreography again. Knowing when to step, when to glide, and when to stand still.

Because beyond the daily noise – be it the tariffs, elections, and ESG whiplash – the megatrends remain unchanged: traceability, accountability, and credible AI for sustainability. These are not headlines; they’re tectonics.

By staying aligned with those, we can improvise freely without losing structure. The painter Gerhard Richter once said,

“ I blur things to make everything equally important and equally unimportant. I blur things so that they do not look artistic or craftsmanlike but technological, smooth and perfect. I blur things to make all the parts a closer fit. Perhaps I also blur out the excess of unimportant information.”

We build similarly: balancing focus and ambiguity, precision and patience, the present invoice and the future law.

3. The virtue of scarcity

Schopenhauer believed that the world is driven by Will: an endless striving, often irrational, but undeniably alive. Bootstrapping is a daily proof of that will. When resources tighten, instinct sharpens. You learn what truly matters: which problems to solve, which ambitions to delay, which truths to face. Scarcity rewards transparency because we simply don’t have time for fake news. It’s uncomfortable, but it’s also deeply clarifying. And ironically, it makes a company more bankable.

Because investors know what philosophers do: that endurance outlasts enthusiasm. When the hype subsides and valuations deflate, fundamentals inflate. The startups worth funding will be those that proved they could live without it. Seven years from now, when today’s megafunds look for exits, the survivors will be those who built with constraint and conviction — companies that turned liquidity into art.

We stand in that liminal space: where cash flow meets culture, where sustainability is not a slogan but a system. We’re not building a myth; we’re building an organism. One that learns, breathes, and lasts.

Conclusion — In liquidity we trust

Every age gets the anthem it deserves.

Ours just happens to come with a synth beat and an AI chorus.

But when the music fades, those who stayed grounded in rhythm — who practiced restraint, precision, and will — will still be dancing.

Grow deliberately.

Spend consciously.

Play the long game.

Cash is king. Impact is queen. The rest is choreography.

References

Sustaain Blog — For a Grounded Intelligence: AI God vs. Machine Learning (2024).

Reuters — AI venture funding continued to surge in third quarter, data shows (2025-10-06). VC funding hit US$97 billion in Q3 2025.

Gerhard Richter — Text: Writings, Interviews & Letters 1961-2007, Thames & Hudson, ISBN 978-0500093467.

Arthur Schopenhauer — Le Monde comme volonté et représentation, Bibliothèque de la Pléiade n° 682, Gallimard (2025 édition).

L’article Cash is King — A Thriller in Three Acts for the AI Age est apparu en premier sur Sustaain.

A year later, we’ve seen enough dashboards and declarations to know that progress isn’t measured in slides, but in soil. On our way to Produrable, we write this as a declaration of intent: to stay grounded, connected, and unafraid to speak out — even when the echo feels too loud.

Because there’s irony in all of this.
Irony in traveling to sustainability summits.
Irony in preaching sobriety from the stage.
Irony in building data clouds while longing for roots.

But silence won’t help either.
So we speak — not to dominate the conversation, but to shift its rhythm.

Data as the New Ecology

The world’s sustainability crisis is not just ecological — it’s informational.

We operate in silos: ministries, NGOs, corporates, traders, scientists. Each collects fragments of reality, but few connect them. Data is abundant — meaning is scarce.

At Sustaain, we’re assembling a global database of organizations linked to deforestation. We connect FAO and World Bank datasets with satellite and land-use data, and with corporate data from local cooperatives, communities and households — to reveal invisible risks in supply chains of coffee, cocoa, and other soft commodities.

It’s not about big data. It’s about useful data — data that helps farmers, buyers, and regulators see the same forest.

Recent studies show that environmental and emissions data remain fragmented, inconsistent, and often inaccessible across regions and institutions (Beck et al., 2025). Connecting them is not a technical luxury — it’s a condition for coherent action.

When data behaves like an ecosystem — open, diverse, self-correcting — sustainability becomes something we can actually manage, not just report.

The Problem with AI: The New God

AI has become the new religion. It promises transcendence, efficiency, and salvation — if only we surrender enough faith (and data). But AI is not divine. It’s probabilistic. It operates on uncertainty, not absolutes — a fluid technology too often locked into rigid, vertical deployments, and all too often today misused for greater power and ever greater intensity.

We’re told there’s “AI for ESG,” “AI for compliance,” “AI for supply chains.” In reality, none of these worlds stand apart. Nature, trade, and governance don’t obey those boundaries.

That’s why we prefer machine learning as network, not oracle.
Research in distributed and energy-harvesting systems shows how intelligent networks can learn collectively across constrained environments, improving resilience while reducing environmental cost (Güler & Yener, 2021).
The forest outlives the tower. The network is wiser than the god.

So let’s stop worshipping AI. Let’s use deep learning to listen — to adapt, to doubt, to collaborate. Intelligence should flow horizontally, like mycelium — not beam down from a server rack.

Ô Guérillières – Toward a Techno-Feminism

If the first wave of technology was about control, the next must be about care.

A techno-feminist perspective isn’t decorative; it’s structural. It resists extraction and domination. It values empathy, plurality, attention — not conquest. We need slow companies that grow like rhizomes (Deleuze would approve): non-linear, adaptive, decentralized. This isn’t poetry; it’s systems thinking. The only way to make global supply chains resilient is to let intelligence circulate, not concentrate.

Progress, then, becomes a form of listening — digitizing a bill of lading, validating a farm polygon, harmonizing KPIs across geographies — all small, tangible acts that restore coherence in the noise.

Stay Grounded

Data is not power — connection is.
AI is not wisdom — attention is.
The future is not vertical — it’s rhizomatic.

We carry our paradoxes: screens and trains, voice and doubt, ambition and restraint. But acknowledging contradiction is not weakness — it’s consciousness.

To be sustainable is to stay in contact — with others, with earth, with uncertainty.

So keep questioning. Keep connecting.
And whenever possible — take the train.
It keeps you close to the ground.
And that’s where change begins. 🌱

References

• Beck, M. W. et al. (2025). Addressing data gaps in sustainability reporting: A benchmark dataset for greenhouse gas emission extraction. Scientific Data (Nature). https://www.nature.com/articles/s41597-025-05664-8

• Güler, B., & Yener, A. (2021). Energy-Harvesting Distributed Machine Learning. arXiv:2102.05639. https://arxiv.org/abs/2102.05639

• “Le Monde Nous Appartient” by Draga (Ô Guérillières, 2025)

• “God Only Knows” by The Beach Boys (Pet Sounds, 1966).

L’article For a Grounded Intelligence est apparu en premier sur Sustaain.

A Market That Touches Billions

Coffee is not just a beverage—it’s one of the world’s most valuable and widely traded agricultural commodities. The global coffee market is estimated at over $200 billion annually, with 170+ million bags produced each year.

More than 12.5 million farming households grow coffee across 70 countries, the majority of them smallholders. Yet the way value is distributed along the chain is strikingly unequal:

• Producers capture less than 10% of the final retail value.

• Exporters, traders, and roasters dominate margins, with major international players like Nestlé, JDE Peet’s, and Starbucks shaping global demand.

• Consumption is concentrated in Europe and North America, while production is centered in Latin America (Brazil, Colombia, Honduras), Africa (Ethiopia, Uganda), and Asia (Vietnam, Indonesia).

This combination of global reach and highly structured trade—organized around exchanges like ICE in New York and LIFFE in London—means that innovations tested in coffee can scale rapidly across geographies.

A Sector Under Pressure

Coffee production sits at the intersection of some of the world’s most pressing sustainability challenges:

• Deforestation: Expansion into forested areas remains a key concern, particularly in Latin America and Southeast Asia.

• Climate vulnerability: Rising temperatures and shifting rainfall patterns threaten yields and farmer livelihoods.

• Fragmented data: Despite decades of sustainability initiatives, from voluntary certifications (Fairtrade, Rainforest Alliance) to industry-driven commitments (through the Global Coffee Platform or ICO’s Coffee Public-Private Task Force), the sector still lacks consistent, reliable, and scalable data.

Regulation adds a new layer of urgency. In the context of the EU Deforestation Regulation (EUDR) implementation, Importers and traders now need robust, verifiable data on origin and land use—a demand that current systems struggle to meet.

Where We Saw the Opportunity

Coffee’s scale and structure make it a proving ground for innovative approaches to supply chain data. We saw the chance to build solutions that could have immediate impact for both regulators and industry.

Our current focus is twofold:

1. A Global Coffee Detection Model
   Using satellite imagery and probabilistic methods, we are developing a consistent baseline for identifying coffee production areas. Unlike fragmented national datasets, our model provides uniform coverage, enabling actors across the value chain to work from the same reliable reference.

2. A Flexible Supply Chain Model
   We are designing a comprehensive yet adaptable framework that can support compliance with EUDR and other emerging regulations—while recognizing the realities of complex supply chains. Our model balances traceability with feasibility: strong enough to satisfy regulators, but pragmatic enough not to block entire flows of coffee that are vital to livelihoods and markets.

Rapid Impact, Lasting Change

By targeting coffee first, we can demonstrate what’s possible: an approach that combines global detection with supply chain modeling to create actionable, regulator-proof data. The benefits extend beyond compliance:

• Operational efficiency: less time lost chasing incomplete supplier declarations.

• Bargaining power: stronger positions in negotiations with downstream buyers.

• Competitive edge: the ability to meet regulatory and customer expectations ahead of the curve.

And while coffee is our entry point, it won’t be our endpoint. The tools we’re building—scalable detection models, flexible supply chain frameworks—are designed to apply across other high-risk commodities. Success in coffee will open the door to cocoa, palm oil, soy, and beyond.

Coffee First, But Not Coffee Only

The coffee sector is already supported by influential organizations such as the International Coffee Organization (ICO), the Global Coffee Platform (GCP), and national sustainability programs. By aligning with these efforts and adding the missing piece—consistent, scalable data—we believe we can accelerate sector-wide progress.

Coffee matters because it touches billions of people, from farmers to consumers. By starting here, we’re showing that better data isn’t just a technical solution—it’s a lever for systemic change.

L’article Why We Chose Coffee est apparu en premier sur Sustaain.