You Can't RECODE a Blind Spot

Why procurement's AI layer is only as good as its signal layer

The short answer: most procurement teams do not lack data. They lack timely, filtered, external signal. Internal systems capture spend, contracts and Tier 1 supplier records well. But the disruptions that actually halt production originate outside the enterprise: in a Tier 3 supplier's factory, in a local-language news report, in a regulatory filing or court docket that no internal system indexes. Layering AI agents on top of that gap does not close it. It reaches the blind spot faster.

‍

DPW named the problem. Most of the field is solving the wrong half of it.

DPW's theme for its 2026 Amsterdam edition is RECODE, and the framing is unusually blunt for a conference theme: organisations keep layering AI on top of broken processes, fragmented data and outdated operating models, producing a great deal of activity and very little progress.

The diagnosis is correct. But walk the floor at RAI Amsterdam on 30 September and nearly every solution on display will be recoding the same thing, the decision layer. Agents. Orchestration. Workflow automation. Approval routing. Intake management.

Very few are recoding the sensing layer: the question of what those systems are actually looking at.

This distinction matters because the two layers fail in opposite ways. A weak decision layer produces slow decisions, which are visible and therefore fixable. A weak sensing layer produces fast, confident, well-formatted decisions that happen to be wrong and those are invisible until the shipment doesn't arrive.

Procurement has spent five years automating the decision layer on top of a sensing layer that has barely changed since the supplier questionnaire.

‍

The two failure modes procurement teams actually describe

Ask procurement and supply chain leaders what is wrong with their risk tooling, and the answers cluster into two complaints. Neither of them is "we need more automation."

1. The signal arrives late.

Risk alerts land after the disruption has already reached the production line. At that point the system is not performing risk management; it is performing incident reporting. The team learns what happened, not what is about to.

2. The signal arrives unfiltered.

Where feeds do exist, they deliver volume rather than prioritisation – thousands of undifferentiated alerts, no severity weighting, no context, no relationship to the specific supply base that matters. The filtering work gets pushed back onto a procurement analyst who now spends their week triaging notifications. That is precisely the work the technology was purchased to remove.

Both failures share one root cause. The systems are watching the wrong perimeter. They watch what the enterprise already knows about itself, at the tier where the enterprise already has contracts.

What visibility below Tier 1 is actually worth

Lead time is the only real currency in supply chain risk, and unlike most things in this category it is measurable. Three examples from our own monitoring:

‍

‍

‍

Four observations about that table.

None of these signals originated in an ERP, a spend cube, or a supplier questionnaire. Three of the four surfaced first in a language most Western procurement teams do not monitor. All four were entirely public at the time of detection — no privileged access, no proprietary feed, no insider. And in every case the affected downstream organisations had the budget, the tooling and the mandate to act. What they did not have was the signal.

That is the sensing layer gap in one sentence: this is not a prediction problem, it is a reading problem.

‍

What a functioning sensing layer requires

Four criteria, ordered by how frequently they are missed:

1. Language coverage that matches your actual supply base. If Tier 3 exposure sits in Vietnam, Bangladesh and Türkiye and monitoring is English-only, the visibility is decorative. Semantic Visions monitors sources across various languages, which is why the Bangladesh and Czech signals above were detectable at all.

2. Breadth beyond mainstream media. Mainstream outlets report disruptions once they are already disruptions, that is the definition of news. Early signal lives in regional press, trade publications, local government notices and regulatory filings. We process 2 million sources a day from more than 270,000 monitored domains.

3. Relationship mapping, not just entity monitoring. Knowing that a company had a fire is useless information unless you also know it supplies your supplier's supplier. Risk relevance is a function of the graph, not the event. We map relationships across more than 10 million companies and beyond five tiers of connection.

4. Severity weighting at the point of ingestion. Filtering must happen before the alert reaches a human, across a classified taxonomy, in our case more than 720 distinct event types. A tool that forwards everything has not solved the problem; it has relocated the workload and added a licence fee.

‍

The uncomfortable implication for procurement AI programmes

If the four criteria above are not met, the sophistication of the decision layer is close to irrelevant. An agent reasoning over incomplete external data will produce fluent, well-structured, timely recommendations built on a supply base it can only see one tier deep. It will do this quickly and at scale, and nothing in the output will indicate that anything is missing.

This is the specific risk that RECODE, taken seriously, should surface: automation applied to an incomplete input does not fail loudly. It fails silently, and it fails faster than the manual process it replaced.

The order of operations matters. Fix what the system can see before optimising what it does with what it sees.

‍

About the DPW Awards 2026

Semantic Visions has been shortlisted for the Growth Stage Award at the DPW Awards 2026 – the annual programme run by DPW in partnership with Deloitte, recognising technology companies reshaping procurement and supply chain. Finalists are announced on 7 September 2026, with winners announced on stage at DPW Amsterdam, held 29 September – 1 October 2026 at RAI Amsterdam.

Semantic Visions has been building open-source intelligence infrastructure since 2011. More than 400 customers use svEye signal today, including through integration with SAP Ariba.

‍

Frequently asked questions

What is the difference between the sensing layer and the decision layer in procurement AI?

The decision layer covers agents, workflow automation and orchestration, systems that act on information. The sensing layer is what supplies that information. Most enterprise procurement AI is built on internal data such as spend, contracts and supplier master records, and therefore inherits the organisation's existing blind spots regardless of how sophisticated the decision layer becomes.

‍

Why do multi-tier supply chain risks go undetected?

Because visibility below Tier 1 is typically assembled manually and refreshed infrequently, and because the earliest indicators of Tier 2 and Tier 3 disruption appear in local-language sources, regional media and public filings that fall outside the scope of most monitoring tools.

‍

How much advance warning is realistically available before a supply chain disruption?

It varies by risk type. Physical disruptions such as flooding or industrial fire have surfaced in local reporting weeks ahead of downstream impact — 22 days in the case of the 2024 Novelis flooding event. Financial distress typically shows detectable signal patterns over much longer horizons, in some cases more than a year before formal insolvency proceedings.

‍

Can public LLMs be used for supply chain risk monitoring?

Not reliably on their own. General-purpose models are trained on data with a fixed cutoff and do not continuously ingest regional, local-language or trade sources at the volume required for early detection. They can reason well over signal they are given; they are not a substitute for the sensing layer that produces it.

‍

When and where is DPW Amsterdam 2026?

29 September – 1 October 2026 at RAI Amsterdam, with the main conference programme running 30 September – 1 October. The 2026 theme is RECODE.