Gains:
- Ability to establish Scope 3 by prioritizing suppliers based on spend and risk level and distinguishing between actual data and proxy estimates
- When flagging supplier risk signals with artificial intelligence, treat them as 'signals to be examined' rather than as evidence
- Ability to manage serious human rights/environmental claims through a due diligence process requiring field inspection and legal confirmation
For most companies, the biggest part of their sustainability impact lies not in their own factory but in their supply chain. More than 80 percent of a clothing brand's carbon footprint can come from its suppliers, ranging from fabric manufacturers to shipping. Likewise, the most severe social risks, such as child labor, forced labor and environmental violations, often occur deep in the supply chain. In this unit, you will learn how to use artificial intelligence (AI) to collect and evaluate supply chain data and screen for risks.
Let's clarify two concepts. Scope 3, as we saw in the previous unit, is the indirect emissions in the company's value chain and is generally the largest but most difficult to measure item. Due diligence is the process of identifying, preventing and remediating human rights and environmental risks in a company's supply chain. Regulations such as the EU's CSDDD (Corporate Sustainability Due Diligence Directive) make this mandatory for large companies.
The main challenge of supply chain data
You can measure your own facility's data; But you cannot directly measure the data of hundreds of suppliers, you have to ask them. This creates three problems:
- Data does not arrive or is incomplete. Most smaller suppliers do not keep carbon data.
- The data is unreliable. When a supplier self-reports, there may be overstatement or underreporting.
- A guess (proxy) is required. If actual data is not available, industry average or expenditure-based estimate is used.
This is where AI helps in two ways: (1) quickly drafting surveys and follow-up letters to be sent to hundreds of suppliers, (2) structuring scattered responses and flagging risk signals. But no number given by the supplier is verified, no risk signal is concluded without human review.
Scope 3 category (example)
Data source
Typical method
Purchased goods/services
Supplier survey / spend
Real data > proxy
Shipping and distribution
Carrier records
Distance × load × factor
business travel
travel system
Km/flight × factor
Use of the product sold
Product energy data
Usage model
waste
Waste transfer documents
Quantity × factor
Tip: In Scope 3, the “spend-based” estimate (this much CO₂e per $1,000 spent) is a quick start, but a rough one. When a supplier provides real data, use it; The spend-based estimate is only provisional if there is no actual data and should be marked as "estimate/proxy" in the report.
Step by step: Supply chain workflow with AI
1. Prioritize suppliers. You cannot examine all hundreds of suppliers in the same depth. Focus on those with high spend size and risk (sector, country). AI can produce a draft list with this prioritization.
2. Draft survey and request. Ask AI for a clear, easy-to-fill out supplier data survey outline.
3. Configure the incoming data. Put the answers in different formats into a single table; Mark the missing parts.
4. Scan for risk signals. Have the AI find “red flag” candidates in public news, breach records, and survey discrepancies; But humans verify every signal.
5. Action and pursuit. Establish a corrective action and follow-up plan for high-risk suppliers; Document the process.
Weak prompt / Strong prompt
Weak prompt:
Calculate the carbon footprint of our suppliers.
No data source, no real/proxy distinction, and no verification; AI makes up supplier numbers.
Powerful prompt:
Your role: supply chain sustainability analyst. Input: 15 suppliers' names, industry, annual spend amount, and the actual emissions data they provide (if any). Task: 1) For those who provide real data, use that data. 2) For those who do not have real data, write "spend-based PROXY required", DO NOT FIT the number; specify which factor to get from official source.3) Prioritize suppliers in high spend + high risk industry.Output: table (Supplier | Industry | Spend | Data type[actual/proxy] |Priority | Note). Proxy numbers will be marked in the report.
Prompt for supplier risk screening
Set up a risk screening scaffold for the following list of suppliers. Assess for each supplier (only based on information I have provided, public claim FAKE):- Possible human rights risk from country/sector (high/medium/low)- Possible environmental risk- Inconsistency in survey response/red flagsRule: add note “verification required — no source” for each risk; do not make definitive accusations, just suggest signals to investigate.List: [SUPPLIERS]
The following prompt drafts a data request survey to send to suppliers.
Your role: sustainability sourcing specialist. Task: draft an ESG data survey of no more than 10 questions that small/medium suppliers can easily complete. Include: energy consumption, fuel type, waste, number of occupational safety accidents, commitment to forced/child labor. Tone: simple, clear, easy to complete; Add units and examples to each question.
Caution: Do not have AI verify a risk signal about the supplier (e.g. "child labor") and never use a claim generated by AI as evidence. Such grave claims are only confirmed by reliable source, field inspection and legal process. AI only produces a “signal to be examined”; does not judge.
three mini cases
Case 1 — Honest use of proxy. Only 30 of one furniture manufacturer's 200 suppliers provided actual carbon data. The team used a spend-based proxy for the remaining 170, but the report clearly states, "65% of Scope 3 is a spend-based estimate." The auditor acknowledged this honesty; had it been concealed, the assurance might have been denied.
Case 2 — Prioritization. A food company had 480 suppliers. AI showed that 80% of the spend was concentrated in 40 suppliers. The team focused the deep dive on these 40; resources were not wasted and the highest impact suppliers were reported with real data.
Case 3 — Risk of false accusation. AI flagged a supplier as “possible forced labor”; The reason was only the national average. The team accepted this as a "signal to be examined", not evidence, and requested a field inspection. The inspection found no problems; If the AI's signal was taken as evidence, an unfair accusation and legal risk would arise.
Common mistakes
- Accepting supplier data without verifying it. Self-reporting supplier data requires cross-checking.
- Hiding the proxy. Using spend-based forecasting is fine; Not specifying is a problem in auditing.
- Examining all suppliers equally. Sourcing should focus on high spend and high risk suppliers.
- Mistaking the AI signal for evidence. Claims of human rights/environmental violations are verified by the field and the law, not by AI.
- Skipping Scope 3 entirely. The biggest impact is usually here; It cannot be ignored on the grounds that it is "difficult to measure".
In summary
The supply chain is where most companies' biggest environmental and social impact hides; Scope 3 emissions and due diligence risks are concentrated here. AI; It is a powerful aid in supplier prioritization, survey drafting, structuring scattered responses and risk signal flagging. But supplier data is not included in the report until it is verified, proxies are clearly stated, and serious allegations are confirmed by field inspection. AI generates signals; The human being assumes the judgment and responsibility.
Application task
Prepare a list of 10 fictitious suppliers with name, industry, and annual spend; Add actual emissions data to a few, leave most blank. Ask the AI for a table of prioritization and data type via the powerful prompt above. Then: (1) verify that the number of suppliers requiring a proxy is not made up, (2) select the 3 highest priority suppliers with justification, (3) draft a short data survey to send to them.
checklist
- [ ] I prioritized suppliers based on spend and risk level.
- [ ] I used real data ones; For those who don't have it, I marked the proxy.
- [ ] I evaluated the supplier data not as is, but by cross-checking.
- [ ] I considered the risk signals "to be examined" and did not use them as evidence.
- [ ] I planned a field audit/legal verification step for serious claims.
- [ ] I did not skip Scope 3; I stated the scope and estimation rate in the report.
- [ ] I recorded the entire process with a due diligence document.