Optimizing Inputs for Mapping Agent

Why inputs matter

Mapping Agent (formerly known as Autopilot) matches your text descriptions to emission factors in Climatiq’s database. The quality of that match depends almost entirely on the quality of the description you provide. A clear, specific description gives Mapping Agent the context it needs to find the right factor. A vague description leaves too much open to interpretation.

You will see this reflected in your quality labels: well-described inputs tend to produce stronger matches. That said, quality labels reflect the fit between your input and the available emission factors. A well-described activity can still return review if the database has limited coverage for that activity type.

If you are seeing a high proportion of review labels across your dataset, there are two likely causes: input descriptions that are too vague or ambiguous, or limited data coverage for your activity type. Improving your descriptions is a good first step. If review rates remain high after that, the issue may be coverage rather than input quality.

What makes a good input

A good input description is:

• Specific: it names the actual activity, material, or product rather than a broad category or internal reference
• Human-readable: it uses natural language rather than SKU numbers, supplier codes, or abbreviations that only make sense in your internal systems
• Contextual: include relevant details that help clarify the activity, such as the material, the type of use, or the product category. This can work in both directions - narrowing an ambiguous match, or making an unclear product name or internal code identifiable to the model.

Examples: good and poor inputs

Practical tips

Clean up invoice line items before processing. Invoice data often contains the right information but in a format Mapping Agent cannot easily interpret, such as codes, abbreviations, or composite descriptions. A short clean-up step, replacing codes with plain-language descriptions, can significantly improve match quality across your entire dataset.

Include activity context, not just quantities. Mapping Agent uses your description to identify the right emission factor, not to calculate the final footprint. Quantities and units in the description do not improve matching. What helps is the nature of the activity: what it is, what it is made of, and how it is used.

Split mixed descriptions into separate entries. If a single line item covers multiple activities, for example, “travel and accommodation”, split it into two separate inputs. Each description should represent one distinct activity type.

Test a sample before processing your full dataset. Run a small, representative sample of your data through Mapping Agent and review the matches and quality labels. This tells you where your inputs need improvement before you commit to processing everything. If you have multiple categories or commodity groups in your data, experiment with different input combinations to yield optimal results.

Use your supplier or product names as a starting point. Even if the exact name is not in Climatiq’s database, a supplier product description is usually a much better starting point than an internal code. Translate it into plain language if needed.

If you are processing large volumes and speed matters more than the highest possible match quality, Mapping Agent can be configured to run faster with some trade-off in accuracy. Test both settings on a sample before processing at full scale. See the Mapping Agent API reference for how to configure this using the effort parameter.