Rebuilding Marine Parts Discovery and Order Intake Around AI

Industry
Manufacturing / Retail & E-commerce
Overview
The client is a US marine hardware manufacturer supplying rub rail, trim, and deck hardware to boat builders across the East Coast and to the aftermarket refit community. With a catalog spanning thousands of SKUs and a product line where fit depends on hull length, profile, material, and finish, the company found that growth in catalog depth was outpacing the ability of its storefront and service team to guide customers through it.
Challenge
Shoppers arriving at the site knew their boat but not their part number. Conventional faceted search could not answer the one question most of them came with, so a large share of commercially meaningful traffic converted into support conversations instead of orders. On the order-intake side, purchase orders arrived as PDF attachments inside Salesforce cases and were transcribed and policy-checked by hand. Off-the-shelf document automation platforms were priced for enterprise throughput the client did not have.
Key objectives included:
- Plain-language fitment Let customers describe their boat in plain language and receive a correctly sized, compatible recommendation.
- Accessories in the same interaction Surface compatible accessories and sample orders inside the same interaction.
- AI purchase-order extraction Replace manual purchase-order transcription with AI extraction and human review.
- Order validation rules Validate every incoming order against pricing, lead-time, part-status, and tax rules before it reaches the ERP.
- Stay inside Salesforce cases Keep the entire workflow inside the agent's existing Salesforce case, not a separate tool.
- Replatform support Support a replatform from nopCommerce to Salesforce Commerce Cloud.
The Solution
Sigma Solve designed two connected AI systems built around how the client's customers shop and how its service team works, rather than around generic commerce and document tooling.
Key initiatives included:
- Conversational product discovery An assistant embedded in the storefront product experience.
- Fitment reasoning Converts hull dimensions into correct coil length, including finishing allowance.
- Vector knowledge base Spans product data, installation kits, technical manuals and historical customer interactions.
- Intent-based search With substitution logic and compatible-accessory cross-sell.
- Image-led discovery roadmap Identifying boat make, model and colour from a customer photo.
- AI document extraction For purchase orders, with field-level confidence scoring.
- Side-by-side review interface Click-to-source traceability and correction-as-training feedback.
- Business-rule validation layer Configurable rules covering pricing, lead times, part status, shipping method, and tax jurisdiction.
- Unattended ingestion Via monitored mailbox and FTP, with output in the client's preferred data format.
- Native Salesforce embedding Inside Salesforce cases, with AI-drafted customer responses and human approval.
Outcome
The client moved from keyword search that could not answer fitment questions to an assistant that reasons about the boat, and from manual purchase-order transcription to AI extraction with targeted human review. Order validation shifted from institutional knowledge held by individual agents to a codified rule layer applied consistently to every document, with service agents staying inside the Salesforce case they already work in.