Forward Deployed Engineer

About NegotiateAI (Menlo Ventures Backed)

Today we're building AI for category managers at manufacturing and industrial companies, taking the grind out of indirect spend and delivering enterprises millions in savings.

That's the starting point. Where we're going is general procurement AI that spans the entire function.

About this role

Procurement data is a mess, and every enterprise's mess is different. Part numbers don't match, supplier names are spelled four ways, and the ERP is a decade of customizations on top of SAP. Our platform has to work inside that reality on day one, not after a six month cleanup project.

You'll be the first Forward Deployed Engineer at NegotiateAI: embedded with our earliest customers, building until the product delivers measurable savings against their real data, then pushing what you learned back into the core platform. This is an engineering role, not a solutions or consulting role. You write production code and you own what you ship.

What you'll do

• Embed with customer procurement and IT teams to map how spend actually flows: intake, approvals, category taxonomies, supplier master data, and where the savings are hiding.

• Define crisp success metrics with the customer before you build. Realized savings, categorization accuracy, cycle time, coverage. Then instrument them so nobody has to argue about whether it worked.

• Build integrations into customer systems: SAP, Ariba, Coupa, Oracle, NetSuite, data warehouses, SSO and identity, and the inevitable SFTP drop. Navigate read-only versus write-back postures and design for whichever one the customer will actually approve.

• Go from a raw spend extract to a working pipeline fast: entity resolution on suppliers, part identifier normalization, category mapping, and ground truth you can defend in a room full of skeptics.

• Model customer spend in ClickHouse, which backs our analytics and benchmarking layer, and keep queries fast as datasets grow from one customer's category to an enterprise's full indirect spend.

• Customize the agentic workflows at the core of the product for each deployment: retrieval sources, tools, prompts, and guardrails, then harden them for reliability and observability.

• Build per-customer evals against their own data so we can prove accuracy, catch regressions, and show progress week over week.

• Drive adoption. Run training, write runbooks, and hand off something durable to a champion inside the account instead of becoming the single point of failure.