§0 — The one line

The goal isn’t to become an SFDC/SAP admin. It’s to click through the same case, return, warehouse-receipt, and dashboard screens NVIDIA’s people live in, so the process map stops being abstract and the pain points become things you’ve physically felt. Salesforce is free and takes an afternoon; SAP you can read in an afternoon and drive in a day if you decide it’s worth it. Everything below maps to a numbered step in the RMA process map and the pain point (P1–P23) it illuminates.

A grounding caveat, so you don’t over-trust the labels below. Exact Trailhead module titles and SAP Best Practices scope-item codes drift between releases. Where I name one, treat it as a search term to confirm in the tool, not a guaranteed string. I’ve flagged the ones I’m inferring. NVIDIA’s actual editions/versions are open questions (see process map §3) — this is a generic environment to learn the workflow shape, never their config.


§1 — Setup (one-time, ~30 min)

#Do thisWhereNote
S1Create a free Developer Edition orgdeveloper.salesforce.com/signupPermanent, full-featured, never expires. Your SFDC sandbox.
S2Create a Trailhead account + spin up a Playgroundtrailhead.salesforce.comGuided modules with a throwaway org attached. Use for the guided trails.
S3Bookmark SAP Best Practices Explorer / Process Navigatorsearch “SAP Best Practices Explorer”Read-only process test scripts. No install. This is Tier-1 SAP.
S4(Optional, defer) Register for SAP Cloud Appliance Librarycal.sap.comOnly if you decide to drive SAP hands-on (§5). Deploys S/4HANA to your own AWS.

Do S1–S3 before session 1. Skip S4 until you’ve decided the appliance is worth the cost/time.


§2 — The lab, mapped step-by-step to the RMA process map

Read the middle column as “the exercise,” the right column as “why — what you’ll feel.” 🅝/🅒/🅒🅜 side-markers carried from the process map.

Salesforce half — the front of the flow (steps 2–5) + the dashboard NVIDIA actually asked for

Map stepHands-on exerciseSandboxWhat it lets you feel
2 🅒 Customer opens RMA via portalBuild a Case in Service Cloud; stand up a minimal Experience Cloud portal page that lets an external user file oneSFDC Dev org / PlaygroundThe “front door.” Note how much structure the case form imposes — then recall P14: “you don’t want xAI or OpenAI to fill the form.” You’ll see exactly what NVIDIA is asking hyperscalers to do.
3 🅝 Three-gate approval gauntlet (case mgmt → quality → finance)Build a multi-step Approval Process (or a Flow with approval steps) on the Case; add serial-number + entitlement fields to checkSFDC Dev orgThis is the <1%-rejection gauntlet made literal. Building three sequential gates yourself makes P23 (pure latency, not a filter) obvious — and shows why Greg’s “auto-fast-track known-defect batches” is a routing rule you could actually configure.
4 🅝 Disposition: standard RMA vs ARMAExplore Salesforce Order Management → Returns / RMA; create a Return Order; branch standard-vs-advance-replacementTrailhead trail: search “Salesforce Order Management” + “Returns”Advance Replacement is the whole “two decoupled clocks” story. Creating a return order that ships a replacement before the defective unit returns is the ATC promise from the four-terms primer §4 in software form.
5 🅝 Ship replacement (DN, tracking #, serials — today via email)Generate a fulfillment/delivery record off the Return Order; note every field (ship-to, serials, tracking) that Order Mgmt captures in-systemSFDC Dev orgThis is where P1 (the spreadsheet cascade) begins in real life. Seeing Salesforce hold these fields natively makes the pain concrete: NVIDIA drops to email exactly here instead of letting the system carry it.
P7/P8 cluster-2 dashboardBuild Reports + a Dashboard on open Cases/Return Orders (count, aging, day-31 flag); add a Flow that fires an alert past SLATrailhead: “Reports & Dashboards for Lightning Experience” + “Flow Builder”This is the thing Lonny + Greg sketched at the end of the meeting (“we don’t know the score of the game”). Building even a toy version shows how close the platform already is to their ask — and where the missing data (from the SAP half) blocks it.
(context)Skim Manufacturing Cloud (sales agreements, account-based forecasting)Trailhead: “Manufacturing Cloud”Optional. NVIDIA is a manufacturer; this is the industry-cloud framing of demand/commit. Light touch.

SAP half — the back of the flow (steps 8–16)

Tier 1 (read, no system): pull these Best Practices test scripts and follow the click-path on paper. Names are descriptive — confirm the exact scope-item code in the Explorer (⚠︎ code not verified = don’t quote the code until you’ve looked it up).

Map stepRead this processSAP moduleWhat it lets you feel
8–9 🅒🅜 Defective ships back, ASN should fire; CM receivesInbound Delivery / ASN processing; EDI inbound ⚠︎ code not verifiedEWM (inbound) + IDoc/EDIP2 made concrete: the ASN is a standard, boring, solved SAP document. Reading how routinely it’s supposed to flow shows how far “email/paper” is from baseline.
10 🅒🅜+🅝 Disposition: reman / repair / refurbAdvanced Returns Management (ARM) — returns receipt, inspection & disposition ⚠︎ code not verifiedSAP ARMARM is SAP’s RMA module. Its inspection-code → follow-up-activity logic is precisely the three-category routing Greg described. This is the single most on-point SAP artifact to read.
11 🅒🅜 Repair on consignment + turnkey materialsConsignment stock + subcontracting / repair order; production order issuance ⚠︎ code not verifiedMM + PPP3 (no PO issued before CM starts) and the consignment-vs-turnkey split from the four-terms primer §2. You’ll see the system-of-record PO that NVIDIA’s CMs work without today.
12 🅝 Material reconciliationPhysical inventory / cycle counting; goods movement (MIGO)MMThe “suspect sheets” (P4/P5) fix. Cycle counting is a first-class SAP function — reading it shows the discipline NVIDIA is doing by hand. Ties straight to four-terms primer §1.
13–14 🅝↔🅒🅜 Planning loop / repair commitDemand-driven / service-parts planning overview ⚠︎ code not verifiedIBP / planningThe unmeasured CM↔Planning loop (P6) and the $2M SAP deal. Read for the shape of the signal that’s missing, not the detail.
15–16 🅝 Repaired unit → services pool → next customerOutbound delivery; Transportation Management freight orderEWM (outbound) + TMCloses the loop. The TM freight-order screen is the Expeditors leg; the Dallas↔Guadalajara cross-border move is the named bottleneck.

§3 — Suggested session plan

Four focused blocks. Salesforce first — it’s where NVIDIA’s flow genuinely starts, and it’s free.

  • Session 1 (½ day) — SFDC front half. Setup S1–S2, then map steps 2→5. Build one case, run it through a 3-gate approval, create a return order + replacement. Outcome: you’ve driven the exact path a Lonny-team frontline person drives.
  • Session 2 (½ day) — SFDC measure-the-loop. The P7/P8 dashboard + SLA-breach Flow. Outcome: a toy of the thing Greg + Lonny asked for; a felt sense of what data the SAP half has to supply.
  • Session 3 (½ day) — SAP on paper. S3, then read the six Tier-1 scripts above, ARM and cycle-counting first. Outcome: you can speak to how the “solved” version of steps 8–16 looks, and name exactly where NVIDIA diverges.
  • Session 4 (1 day, optional) — SAP hands-on. Only if session 3 leaves you wanting to drive EWM/TM/ARM. See §5.

§4 — The stitch-back (do this at the end — it’s the real deliverable)

After the sessions, re-open the process map and, for each of the 16 steps, write one line: “In a clean SFDC/SAP world this step is [the screen I built]. NVIDIA does it in [email/spreadsheet/suspect sheet] instead.” That single-page delta — clean-system vs. NVIDIA-reality, step by step — is the artifact that turns two afternoons of clicking into something you can show Greg + Lonny. It’s also the sharpest possible input to a Phase 1 scope: the wedge is precisely the set of steps where the delta is widest (cluster 1: steps 5→13).


§5 — If you decide to drive SAP hands-on (Tier 2)

Heavier — worth it only if the paper read leaves a real gap.

  • Deploy the S/4HANA Fully-Activated Appliance via SAP CAL (cal.sap.com) into your own AWS account. Pre-loaded with the Global Bike (GBI) demo dataset; includes embedded EWM and TM.
  • Cost is AWS compute, not SAP license — these are large instances (32–64 GB RAM, ~a couple $/hr). Suspend the instance between sessions or the meter runs.
  • Priority transactions to actually click, in process-map order: ARM returns cycle (step 10) → EWM inbound goods receipt (steps 8–9) → cycle counting / physical inventory (step 12) → EWM outbound + TM freight order (steps 15–16).
  • Expect SAP GUI / Fiori friction. Budget the first two hours for “where is anything.” The GBI data means you’re not building master data from scratch.

§6 — Honest ROI read

Do the Salesforce sessions for real — cheap, fast, and case management is genuinely where NVIDIA’s flow lives, so it builds credible intuition per hour better than anything else here. For SAP, the Tier-1 paper read is the high-ROI move; only stand up the appliance if you find yourself needing to feel a specific screen (most likely ARM disposition or EWM inbound). Standing up S/4HANA to understand the workflow is often more setup than insight — the test scripts carry most of the signal.

Open items to confirm as you go

  • The ⚠︎ code not verified scope-item codes — look them up in the Explorer before quoting any code to NVIDIA.
  • Whether Salesforce Order Management Returns or a custom object model better mirrors NVIDIA’s actual SFDC (process map §3: we don’t know their edition or whether the RMA object is standard or custom). Note which as you build.
  • If any exercise surfaces a question only Lonny/Greg/Alex can answer, drop it into the process-map §7 question bank rather than guessing.

Sources: RMA process map 2026-06-20; four-terms primer; repair-flow economic model. Environments: Salesforce Developer Edition + Trailhead; SAP Best Practices Explorer; SAP CAL (S/4HANA appliance, Global Bike dataset).