Technical Documentation & FAQ

MarginArc reads from standard Salesforce objects: Opportunity (Amount, Stage, CloseDate, Owner), Account (Industry, AnnualRevenue, Segment), and a lightweight custom object for MarginArc-specific fields. We also access closed-won and closed-lost history for model training. All access is read-only to your CRM data — MarginArc only writes to its own managed package objects (recommendations, scores, audit logs).

You get margin recommendations from Day 1. The SignalLayer phase uses Bayesian priors and OEM baseline data to generate useful recommendations immediately, even with zero historical deal data. As your CRM history builds, accuracy improves from 68% (Day 1) to 79% (Month 1-3), 88% (Month 3-6), and 94% (Year 1+). Most customers report measurable margin improvement within the first 30 days.

Only 6 fields per deal, all quick selections — no free-text entry required:

• Deal Registration — Select the registration tier (e.g., Premium Hunting, Renewal)
• Services Mix — Approximate services percentage of deal value
• Solution Complexity — Single-vendor, multi-vendor, or full-stack
• Competitor Count — How many VARs are competing (0-5+)
• Competitor Identity — Multi-select picklist of known competitors
• Displacement Flag — Are you protecting an incumbent position or attacking?

The remaining 25 signals are collected automatically from your CRM, account settings, or the MarginArc Network.

No. Never. Your raw deal data never leaves your Salesforce org. The MarginArc Network uses federated learning (only model gradients are shared, not data), differential privacy (mathematical noise injection with ε=1.0), and a Conflict Firewall (SHA-256 hashing) that automatically excludes your own data from any network query results. It is mathematically impossible for another participant to see your deal data.

The MarginArc Network is an anonymous data cooperative. Each participating VAR contributes anonymized, aggregated intelligence to the collective — think of it like Waze for pricing. The network provides five signals: Peer Win Rates, Network Margin Bands, Regional Variance, Deal Velocity Norms, and Competitive Tactics. All data is processed through differential privacy (noise injection), federated learning (local training, shared gradients), and conflict firewalls (you never see your own data in aggregates). The more VARs participate, the better the intelligence for everyone.

MarginArc is designed for this exact scenario. On Day 1, the SignalLayer phase uses Bayesian priors seeded by OEM baseline margins, industry benchmarks, and the MarginArc Network's aggregate data. You get actionable recommendations immediately. As your closed-deal history accumulates, the model progressively shifts from prior-based estimates to data-driven predictions. Even organizations with as few as 20-30 closed deals in their CRM begin seeing meaningful accuracy improvements.

Accuracy depends on your data maturity phase:

• Day 1 (SignalLayer): 68% — Bayesian priors + OEM baselines
• Month 1-3 (MatchIQ): 79% — kNN similarity + logistic win probability
• Month 3-6 (PricePoint): 88% — Stacked ensemble + MarginArc Network
• Year 1+ (AdaptIQ): 94% — Online learning + drift detection

Accuracy is measured as the percentage of recommendations that fall within 1.5 margin points of the observed optimal margin on closed deals.

Absolutely. MarginArc is a co-pilot, not an autopilot. Reps always set the final margin. The recommendation, Deal Score, and win probability are guidance tools. When a rep overrides the recommendation, MarginArc tracks the deviation and uses the outcome (win or loss) as training data to improve future recommendations. Over time, the Compliance Score signal measures how often a rep follows recommendations and correlates that with win rates.

MarginArc maintains a proprietary OEM database covering all major IT vendors including Cisco, HPE/Aruba, Dell, Palo Alto Networks, Fortinet, Microsoft, VMware/Broadcom, Lenovo, Juniper, Arista, and dozens more. The OEM database provides baseline margin ranges by product line and is continuously updated. If you work with a vendor not yet in our database, MarginArc uses your historical data and network aggregates to generate recommendations for that vendor within the first few deals.

MarginArc tracks each OEM's fiscal calendar and calculates a "Quarter-End Timing" signal that captures how many days remain in the OEM's fiscal quarter. This is combined with historical seasonal patterns and the "Program & Rebate Shifts" signal to dynamically adjust margin recommendations. For example, if Cisco historically offers better back-end rebates in the last 3 weeks of their fiscal quarter, MarginArc's recommendation will reflect the expected effective margin uplift.

AUTO signals (17 of 31) are collected and calculated entirely by MarginArc with zero rep effort. They pull from CRM fields, compute historical metrics, and detect patterns automatically. REP signals (6 of 31) require the rep to make a quick selection on each deal — things only the rep would know, like competitor count and deal registration tier. ACCT signals (2 of 31) are set once per account by sales leadership. NET signals (6 of 31) come from the MarginArc Network. In total, the rep only needs to fill in 6 fields per deal.

The Deal Score is a weighted composite of five factors: Margin Alignment (35%), Win Probability (25%), Risk-Adjusted Value (20%), Deal Structure (10%), and Competitive Position (10%). The score ranges from 0-100 and is color-coded: Red (0-39, Needs Work), Orange (40-59, Below Average), Yellow (60-74, Fair), Green (75-89, Good), and Bright Green (90-100, Optimal). See the Deal Score section above for a detailed breakdown of each factor.

Win Probability estimates the likelihood of closing the deal at a given margin percentage. It uses a logistic function that models the inverse relationship between margin and win rate: higher margins reduce win probability. The model is calibrated against your actual closed-won and closed-lost deal history. Win Probability is one of the five inputs to the Deal Score and is also used to calculate Risk-Adjusted Gross Profit (margin x win probability x deal size), which represents the expected dollar value of the deal.

Every closed deal — won or lost — becomes a training example. MarginArc tracks the margin that was set, the recommendation that was given, and the outcome. This feedback loop continuously refines the model. After Year 1, the AdaptIQ phase uses online stochastic gradient descent for real-time updates and DriftGuard (CUSUM algorithm) to detect when market conditions shift, triggering automatic model recalibration.

No minimum deal size is required. MarginArc generates recommendations for any opportunity in your CRM pipeline. However, the economic impact is naturally greatest on larger deals where even a 1-2% margin improvement translates to significant dollar amounts. Some customers configure MarginArc to only display the component on deals above a certain threshold (e.g., $25K) to reduce noise for small transactional orders.

The MarginArc Network uses a three-layer privacy architecture. First, federated learning trains models locally in your Salesforce org and only shares encrypted model gradients (mathematical derivatives), not raw data. Second, differential privacy adds calibrated statistical noise to every network query, making it impossible to isolate any individual deal. Third, the Conflict Firewall uses SHA-256 hashing to ensure your own organization's data is always excluded from the aggregates you receive. The result: you get rich competitive intelligence without any organization being able to identify another's specific deals.

MarginArc is a Salesforce managed package that supports Enterprise, Unlimited, and Performance editions of Salesforce. The Lightning Web Component (LWC) requires Lightning Experience (not Classic). MarginArc does not require any additional Salesforce add-ons or licenses beyond your standard CRM seats.

Typical implementation takes 1-2 weeks:

• Day 1-2: Install the managed package and configure OEM/product mappings
• Day 3-5: Set account-level signals (Customer Segment, Relationship Depth) for your top accounts
• Day 5-7: Train reps on the 6 input fields and the Deal Score component
• Day 7-14: Pilot with a small team, validate recommendations, adjust mappings

Reps start receiving recommendations from Day 1 of going live. The model accuracy improves automatically over time with no additional configuration.

Yes. MarginArc is designed for the entire sales organization. Every rep sees the MarginArc component on their opportunities. Sales managers see team-level analytics including compliance rates, margin trends, and deal score distributions. Leadership gets dashboard views showing aggregate margin improvement, forecast accuracy, and competitive positioning. Role-based access ensures reps only see their own deal data while managers see their team's aggregates.