01 — Neural
Perceive
Read the mess.
Extract entities and intent from CRM notes, vendor price-lists, quote PDFs, supplier emails, tariff bulletins, and contract terms — the unstructured matter of a reseller's day.
The math isn't the missing piece. The missing piece is that uncertainty is currently implicit, unresolved, and discovered too late — which creates volatility, manual reconciliation, and erodes trust in what should be a daily decision signal. This proposal describes the Commercial Steering Engine: a governed daily control loop that extends the existing data aggregation layer into a decision system. Phase 1 delivers the resale workflow — the largest and most volatile line of business, and the fastest place to prove value without disrupting frontline flow.
Neurosymbolic AI is a hybrid architecture that combines the pattern recognition of neural networks with the logical rigor of symbolic reasoning.
Large language models reason probabilistically — excellent at reading a messy opportunity note or summarizing a vendor bulletin, but they cannot show their work in a form a CFO would accept at quarter-end. Symbolic systems are the opposite: rigid, explicit, and fully auditable, but unable to parse a procurement email on their own.
Neurosymbolic systems use the neural side to perceive — pulling structured facts out of CRM, ERP, vendor portals, and inboxes — and the symbolic side to decide — applying your pricing policy, vendor programs, and forecasting discipline with deterministic precision. The neural model proposes. The symbolic engine disposes.
Extract entities and intent from CRM notes, vendor price-lists, quote PDFs, supplier emails, tariff bulletins, and contract terms — the unstructured matter of a reseller's day.
Map every fact into a formal model of the commercial world: what a Customer is, what a VendorProgram is, how PriceLock relates to LandedMargin, which Renewal feeds which ForecastPeriod.
Execute pricing guidelines, discount authority, rebate eligibility, FX posture, and forecasting discipline as logic. Every conclusion traceable to a named rule and a specific fact.
Generate deal memos, pricing rationales, rep coaching notes, and exec-ready forecast narratives grounded in the reasoner's output — never hallucinated, always traceable.
The difference isn't about accuracy on any single task — it's about the kind of cognition each system performs. An LLM generates the most plausible next token. A symbolic reasoner evaluates the truth value of propositions against a formal knowledge base. For a reseller, that gap is the difference between a forecast that reads well and one you can bet the quarter on.
A rep's opportunity note — "Big refresh, Cisco + Azure, maybe Q2, customer haggling for 12 pts off list" — arrives as text. The model tokenizes it and passes the tokens through dozens of transformer layers, each applying attention across the sequence. At each position it produces a probability distribution over its vocabulary and samples the next token.
There is no separate "reasoning step." What looks like reasoning — chain-of-thought — is the model generating text that resembles reasoning it has seen in training data. It often gets the answer right because the pattern is well-represented. It sometimes gets it confidently wrong for the same reason.
The model has no persistent representation of your pricing policy. Even if you paste the policy into context, the model's attention to it is probabilistic; there is no guarantee rule PR-RENEWAL-02 will fire when it should.
The same note arrives. A neural extractor parses it into structured facts: Opportunity(id=O1, customer=C-Acme, vendors=[Cisco,Azure], closeQ=inferred, discountReq=0.12).
These facts are asserted into the knowledge graph. A reasoner evaluates every applicable rule. PR-DISC-04 is represented as executable logic: IF discount > authority(rep) THEN require_desk_review. FX-PROP-01 propagates today's EUR/USD into the quote. RB-TIER-03 checks whether adding this deal tips Cisco Gold to Platinum — unlocking 180 bps of backend rebate.
The conclusion isn't generated — it's derived. Every derivation produces a proof: the fact, the rule, and the binding. That proof is what makes the number defensible, consistent, and reproducible.
System 1 is pattern recognition at speed — the brain's autopilot. Fluent, confident, excellent for familiar problems. It is also where bias lives. When the pattern matches something common, System 1 is fast and accurate. When the situation is ambiguous — a new hyperscaler program, a novel vendor promo, a customer whose pattern has shifted — it produces a confident answer that can be materially wrong. A large language model is System 1 made mechanical.
System 2 is effortful thinking — applying policy, working through multi-step pricing math, reconciling a forecast against portfolio constraints. Slower and more expensive, but it is the only kind of thinking that produces an auditable trail of why a number was what it was. Forecasting and pricing at this scale are System 2 disciplines. A neurosymbolic system operationalizes System 2 — and can apply several modes of formal reasoning, the right one for the question.
Because a neurosymbolic system has an explicit model of the commercial world, it can apply the reasoning mode that fits the question — not just probabilistic association. Each of these shows up somewhere in a reseller's week.
Given facts and rules, derive what must follow. Certainty, not probability.
Given incomplete observations, infer what is most likely. Used to fill gaps in CRM.
Reason about cause and effect, not just correlation. Trace margin loads to their origin.
Find a price that satisfies customer target, margin floor, vendor program, and rep authority simultaneously.
With a structured world model, simulate changes — what happens if the dollar strengthens, if a renewal slips, if tariffs land.
Think of an ontology as the vocabulary and grammar of your business, written in a form a computer can reason over.
When an account executive says “a mid-market cyber refresh, Palo Alto through TD Synnex on the Platinum tier, with three years of managed services and a marketplace co-sell to Azure,” every noun is loaded with meaning — and every noun connects to others. Vendor tier unlocks rebate eligibility. Distributor choice shapes landed cost. Marketplace co-sell affects quota counting and incentive. Managed services unlock recurring revenue, which changes the renewal forecast shape.
An ontology makes that web explicit. It names the classes (Customer, Opportunity, Quote, LineItem, Vendor, Distributor), the properties (TCV, price-lock expiry, margin floor, tier status), and the relationships (a Quote hasLineItem, an Opportunity routedVia a Distributor, a Contract rolls_into a ForecastPeriod).
Without an ontology, the LLM knows the words. With an ontology, the system knows the meaning. Rules become composable and inheritable — a rule that applies to "all marketplace co-sell opportunities" automatically applies to anything classified as such. You write the rule once, against the concept.
Most importantly, ontologies compose across domains. A deal ontology plugs into a vendor-program ontology plugs into a services & renewals ontology plugs into a forecast ontology. The same structure that powers pricing decisions powers renewals management, vendor scorecards, and CFO-grade forecast reconciliation.
An LLM might see "Cisco Catalyst" and "C9300-48UXM" as unrelated strings. An ontology knows they resolve to the same SKU concept, that the SKU belongs to the Enterprise Networking family, and that family carries a 22% target margin floor — not the 14% applied to commodity endpoints.
A rule attached to Marketplace Co-sell Opportunity automatically applies to every deal classified as such — without re-authoring across AWS, Azure, and GCP variants. Your pricing manual becomes a small, maintainable tree instead of a sprawling SharePoint doc.
The same Customer anchors opportunity pricing, renewal forecasting, services attach rates, and rebate eligibility. One customer record resolves to one entity everywhere — the foundation of a reconciled forecast instead of five conflicting spreadsheets.
Two domains create most of the erosion in the daily decision signal: pricing — where margin guidance is uneven and guardrails are enforced manually — and forecasting — where updates are late, effort is high, and stability is low. They show up as different problems in leadership meetings but they are one commercial motion seen from two sides. Phase 1 resolves both together.
What leadership is asking for is confidence — that the numbers are stable enough to steer the business with, that pricing reflects intent rather than urgency, and that risk is surfaced early rather than concentrated in the last two weeks of the quarter.
A real operating engine already runs daily: the existing data aggregation layer produces point-in-time snapshots with auditable history and anchors the business on a sound construct — plan, actuals, invoiced, accruals, backlog, a clear variance-to-plan, and pipeline used to assess whether the gap can realistically be closed. Phase 1 does not replace that backbone. It turns it into a decision system by adding the five control capabilities the steering loop requires.
Daily-running engine with auditable history and a sound variance-to-plan construct. This is the foundation Phase 1 builds on — and the reason Phase 1 can be delivered without frontline disruption.
Each capability is the thing that turns daily data into daily decisions. None replaces an existing process — all five bolt on to the aggregation layer and operate as governed control loops.
The Commercial Steering Engine combines pricing and forecasting into a single governed control loop — turning strategic goals into tactical deal-level reality, and turning tactical reality into a reliable leadership signal. It goes from isolated optimization in silos to a globally optimized commercial outcome landscape.
What follows is the full engine anatomy: the operating model, the data blueprint that feeds it, the four artifacts leadership actually receives, the mechanics of how deals get priced and how the forecast gets built, the scenario engine that answers "what if," and the patterns of commercial steering this finally resolves.
Continuous ingestion from CRM, ERP, vendor portals, distributor feeds, FX & tariff sources. Rule engine re-evaluates the whole commercial state every 15 minutes.
One page each morning. The committed number with confidence band. A short list of fast movers that actually need leadership attention. The lock-risk exposure and margin at stake.
Deal desk, renewals, and pricing ops each receive a prioritized queue of exceptions with reason codes. Conforming deals flow without intervention; only true out-of-bounds items require review.
Minimal disruption is the design principle. Sellers work in the same CRM. They get faster quote turnaround on conforming deals and clearer guidance on the ones that need extra thought — delivered as a nudge in the tool they already use.
The system is invisible to the 140 sellers. It is indispensable to the 18 people in deal desk, pricing, renewals, and FP&A. And it delivers a one-page morning brief to the 9 executives who steer the business.
The engine ingests the existing aggregation layer (plan, actuals, invoiced, accruals, backlog, pipeline, variance-to-plan) and extends it with four additional streams that today live in spreadsheets, PDFs, vendor portals, and people's heads. Each stream is normalized into the ontology — so rules fire the same way regardless of where the fact came from.
In resale, the vendor price is the cost. The engine treats live vendor-price feeds with the same rigor as the margin-side constraints — continuous ingestion, price-lock windows tracked explicitly, tariff and FX overlays propagated into landed cost before a single margin calculation runs.
Rules express what must be true — not which system tells us it is true. An adapter layer absorbs ERP migration and upstream schema changes. The workflow survives the transition; rules are authored once and survive any source swap beneath.
The existing daily engine already produces these. Phase 1 ingests them unchanged — no disruption to current production flow.
CRM opportunity states, quote line items, stage transitions, backlog not yet delivered. Feeds gap-closure assessment and deal-level pricing.
The commercial playbook, expressed as executable logic. Not a PDF the model might attend to — a knowledge base the reasoner must satisfy.
Wins and losses, cycle times, deal shapes, discount-to-win-rate mappings. The foundation for predictive pipeline and win-rate modelling.
The qualitative matter of a sales cycle. The neural front-end extracts facts; the symbolic layer asserts them into the ontology for the reasoner to consume.
Context the business cannot see from its own books alone. FX spot rates, tariff schedules, vendor program announcements, regional demand indices.
The vendor price is the cost in resale. The engine treats live vendor-price feeds with the same rigor as margin-side constraints — continuous ingestion, price-lock windows tracked explicitly, tariff and FX overlays propagated into landed cost before any margin calculation runs.
Each artifact is generated continuously and read on a fixed cadence. No one logs into a tool to find out what's happening — the artifact arrives, and the decisions are waiting inside it. Below, sample content as it would appear to the executive team on the morning of this proposal.
| ID | Customer & deal | TCV | Close & conf. | Signal | Posture | Impact |
|---|---|---|---|---|---|---|
| O-4417 | Helvetica Bank AG Cisco · PaloAlto · Azure · Mgd Svc | $9.60M | 04 Jun · 88% | Tier lift | Prioritize close · triggers Cisco Platinum unlock | +$1.05M |
| O-4189 | Kestrel Manufacturing Dell HW · 3y MSA | $6.40M | 12 May · 72% | Lock exp | Re-quote w/ 6% HW rebalance · preserve customer price | +$182k |
| O-4032 | Meridian Health Azure EA · 3y renewal | $4.20M | 30 May · 64% | Disc req | Desk review · 18% ask above rep auth; counter 12% + attach | +$91k |
| O-3891 | Lindqvist Industri AB NetApp · EUR billed | €3.10M | 21 May · 58% | FX drag | Hedge EUR · repost USD landed margin to deal | +$74k |
| O-4512 | Coastal Logistics Palo Alto NextWave | $2.80M | 18 May · 81% | Co-sell | Route via Azure Marketplace · MDF + quota double | +$142k |
| O-4298 | Northbridge Research HPE · 2y svc attach | $2.60M | 09 May · 77% | Svc gap | Attach managed services · pull-through at 4.2% margin | +$109k |
| O-4401 | Brightforge Studios Endpoint refresh · SMB | $1.90M | 27 May · 69% | Lock exp | Re-quote at current street · margin at risk −180 bps | +$58k |
| O-4477 | Saito Manufacturing K.K. Cisco DC · JPY billed | ¥412M | 02 Jun · 63% | Stage stall | Stuck in Evaluate 19d · trigger AE escalation + decision-maker meeting | slip risk |
| O-4345 | Prairie State Credit Union Renewal · no-touch | $1.40M | 06 May · 52% | Churn sig | Proactive engage · NRR trailing 88%, exec call this week | +$1.18M def. |
Showing 9 of 12 fast movers · ranked by marginal Q2 impact · +3 additional in steering queue for Thursday
Each change justified by trailing 4-week recall/precision on its rule firings
Every live deal sits at the intersection of competing economic drivers: customer target, product-category margin floor, rep discount authority, vendor program eligibility, landed-cost reality, strategic intent.
The cost side is re-evaluated with every tick of the VendorPriceFeed. Price-lock windows are tracked explicitly, tariff and FX deltas propagate into landed cost before the margin calculation runs. The vendor price is the cost — it gets the same rigor as the margin constraint.
Where managed or professional services are delivered by a third party, the Subcontractor cost-to-serve is a first-class input. Pass-through margin on subcontracted work is enforced the same way as margin on resale line items — no hidden leakage in reconciliation.
Rep pipeline plus variance-to-plan, trued up against actuals, accruals, and backlog — with experienced judgment applied at commit meetings to smooth known distortions. Works until it doesn't. The miss is discovered at quarter-end.
Same aggregation-layer elements plus live vendor pricing, market signals, and confidence-weighted pipeline — composed into a full P&L: Revenue layers on top, Cost layers in the middle, Gross Profit falling out at the bottom. Every line is a named operation; click to see the driving facts.
Signed contracts landing this quarter. 312 contracts · 94% confidence band on recognition timing. Source: aggregation layer + contract ontology.
Backlog from prior quarters expected to deliver in Q2 — weighted by historic conversion timing. 48 deals · 91% confidence. Held back: $2.1M pending fulfillment blocks.
New bookings expected to close and recognize revenue within Q2. Confidence-weighted against historical win rate by deal shape. 68% confidence band — the most volatile revenue layer.
Work delivered but not yet invoiced — standard accrual treatment from the aggregation layer. High certainty; bookkeeping convention only.
Live vendor pricing × landed volume per SKU. Price-lock windows tracked per quote; expired locks repriced at street. 118 quotes on expired locks today — margin impact −$634k if unaddressed.
Cost of third-party-delivered managed/pro services at 36mo contract terms, amortized to Q2 period. Pass-through margin policy enforced; no subcontractor work booked below floor.
In-house managed services and professional services delivery cost — utilization × blended rate from the services ontology. Held within band; no attention required this cycle.
Net cost impact from EUR/GBP/CHF movement against USD baseline. $46.8M of European HW exposure at Q2-open rates. Hedge recommendation sits in today's leadership pack.
US → CH networking-HW tariff (3.2% on 38% of BOM for affected deals). Propagated into landed cost before margin calculation runs. 29 deals affected in current book.
Cisco / PANW / Microsoft backend rebates, probability-weighted against attainment. Cisco Platinum unlock (+$1.05M) contingent on Helvetica Bank O-4417 closing by 06-30 — tracked in fast-movers.
Standard reserve policy applied to revenue layers. No unusual movement this cycle. Aggregation-layer convention.
± $3.6M total — $0.8M from in-period booking uncertainty, $1.2M from vendor price-lock exposure, $0.9M from FX movement window, $0.4M from rebate attainment, $0.3M residual. Each component tracked to its driving ontology nodes.
Today: 15.3%. Target (floor): 14.2%. Optimum (per scenario engine): 17.8% — achievable by tightening floor and pulling the 104 low-touch renewals into proactive engagement. This is the number the business is steering against.
A single GP$ number debated at quarter-end meetings, with no named decomposition and no traceable link from line to driver. The miss is discovered too late to steer.
Revenue layers sum to top line. Cost layers sum to COGS&S. Gross profit falls out. Confidence decomposed by source. The KPI — Resale Gross Profit Margin — sits at the bottom, computed from the same facts every morning, defensible every morning.
A higher target margin lifts gross margin percent but compresses win rate; at the extreme it reduces both revenue and absolute gross profit dollars. A lower margin floor wins more deals but leaks margin on the ones that would have closed anyway. Today these tradeoffs are debated in meetings with anecdote and instinct. The engine makes them visible and quantified.
Because the entire commercial state lives in the ontology, leadership can perturb a driver — target margin, margin floor, vendor cost — and see every connected number respond with confidence bands attached. Every answer is decomposable to the rules that produced it. This is what turns leadership meetings from investigation into steering.
Every reseller commercial team develops a set of reflexes to compensate for unreliable signal — panic concessions near quarter-end, tribal-knowledge explanations in leadership reviews, spreadsheet reconciliation that replaces steering. These reflexes are expensive, and they compound. Below: the five patterns that most visibly block commercial steering today, and how the system resolves each. The full nine-pattern inventory, including timing integrity and policy drift, is available below.
Broad concessions used to "buy the quarter" because signal reliability is low and decisions are made under time pressure. Discounts land on deals that would have closed anyway.
Confidence bands plus a short fast-movers list isolate where intervention actually matters. Concessions become targeted posture decisions, not blanket behavior — applied to the 12 deals where they change outcomes, not the 941 where they don't.
Prevents avoidable margin give-up. Concentrates concessions where they truly change outcomes.
The daily pack replaces debate with a short list of actions and owners.
"Why is this weird?" gets answered by whoever happens to know, or deferred because context lives in heads and inboxes. Decisions wait on backchannel pulls.
The workflow produces a structured what changed, why narrative and traceable reason outputs, reducing reliance on who happens to be in the room.
Less decision latency; fewer costly delays that turn into urgency-driven margin erosion.
Meetings stop being investigative — they become steering sessions.
Meeting time is spent reconciling numbers, not deciding, because the forecast is hard to trust. Nine-and-a-half FTE-weeks per quarter-end chasing spreadsheet variance.
Signals are normalized into a daily decision-ready dataset and variance is decomposed into explicit named drivers. Reconciliation is a continuous graph operation, not a quarter-end sprint.
Cuts hidden operating cost and reduces compounding errors that create late-cycle surprises.
Leaders get a stable daily delta narrative instead of conflicting numbers.
Price-protection windows expire, margin compresses quietly, and re-quotes pile up late. The damage is visible only in the margin print, by which point it's locked in.
Price-lock exposure is tracked explicitly on a 7-day horizon and surfaced early in the daily lock-risk list, with recommended posture and owner routing. Action happens while there is still time to respond cleanly.
Prevents margin leakage, reduces timing-driven re-quotes, protects profitability under volatility.
Lock-risk becomes a daily control item — not a late surprise.
Pricing relies on manual checks with slow cycle times, frequent exceptions, and uneven outcomes across reps and regions. Deal desk drowns in routine approvals.
Guardrails become executable logic. Conforming deals flow quickly without desk intervention; only true out-of-bounds items require review, each with an explicit reason code and owner.
Faster quoting for in-bounds deals, fewer exceptions, higher margin predictability.
Fewer "everything is an exception" escalations reaching leadership.
The five cards above surface the most visible patterns. Four additional patterns show up less often in leadership view but cost the business as much as those above — particularly around late pricing changes, policy drift, and the deferred clarification loops that chew up cycle time.
| Pattern | Before · baseline | After · steering workflow | Bottom-line impact | Why execs feel it |
|---|---|---|---|---|
| Late-cycle panic concessions | Broad discounts used to "buy the quarter" because signal reliability is low. | Confidence bands + fast-movers list concentrate concessions where they change outcomes. | Prevents avoidable margin give-up. | Daily pack replaces debate with actions and owners. |
| Tribal-knowledge explanations | "Why is this weird?" answered by whoever happens to know; else deferred. | Structured "what changed, why" narrative with traceable reason outputs. | Less decision latency; fewer urgency-driven margin losses. | Meetings become steering, not investigation. |
| Deferred clarification loops | If unclear, leadership defers and burns hours waiting for sales or deal-desk explanations. | Missing prerequisites become explicit requests with owners, due-bys, and expected impact. | Reduces rework; prevents late escalations that force concessions. | Fewer "we'll get back to you" moments. |
| Spreadsheet reconciliation replaces steering | Meeting time spent reconciling numbers because the forecast isn't trusted. | Normalized daily dataset; variance decomposed into explicit named drivers. | Cuts hidden operating cost; reduces compounding errors. | Stable daily delta narrative, not conflicting numbers. |
| Silent price-lock erosion | Lock windows expire, margin compresses quietly; re-quotes pile up late. | Lock exposure tracked on 7-day horizon, surfaced early with owner routing. | Prevents margin leakage; protects profitability under volatility. | Lock-risk becomes a daily control item. |
| Manual guardrails & approval thrash | Manual checks; slow cycle times; uneven outcomes across reps and regions. | Guardrails as executable logic; conforming deals flow; only true out-of-bounds items reviewed. | Faster quoting; fewer exceptions; higher margin predictability. | Fewer "everything is an exception" escalations. |
| Late pricing changes destabilize forecast | Pricing shifts late in the cycle inject noise into forecasting and distort steering. | Forecast confidence informs pricing posture; unstable high-impact deals get tighter control earlier. | Fewer end-of-period swings; fewer reactive pricing moves. | Fewer surprise deltas and last-minute reversals. |
| Policy drift — "exceptions become the norm" | Unstructured overrides become routine; governance collapses into urgency management. | Overrides become reason-coded decisions; weekly steering tunes thresholds based on recurring failure modes. | Shrinks exception surface area; reduces recurring leakage drivers. | Weekly pack makes recurring failure modes visible and actionable. |
| Timing integrity — "no dates," quarter misplacement | Timing and linkage gaps distort forward views and force manual cleanup; trust erodes. | Forecast-ready gating prevents structurally incomplete items from contaminating the forecast; remediation is routed early with owners. | Less artificial variance; more stable commit accuracy; fewer late-cycle cleanups. | Execs stop arguing "is this real" — they start deciding "what do we do." |
The GP First Instance is the first productionized run of the Commercial Steering Engine, focused on the Resale Gross Profit Margin KPI. What follows is the first instance walkthrough — a live Q2 review against today's book. The CFO wants a committed number. The CRO wants pricing discipline. The COO wants to know which deals are real. The rep roll-up arrived this morning — $378M in open pipeline across 941 opportunities — and, as every quarter, a material chunk of it isn't in a state anyone can reason over. Dates missing. Prices stale. Discounts applied without context. Renewals treated as commodity line-items.
Below: the snapshot. Then: how the process traditionally runs, how a pure LLM would handle it, and how a neurosymbolic system handles it — not just producing a cleaner number but reasoning across the enterprise ontology to surface specific actions the commercial team can take this week. Every figure traceable to its source.
Before introducing any AI, here's the workflow the commercial ops and FP&A teams run every quarter. The process is sound — but it's slow, uneven across reps and regions, and heavily dependent on manual reconciliation in the final two weeks.
Each rep updates CRM opportunities — stage, close date, commit flag. Quality varies by rep and by geography.
Deals above authority are escalated to deal desk. Discount requests reviewed; pricing is re-quoted where vendor terms have shifted.
Finance reconciles CRM rollup against ERP-booked revenue, vendor rebate accruals, and service backlog. Excel-heavy.
Renewals managers work the top-dollar expiries. Long tail under $50k gets automated quotes at list-minus-standard-uplift.
CRO, CFO and region leads hold a two-hour call. Final commit is a haircut on rep rollup based on experience.
Quarter-end takes 9–10 finance FTE-weeks. Two reps looking at similar deals routinely produce different discount decisions. When a number is challenged — by a CFO, an auditor, or a vendor asking for a rebate true-up — the rationale lives in an email thread or a spreadsheet with a broken macro. Last quarter's miss was −7.8% to commit; no one could tell you which segment or which rule drove it.
A pure LLM approach: feed the pipeline rollup into a long-context prompt along with the pricing policy and last quarter's actuals, and ask for a committed forecast and pricing recommendations. Fast, fluent, and dangerous to rely on. Here is what the model produced.
The output reads like something a chief of staff might draft. It is also materially incorrect on several counts that an experienced commercial leader would catch — and that the CFO will discover three weeks into the quarter, when the actuals don't match.
The neurosymbolic system runs the rollup through five stages: neural extraction into the ontology, symbolic enrichment from live external sources, rule-engine evaluation against policy, neural explanation grounded in the proof, and finally cross-domain reasoning to produce specific actions. Every step is inspectable.
| Rule | Condition | Inputs | Outcome |
|---|---|---|---|
| FC-DATE-01 | Missing closeDate: infer from deal-shape abductively or downgrade commit | 212 opps · avg TCV $254k | Inferred |
| PR-LOCK-02 | If quote.validUntil < today: refresh cost basis, recompute margin, flag re-quote | 118 quotes · $36.2M | Re-quote |
| PR-DISC-04 | If discount > rep_authority(segment,tier): route to deal desk | 78 opps · $24.8M | Refer |
| PR-TARGET-07 | Margin floor differs by (product_family × customer_segment × strategic_intent) | Helvetica · Strategic-Retain · Cyber | Floor 16.2% |
| RB-TIER-03 | Cisco Platinum threshold: close $6.2M gap before 2026-06-30 to unlock +180 bps | Q2 Cisco pipeline = $58.4M | Unlock $1.05M |
| FX-PROP-01 | EUR/GBP/CHF moves > 1.5%: recompute landed margin on open EU-billed quotes | 108 quotes · $46.8M | Reprice |
| RN-RISK-05 | Renewal within 90d · NRR < 95% · no-touch: flag for proactive engagement | $17.2M · 104 contracts | Engage |
| MS-ATTACH-02 | HW deal > $1M without managed-services attach: flag pull-through opportunity | 29 deals · $47.2M TCV | Attach |
| CO-SELL-01 | Marketplace co-sell eligible: route through hyperscaler for funding + quota credit | 19 deals · $10.6M | $478k funding |
| FC-RECON-09 | CRM ↔ ERP ↔ vendor-portal reconciliation as a continuous graph operation | 941 opps · 780 contracts | Continuous |
Unlike the LLM, the NSAI system didn't stop at a fluent paragraph. It filled in the missing CRM dates with abductive inference, caught every expired price-lock and recomputed landed margin, applied differentiated margin floors by segment × product × intent, and then reasoned across six ontologies — pricing, rebate, renewal, marketplace, FX, and services — to produce five specific moves the commercial team can make this week. Every number traces to a rule ID and a source.
Press Run. Both systems receive the Q2 resale pipeline at the same moment. Watch the LLM sprint to a fluent commit while the neurosymbolic reasoner extracts, enriches, evaluates rules, and composes a proof tree — deductively, causally, and by constraint-satisfaction — to produce actions for the week. Every number traced to a source.
The path from today's commercial operation to the full steering system is deliberately incremental. Nothing rips and replaces. Nothing requires sellers to learn a new tool. Each phase builds on the prior one, delivers value on its own, and can be paused or re-scoped without losing ground. Four phases, sequenced to build confidence:
The broader context: the business is in an ERP migration window with multiple systems underneath forecasting and pricing scheduled for replacement. The approach is designed to work now, using current sources, while remaining portable as schemas change — decision logic is deliberately separated from source-specific implementation, so the workflow survives the transition. New sources plug in through an adapter layer; rules do not need to be rewritten.
The first productionized run of the Commercial Steering Engine, focused on the Resale Gross Profit Margin KPI. One commercial motion, end-to-end, shipping in ten weeks — proving the loop works before we expand scope.
Move additional resale scope into the engine — more product families, more customer segments, more vendors. Each addition reuses the ontology core; only the local rules and data sources change.
Bring additional regions into the engine. Each geo inherits the same ontology and artifact cadence; local FX, tariff, and distributor relationships plug in through the adapter layer.
Apply the same engine shape to professional services, managed services, and staff augmentation. Different commercial primitives, identical structure — one ontology, many motions.
Fixed-scope IT & communications project work — design, deployment, migration, implementation. The commercial primitives are different in shape but identical in structure to resale.
Recurring-revenue IT & communications operations — MSP, MSSP, NOC/SOC, cloud ops. Recurring-revenue forecasting rides on the same contract & renewal ontology already proven in resale.
IT & comms resource placement — contract staffing, T&M engagements, specialist sourcing. Most margin-sensitive of the three; benefits most from live rate & availability signals.
The same ontology, the same reasoner, the same four artifacts — applied to a different commercial motion. What gets built for resale in Phase 1 becomes the platform that runs every commercial decision across every line of business.