Pricing Authority Stack
Most manufacturers and distributors have built execution and part of decisioning. Authority still lives in Slack, spreadsheets and tribal knowledge. That gap was tolerable when every exception needed a human click. It stops being tolerable when AI can draft, recommend or commit a price before the controls are explicit.
Four jobs, not one system
The useful split is not ERP versus CRM. It is which part of the pricing decision each layer owns, and whether that ownership is written down where humans and agents both have to respect it.
ERP, CRM and CPQ. Not enough on its own when two systems can produce a different number for the same deal. The usual tell is a standing reconciliation workstream, where the invoice does not match the signed quote and nobody chose that outcome.
Guides, waterfalls, rules and optimization. Pricefx is one illustrative implementation, not the only valid answer.
Decision rights, thresholds, override paths and agent policies. Usually lives in heads and informal channels until an agent needs it in writing.
Actor, time, policy result, rationale, final action. What internal-control testing asks for first, and what answers why a model priced one customer differently from another.
Illustrative architecture
The Pricing Authority Stack
Three layers. Most enterprises have built one and a half.
Decisioning
Pricefx · PROS · Zilliant · Vendavo · AI models
Where the price actually gets calculated, optimized, and segmented.
Execution
SAP S/4HANA · Salesforce Revenue Cloud
Where the price gets posted, invoiced, booked.
governs
Execution is mostly built. Decisioning is partly built. Authority lives in tribal knowledge, Slack, and spreadsheets.
Everything above rests on inputs that agree. Customer master, product hierarchy and landed cost are where most of these programs actually stall, which is why the data work usually comes before the authority work. Start with the data assessment →
The objection worth answering
Every system in that figure ships approval workflow. The gap is not that approvals are missing. It is what they are attached to.
Approval workflow, as built
Authority, as a layer
This is segregation of duties applied to price commitment. The principle is old. What is new is that the caller is no longer necessarily a person, and a control that only exists on a screen cannot see a caller that never opens one.
Where authority lives
A system, not a document describing one. Three patterns hold up, trading enforcement strength against how much has to change.
Strength. Closest to the logic, fastest to stand up.
Limit. Governs only callers that come through it. Anything pricing directly in CRM or ERP sits outside.
Strength. One policy, enforced identically whichever system or agent asks.
Limit. It is a build, and partial adoption is worse than none because it produces a control people believe in.
Strength. Fastest answer to the near-term risk. ERP and CRM change nothing.
Limit. Governs machine callers well and human paths not at all. A layer to add, not a place to stop.
Most mid-market programs start at A for human paths and C for machine paths, then converge. The one to avoid is a policy document that nothing enforces, which reads like authority in a review and is invisible at the moment of commitment.
The artifact
The smallest useful form of an authority layer, and the first thing worth writing down. If it cannot be filled in without an argument, the argument is the finding.
| Actor | May do | Up to | Escalates to | Evidence |
|---|---|---|---|---|
| Inside sales | Submit at guided price | Guide floor | Deal desk | Actor, quote, guide result |
| Field sales | Submit below guide | Named band, per segment | Regional manager | Above, plus rationale |
| Deal desk | Approve exception | Named threshold band | VP Pricing | Above, plus policy version |
| Platform rule | Auto-approve in guide | Guide envelope | None, logged | Rule id, inputs, output |
| CRM or chat agent | Draft and recommend | No commit right | Human submitter | Prompt, data, recommendation |
| API or MCP caller | Submit inside policy | Band of its named owner | That named owner | Caller id, policy result |
| Anything unlisted | Nothing | Not applicable | Deal desk | Denied, attempt logged |
Illustrative. The values are placeholders and are set per business. The column headings are the point, and so is the last row.
Three things recur the first time a client fills this in. Two actors turn out to hold the same right through different systems. At least one path has no named escalation. And the machine callers already running in production do not appear anywhere on the sheet.
External AI as a pricing actor
Chat assistants, CRM agents, Slack or Teams helpers and API or MCP callers are new actors in the pricing path. Treat them as callers that must operate inside authority, with draft, recommend, submit and commit held as four separate rights.
Forrester’s 2026 B2B predictions expect roughly one in five B2B sellers to be drawn into agent-led quote negotiations during 2026, and project more than ten billion dollars of B2B loss tied to ungoverned use of generative AI. Forrester, October 2025.
For RevSeekr’s own pilot design and data boundaries, see Trust & AI use. The Pricing Intelligence Pilot is an expert-reviewed evidence workflow, not an unsupervised price-setting product, and it does not accept client datasets through this website.
Where you stand
Four lines you can run in your next commercial meeting
Four answers inside a minute means authority exists somewhere and someone wrote it down. A pause on any one of them tells you which column of the matrix to start with, and it is the same finding the diagnostic scores formally.
Source essays
Why overlapping platform claims produce a number without producing an architecture.
Five workflows that weaken when the screen was the control.
A platform conceding reasoning and keeping governance. The same split, one layer up.
Use a diagnostic to see where readiness or leakage sits today. Use a consultation when you need to make authority, decisioning and audit explicit before AI or another platform release forces the issue.