Beyond PLM examines outcome-based pricing as AI changes PLM work
AI-generated summary
Beyond PLM argues that as AI takes over more product-information preparation and coordination, PLM pricing should shift from counting software users toward the responsibility a platform assumes for completed work and trustworthy product records. Author Oleg, who discloses that he is OpenBOM’s co-founder and CEO, presents this as a proposed direction rather than an established industry model. AI challenges three linked businesses: manufacturers’ engineering work, implementation partners’ billable configuration and integration hours, and vendors’ named-user licenses. The discussion starts with Jakob Strømann-Andersen, Director of Innovation and Sustainability at Henning Larsen, questioning rising Autodesk license costs while architects still sell hours that AI could reduce. Commenter Dominic Seah raises problems with selling outcomes, including defining success, allocating credit among providers, and sharing losses as well as profits. Oleg agrees that PLM cannot take responsibility for broad results such as profitable product launches. Instead, an outcome should be bounded: preparing an engineering change package, checking agreed rules, routing exceptions, collecting approvals, and delivering it to suppliers. Contracts must identify applicable configurations, decision authorities, missing-information procedures, and completion evidence. Supplier acknowledgment proves receipt—not manufacturability or production success. The author expects automation to reduce revision comparison, BOM reconciliation, attribute checking, supplier-package assembly, and follow-up work. However, freed engineering hours often become capacity for additional releases or alternatives rather than cash savings. Competition also tends to turn automation into a standard expectation, as happened with automatically generated BOM reports, PDFs, STEP files, and other CAD derivatives. His argument is that labor savings attract customers initially but are unlikely to sustain differentiated pricing. More durable value comes from maintaining lifecycle information, connecting engineering decisions to observed results, and packaging accumulated expertise with accountable human approval. Customers will also increasingly expect vendors to supply domain-specific automation centrally rather than leaving users to build macros and integrations. A hypothetical change to a machined housing illustrates that distinction. Automation prepares and distributes revised drawings and BOM information, but a supplier subsequently questions an inspection requirement. Engineering clarifies it, and production later reports whether the requirement was understood and achievable. Connecting the revision, question, approved interpretation, approver, and production result could let the system flag similar ambiguity before the next release. Oleg calls this “Product Memory.” He cautions that one supplier interaction does not establish a universal rule: lessons need their material, process, or customer-specific limits preserved and evidence that subsequent checks help. Trusted records and learning must be earned, while respecting customer ownership of product data. The pricing discussion contrasts seats, tokens or API calls, completed transactions, managed information, and hybrids. Seats measure human access; tokens measure vendor consumption rather than useful results; transaction pricing must accommodate complexity, rework, and exceptions. Customer-support software illustrates the uncertainty: Gorgias starts at $10 for 50 monthly tickets, while AI-resolved conversations incur an additional resolution charge and still count as tickets. Helply moved from $1 per ticket with unlimited seats to $0.50 per outcome, then toward credit-based tiers. Oleg favors managed items, BOMs, CAD files, configurations, or assets as a more stable proxy for PLM responsibility. Outcome delivery need not mean per-outcome billing; a subscription can fund it. That approach has limits. Managed-record charges must fund identifiable maintenance of structures, revisions, suppliers, documentation, permissions, and integrations—not merely storage with a markup. Charging separately for every revision, relationship, or archived product could discourage preserving the history that makes Product Memory valuable. OpenBOM combines records and seats, charging for part numbers and CAD files rather than revisions and changes. The author acknowledges his commercial perspective and stresses that record counts establish service scope, not reliability or demonstrated improvement. Human judgment remains another cost and accountability boundary. AI services need explicit stopping rules and authorized reviewers for questions involving tolerances, sourcing, production methods, or customer requirements; difficult exceptions can materially increase delivery costs. Oleg concludes that seats may remain useful within hybrids but should not carry the entire business as agents do more work. His longer-term expectation is that trusted records, supplier collaboration, and evidence-based learning will outlast basic labor-saving advantages—and that pricing should reward preserving context rather than penalize it.