Summary

Gen AI holds genuine promise to transform Configure Price Quote (CPQ) software tools, but the market is long on rhetoric and short on proof. The traditional CPQ paradigm of human sales reps configuring a product, the system calculating a price, and generating a quote remains intact across virtually all vendors, old and new. Gen AI is being applied to make each step faster, not questioning whether those steps are needed in the first place. The market is drowning in “agentic” and “AI-native” marketing that, upon closer inspection, describes conventional workflow automation with a natural language interface bolted on top.
Gen AI offers the promise that pricing, packaging, and selling motions can be re-thought through the lens of highly automated agentic approaches while enabling a human in the loop to still apply checks and balances. Applied in novel ways, Gen AI has the potential to fundamentally transform how enterprises price, package, and sell products and services.  The customer reality of today’s CPQ tools, however, cannot be ignored. Current CPQ projects can be tenuous with longer-than-expected implementation cycles, higher-than-budgeted costs, real risks of re-implementation and/or outright failure, and considerable ongoing change management costs. These issues are particularly acute for rapidly growing, highly complex businesses with rapid rates of internal change and dynamic market conditions. If the marketing hype around Gen AI in CPQ holds up, it could deliver significant value for customers and be highly disruptive to the current CPQ market. 

The impact of Gen AI on CPQ can be viewed through the prism of two major dimensions:

  1. The Conceptual Rethink: How will companies fundamentally rethink and restructure approaches and processes for pricing, packaging and selling?
  2. The Practical Reality: How will CPQ software suppliers take advantage of Gen AI as an opportunity to create value for customers?

Software buyers will assess each of these change vectors on the merits of risk, timing and ROI/TCO. Buyers assessing new options in CPQ should ask if there is a quantifiable return within a reasonable (6-12 month) time horizon, if internal rates of return (IRR) exceed the average of other investment options, and if there is an option that does not significantly increase risk.

The Conceptual Rethink

Today’s Configuration Problem and Why It’s Ripe for Rethinking

Traditional CPQ configuration rests on a deterministic rules engine architecture dating back decades. The core mechanics are product hierarchy models, attribute-based selection, constraint rules (e.g., compatibility, exclusion, inclusion, validation), guided selling wizards, and BOM generation. Every permutation must be explicitly encoded.

The fundamental pain points this creates are well-documented but worth restating since they define the areas of attack for AI. These areas of attack are described below.

The “configuration tax” is the real killer. Complex manufacturers and enterprise software vendors routinely maintain 500–5,000+ configuration rules. Every product change, new option, and retired SKU requires rule updates, often by specialized CPQ administrators who are scarce and expensive. The maintenance burden scales superlinearly with product complexity. A company with 200 products and 50 options per product doesn’t have 10,000 rules. It has tens of thousands of interaction rules because of cross-product dependencies.

The knowledge bottleneck is the consequence of configuration expertise living in the heads of experienced sales engineers, deal desk admins, and product managers who understand which combinations actually work in the field versus which combinations are merely technically valid. Traditional CPQ captures the “technically valid” constraints, but not the “practically optimal” ones that live within tribal knowledge.

The interaction model forces sales reps to think in terms of the vendor’s product structure rather than the customer’s problem. A rep navigates menus, selects options, validates against rules, essentially acting as a human adapter between customer needs and product taxonomy. This is backwards, especially in light of modern technology.

Today’s customer wants more agency. Customers of all stripes (B2C, B2B, and channel resellers/partners) want the option to configure, price, and provision their own products/services independent of a sales rep. The concept of “user” is no longer defined as a sales rep.

When thinking about applying and/or rethinking CPQ with Gen AI, it is useful to approach it by asking three questions: What can be done now? What can be done with modest augmentation? What can be done through a fundamental rethink of CPQ as a process?

Layer One: What Can Be Done Now (Almost)

There are opportunities to apply Gen AI to the existing configure-then-price-then-quote paradigm without changing it. Most vendors are here or heading here.

Natural language configuration entry. Instead of navigating product hierarchies and option menus, a rep types or speaks: “Customer needs a mid-range server cluster for a healthcare data warehouse, must be HIPAA-compliant, 500TB usable storage, three-year lifecycle.” The LLM maps this to product attributes and generates a starting configuration. This is genuinely useful but fundamentally just a better input method. The underlying rules engine still validates and constrains selections.

Intelligent validation with explanation. Instead of a cryptic error code when a configuration fails validation, the system explains why in natural language and suggests the nearest valid alternative. “You selected the 48-port switch with the high-density fiber module, but this combination exceeds the chassis power budget by 15W. Swapping to the standard-density module resolves the issue and reduces cost by $1,200.” This is a straightforward LLM application over structured constraint data.

Rule authoring acceleration. Product managers describe new configuration rules in natural language, and Gen AI translates them into formal constraint logic. This directly attacks the configuration tax. Instead of a CPQ admin coding “IF OptionA = X AND OptionB = Y THEN Exclude OptionC,” a product manager writes: “The high-performance cooling module is required whenever the server has more than four GPUs, except in rack configurations with external liquid cooling.” The AI generates the rule, the admin validates it. This could realistically cut rule maintenance effort by 40-60%, and it’s one of the most practically impactful applications in the near term.

Configuration documentation and rationale. Gen AI automatically generates human-readable explanations of why a configuration is built the way it is. These are useful for complex technical proposals where the customer or their engineering team needs to understand the logic behind the recommended setup. This function is table stakes for any vendor with LLM integration.

Layer Two: Modest Augmentation and Refining How the Process Works

These capabilities shift the configuration process from purely deterministic to hybrid deterministic/probabilistic, which is where things get interesting.

Outcome-based guided selling. This inverts the traditional guided selling wizard. Instead of “What processor family do you want? → What memory configuration? → What storage tier?”, the system asks: “What workload will this run? What are your performance SLAs? What’s your power/cooling envelope?” Then it reasons backward from outcomes to configuration. This isn’t just a better UI. It requires the system to have a model of how configurations map to outcomes, not just a model of which configurations are valid. That’s a fundamentally different knowledge structure.

The hard part comes in building this.  The outcome-to-configuration mapping requires training data most vendors don’t systematically collect. Specifically, it requires post-deployment performance data linked back to the original configuration. Companies in possession of this data (e.g., large infrastructure vendors, cloud providers, industrial equipment manufacturers with IoT telemetry) have a massive advantage.

Probabilistic configuration scoring. Instead of binary valid/invalid, configurations get scored on multiple dimensions: technical viability, deployment risk (e.g., low/medium/high), margin profile, customer fit based on similar accounts, likelihood of requiring post-sale engineering changes. This turns the configurator from a gatekeeper into an advisor. A configuration described as “technically valid but historically problematic for customers in this industry segment” is more useful than just “valid.”

Cross-product and solution-level configuration. Traditional CPQ configures individual products. Enterprise deals are almost never single-product. Instead, they’re solutions composed of hardware, software, services, third-party components, and professional services. Such solutions contain a variety of pricing models, differing terms of duration, and take into consideration historical customer relationships.  Gen AI can reason across product boundaries: “Given that you’re configuring this network infrastructure, here’s the corresponding security stack, monitoring suite, and implementation services that customers with similar architectures typically deploy.” This is where deal size expansion happens, and it’s extremely hard to do with traditional rules engines because the cross-product constraint space is combinatorially explosive.

Configuration knowledge capture from unstructured sources. Sales engineers, field technicians, and customer success teams accumulate enormous tacit knowledge about what works and what doesn’t in real deployments. This knowledge lives in Slack channels, email threads, support tickets, Confluence pages, and people’s heads. Gen AI can ingest these unstructured sources and extract configuration intelligence: “In three separate support cases, customers who deployed Configuration X in humid environments experienced premature component failure. Consider flagging this as a risk when the deployment site is in a tropical or coastal region.” This is one of the highest-value applications because it closes the feedback loop between deployment outcomes and configuration decisions. This critical feedback loop is almost always broken in organizations today.

Multimodal configuration. Imagine the following scenarios: a customer photographs an existing rack, uploads the photo, and the system identifies the most current equipment, infers the configuration, and proposes an upgrade that is a functional equivalent or a replacement configuration. An architect uploads a CAD drawing of a building, and the system configures the HVAC, electrical, or networking infrastructure. Although this example is further out, it is technically feasible for constrained domains, and it completely bypasses the traditional product menu navigation paradigm.

Layer Three: Rethinking What Problem Gets Solved and How by Using Modern Tools

This is where the real innovation could occur. These aren’t vendor roadmap items today. They are architectural possibilities representing genuine category disruption.

From product configuration to problem configuration. The deepest rethinking would be to abandon the idea that CPQ starts with a product catalog and instead starts with a problem model. Customers don’t want to configure a server; they want to run a workload. They don’t want to configure a telecom bundle; they want to serve a subscriber base with certain characteristics. The system’s job isn’t to assemble a product from components but to decompose a problem into requirements and then map those requirements to whatever combination of products, services, and partners solves it.

This is a profound architectural shift because it means the configuration engine doesn’t begin with the vendor’s product ontology. The configuration engine design begins with a domain ontology of the customer’s world. Gen AI is uniquely suited to bridge the gap between natural language problem descriptions and formal product structures. It goes without stating building the domain models underneath is still hard, domain-specific work. Vendors who invest in industry-specific problem ontologies (not just product catalogs) will have a structural advantage.

Self-maintaining product models. Instead of human administrators maintaining configuration rules, the system ingests product engineering documentation (datasheets, compatibility matrices, engineering change orders, technical bulletins) and automatically derives and updates configuration constraints. When engineering releases a new product revision, the AI reads the change documentation, identifies which configuration rules are affected, proposes updates, and flags ambiguities for human review.

This doesn’t just reduce the configuration tax. It changes the economics of product complexity. Today, there’s an implicit ceiling on how many products and options a company can offer because each new option adds design, production and maintenance burdens. If rule maintenance becomes highly automated, companies can offer dramatically more configuration flexibility without proportional cost increases. That’s a market structure change, not just process efficiency improvement.

Configuration as continuous optimization, not point-in-time decision. Traditional CPQ treats configuration as a one-time event at the point of sale. For subscription, usage-based, professional services delivery, and outcome-based pricing models, the optimal configuration changes over time as customer requirements and usage patterns evolve. An AI-driven configuration engine could continuously monitor post-sale telemetry and proactively recommend reconfiguration: “The current configuration is over-provisioned for storage but under-provisioned for computing based on the last 90 days of usage. We recommend shifting to Configuration Y, which would reduce monthly costs by 12% while improving performance for the actual workload.”

This turns CPQ from a sales tool into a customer success tool, and, critically, creates a continuous upsell/cross-sell mechanism driven by genuine customer value rather than sales rep initiative. This changes where revenue capture happens in the customer lifecycle.

Agent-to-agent configuration negotiation. In this scenario, the buyer’s AI agent expresses requirements (e.g., “We need to expand capacity by 40% while maintaining current reliability SLAs and staying within a $2 million budget”), and the seller’s AI agent proposes configurations, negotiates trade-offs, and iterates toward a mutually acceptable solution without a human configuring anything. The “configuration” happens as an emergent outcome of a negotiation between two AI systems understanding their respective constraints.

This is the scenario that would make traditional CPQ genuinely obsolete, and no current vendor is building for it. The technical pieces like LLM-driven negotiation, API-first product catalogs, and constraint satisfaction exist today. However, trust and governance frameworks don’t. No enterprise procurement team today will allow an AI agent to commit to a multi-million-dollar configuration without human review. For lower-value, higher-volume transactions like SMB deals, self-serve tiers, and renewal/expansion motions, this could happen within three years.

Where Real Competitive Moats Will Form

For all the focus on AI, durable competitive advantage will accumulate in three areas outside of LLMs:

  1. Domain-specific training data. The vendor that has the richest corpus of configuration-to-outcome data (configurations paired with deployment success, customer satisfaction, support incidents, renewal rates, etc.) will build models that are genuinely better at recommending configurations, not just faster at accepting input. This data is proprietary, hard to assemble, and compounds over time.
  2. Industry problem ontologies. The vendor that builds the best models of customer problems (not just product structures) in specific verticals will win the outcome-based configuration paradigm. This requires deep domain expertise, not just engineering talent.
  3. Feedback loop architecture. The vendor that closes the loop between configuration decisions and real-world outcomes and does so automatically, continuously, and at scale will have a self-improving configuration engine getting measurably better with each transaction. Most vendors’ architectures don’t even have the plumbing to close this loop.

The Practical Reality

Rethinking vs. Relabeling

The fundamental test of disruption in CPQ is whether a company is changing what problem gets solved and how or merely automating the same sequential Configure → Price → Quote workflow with an LLM chatbot bolted on top. Most current CPQ suppliers fall into the latter category. The CPQ market is currently drowning in “agentic” and “AI-native” marketing that, upon closer inspection, describes conventional workflow automation with a natural language interface. It is important for buyers to understand the dynamics at play to navigate the purchasing landscape with confidence. There are three distinct categories of players on the spectrum of AI penetration in CPQ.

Tier One – Potential CPQ Disruptors to Track

Alguna (YC-backed, founded 2023) is a recent vintage entrant into CPQ with an integrated functional stack claiming to cover all elements of CPQ, usage and outcome pricing, billing, payments integration, metering and revenue recognition.
Alguna claims to be purpose-built for AI/token-based and hybrid pricing models, which absent heavy customization, legacy CPQ tools generally struggle with. The unified CPQ → Billing → Revenue Recognition approach in a single data model is, while not unique, still meaningfully differentiated from the fragmented approach of using multiple vendors to stand up a revenue engine. For companies starting fresh, this eliminates the “quote doesn’t match invoice” syndrome. Unlike many competitors, Alguna claims flat-fee pricing – an attractive model for start-ups trying to conserve cash as they scale. Founders come from fintech/payments infrastructure backgrounds, scaling similar internal tooling at Primer (payments orchestration) and Dojo (UK payments).

Broadn.io (founded 2021) is a vendor claiming to automate responses to RFQs from any channel. Rather than waiting for a salesperson to look at a request and construct a quote, Broadn’s AI agent processes quote requests from any communication channel (email, WhatsApp, web forms, etc.), automatically extract requirements, and generates quotes without human data entry. The vision of eliminating the “sales rep as data-entry operator” paradigm is genuinely different from traditional CPQ. Their claimed position of being manufacturing-focused seems dubious as there is nothing in company or founder backgrounds to suggest strength in this area. The multi-channel NLP intake idea is compelling but brutally hard to execute well. This is especially true for complex B2B configurations where a customer’s emailed requirements are ambiguous and require clarification. Anyone who’s worked in enterprise sales knows that the quote request is rarely clean. Nevertheless, this is an interesting example of how AI can be leveraged to transform long-established process models.  

Tier Two – Interesting Niche Plays to Follow but Not Yet Major CPQ Disruptors

Veles is explicitly not trying to replace current CPQ systems. It is positioned as a plug in for current solutions like Salesforce CPQ/RCA addressing gaps and inefficiencies in current solutions like data alignment, difficulty in c...