You can see it in every roadmap. The flagship product line gets modeled first. The rest stays in spreadsheets with notes in the margins and tribal rules that only three people know. Deals get done, but the cost is hidden: duplicate work, slow coaching, and quiet errors that never make a post-mortem.

We have told ourselves this is normal. It is not. It is a price signal. Modeling is still expensive, so we ration it.

When the cost of reasoning falls, scope is what changes.

The Market Lesson From Machine Translation

Fifteen years ago, a human translator could process about a thousand words a day. That rate set the price of everything from manuals to websites. In 2005, the global translation market sat around three billion dollars. Then computer-assisted and neural approaches arrived. Throughput rose tenfold. Every analyst predicted translators would be displaced. Instead the market expanded toward seventy billion as more content became economically worth translating. Work flooded into spaces that had been ignored because they were too costly to touch.

That is the shape to watch. When an activity gets ten times cheaper, you do not get the same work faster. You get a different market.

CPQ’s Expensive Middle: Where Value Leaks

Configuration is not the bottleneck anymore. Solvers are accurate. The choke point is the middle: turning loose customer language, internal constraints, and pricing strategy into a structured starting point a solver can use. That translation and the modeling beneath it are still rationed. So we apply them to the head of the portfolio and accept an Excel-shaped long tail.

What does that rationing cost you today?

  • Only the top products get real governance. The rest are shadow-quoted with optimistic assumptions that engineering cleans up later.

  • New sellers climb a cliff of implicit rules. Coaching happens by corridor, not by system.

  • Adoption fights gravity. If the official path is slower than the familiar spreadsheet, people route around it.

According to Gartner, 70 percent of large organizations will adopt AI-based forecasting by 2030 to improve decision speed and collaboration. Jan Snoeckx at Gartner framed the value clearly: improved strategic decision making, faster responses to market changes, and enhanced collaboration workflows. The same forces apply to quoting. Teams are not asking for another screen. They are asking for faster, shared understanding that they can trust.

Why This Moment Is Different

We just crossed a threshold where machines can help translate needs into structured inputs at a useful level. Not perfectly. Usefully. That changes the economics of modeling.

Here is the pattern I see working in the field:

  • Use retrieval to gather product constraints, dependencies, and sales arguments from what you already have - price books, CAD notes, service manuals, training decks.

  • Use a language model to sit on top and propose a structured starting point from plain-language needs, plus the narrative that explains why.

  • Use a constraint solver underneath to guarantee what matters: compatibility, pricing math, and documentation.

AI on top, rules underneath. Conversation accelerates. Correctness holds.

One warning sign from the marketing world is worth repeating. Gartner reports that only 5 percent of leaders who use generative AI as a tool in isolation see significant business gains. The lesson is simple: bolt-ons disappoint. Architecture wins. If AI is a widget in your CPQ, expect a novelty spike and a slow fade. If AI is a layer in your system design, expect compounding value.

Leaders feel the stakes. Gartner found that 82 percent of executives expect their company’s identity to change materially in response to AI. That is not a tooling comment. It is a strategy comment. Ewan McIntyre at Gartner put it plainly for CMOs, and it maps to sales leadership too: build hybrid human-AI teams and future-ready leadership capabilities, often without more headcount. In quoting, that means your experts define the boundaries, the system proposes, and humans stay in charge of judgment and relationship.

A Practical Architecture For The Long Tail

Let me be concrete. This is the minimal design I recommend when teams want the long tail out of Excel without gambling margin.

  • Knowledge retrieval: Centralize product, option, and dependency knowledge in a searchable store. PDFs and tribal notes are fair game. Treat this as living content with ownership per module.

  • Translation layer: A swappable LLM that turns needs into structured option sets and narrative. It should show its work. If it cannot explain why it prefers Option B over Option A, it is not ready for your sellers.

  • Deterministic core: A constraint solver and price engine that cannot be overruled by eloquence. Let the LLM be brilliant about the conversation, never about the math.

  • Surfaces: Two thin interfaces that reuse the same brain. One for customers to self-educate and arrive warm. One inside your CPQ to guide sellers to a credible starting point.

  • Change loop: A weekly review where product owners approve new constraints the system inferred from documents or field feedback. No committees, just ownership.

Two craft notes from recent projects:

  • Write variant descriptions that include a “when not” line. It sharpens recommendations because the system can weigh disqualifiers, not just specs.

  • Borrow from supply chain’s “touchless forecasting” idea. Where history is thin, let the system propose starter prices for new bundles based on analogous patterns, then require human sign-off. You get speed without pretending the model is omniscient.

Put AI on top of explicit, testable logic - not in place of it.

Who Moves Ahead When Modeling Gets Cheap

When the cost to build a usable model drops by an order of magnitude, the portfolio you can responsibly model changes. You do not stop modeling your head products. You add the long tail you have ignored for ten years.

Quiet winners will share three traits:

  • They reframe the competitor. The real rival to your initiative is not another CPQ vendor. It is the spreadsheet that has been good enough. Treat Excel as the incumbent channel and measure defection from it.

  • They design for explainability. If the system cannot defend its choices, adoption will plateau at the first complex deal. Narrative clarity beats shiny UI.

  • They scale ownership, not heroics. Governance is not a meeting. It is a named owner per module, a safe path to ship changes weekly, and a test suite that proves nothing broke.

The slow drifters will keep waiting for perfect data before they ship anything. They will run long workshops to capture every edge case. They will treat AI as a feature in a release note. Six months later, the field will be back in Excel and nobody will be surprised.

If you want an external signal that the ground is moving, look at how analysts reshuffle their quadrants. Even the CPQ vendor table is changing shape as pricing and configuration intersect new AI patterns. That is not the point of this article, but it is the weather pattern around it.

What should you do this quarter?

  • Pick one product line that never made it into CPQ because the business case did not clear. Model it in a week. Use the architecture above. Hold the bar at credibility, not perfection.

  • Instrument the path from first need to first price. Time it. Remove one delay that has nothing to do with product complexity and everything to do with handoffs.

  • Expose why prices move. Sellers trust systems that can show their reasoning. Your discount governance will improve as a byproduct.

One last leadership point. Sharon Cantor Ceurvorst at Gartner challenged executives to stop prioritizing execution and lead through strategic insight. In CPQ terms: stop measuring your program by the number of modeled variants and start measuring it by the number of credible conversations the system can support without a meeting. That is the compounding metric.

When the long tail leaves Excel, you do not just sell more. You learn faster than competitors who still ration modeling.

If modeling were ten times cheaper, what would you finally bring in from the cold?