The UI isn’t the bottleneck

You know the screen I mean. A wall of drop-downs and radio buttons that looks the same as it did 20 years ago. Every vendor now sprinkles “AI” on top. A chatbot here. An assistant there. It feels like new paint on the same dashboard.

I’ve sat in vendor briefings where 10 people gave five different stories about what their AI actually does. Impressive demos. Fuzzy answers. Useful for marketing. Not useful for decisions.

The common belief is simple: AI should make quoting faster and admin setup easier. Fine goals. But that’s like putting a faster engine in a car stuck in traffic. Speed isn’t the constraint. Signal is.

Good in each system. Wrong when combined.

CPQ is where ERP, PLM, and CRM promises collide. The friction at quote time isn’t about pretty screens. It’s about truth under pressure.

From quoting tool to Intelligence Engine

Most teams still treat CPQ as a transaction tool that spits out a valid price. I don’t. I treat it as an intelligence engine.

The job of the front-end has changed. It’s no longer just a configuration tool. It should capture why a customer wants something, how sensitive they are to price or lead time, and what alternatives they would accept. It should guide the conversation and record the intent, not just the outcome.

Quotes are flight data recorders. Every interaction contains signals about fit, urgency, budget, and risk. If you don’t learn from it, you fly blind on the next deal.

Legacy CPQ captures the final configuration and the discount. That’s the bare minimum. Modern CPQ should capture the branches we didn’t take, the trade-offs the customer considered, and the nudges that moved them forward.

I’ve run machine learning on years of CPQ data. Won and lost quotes. When you include the reasons and the sequence of decisions, models start to route effort better than intuition. When you only store the last price and the last SKU, there’s nothing to learn.

A quote is a forecast. Treat it like one.

Use the quote to predict probability, cycle time, and margin risk. Then change the workflow based on that forecast. That is value.

The three pillars and what to ask vendors

The mechanism is straightforward when you build for it from day one:

First pillar - LLMs for conversation. Let the user speak naturally. Capture the why behind choices: “I need the shorter lead time even if it costs more.” That’s intent data. Not guesswork.

Second pillar - Symbolic AI for the inviolable rules. Product logic, pricing rules, approvals, compliance. This is where you prevent bad promises. It keeps the conversation honest.

Third pillar - Machine learning on interaction data. Feed the structured intent into models that predict win probability, discount elasticity, and approval risk. Route high-probability quotes for fast-track. Flag margin leakage. Suggest the next best move instead of the next best dropdown.

Legacy systems can bolt on the first pillar. Maybe some of the second. Almost none preserve the rich interaction history required for the third. They record what we sold, not how we got there.

I remember a mid-sized manufacturer where we compared two quarters of quotes. Same products, similar volumes. The only change was capturing a few simple intent fields and training a probability model on the full quote journey. We didn’t change the UI. We changed what the UI captured. Sales spent fewer cycles on low-probability deals. Discount variance tightened. Throughput went up without adding headcount.

Discount is a decision that should be informed by signals, not mood.

Counterpoint: isn’t UI still key for adoption? Yes, if the UI gets in the way, nobody uses it. But the system is only successful if it changes daily behavior. If all we did was modernize the look and add a chatbot, we didn’t change decisions. We decorated them.

What to ask in your next vendor session:

  • Show me how the front-end captures intent, not just selections. Where is the why stored, and in what structure?
  • Prove the separation between conversational logic and product rules. What can the LLM suggest, and what can it never break?
  • Walk me through the dataset you build from each quote. Which fields train your models? Wins and losses included?
  • Which decisions change in real time based on the data? Routing, approvals, pricing protection - show the workflow, not a score nobody uses.
  • How do you keep channel consistency? If one channel lags, customers will find it.

And measure the basics while you do this. Quote revision rate. Approval latency. Discount variance by win probability. Engineering escalations per 100 quotes. If your AI doesn’t move at least one of these within a quarter, it’s theater.

Be pragmatic about data. Data quality is not a debate. It is a measurable operational risk. Automation is removing manual error and freeing time for decisions that matter. Start with 12-24 months of quotes, including losses. Set the box first. Then decide what fits.

I’m skeptical of AI-washing too. I’ve seen the bolted-on assistants that claim 30-40 percent faster setup. Useful, but still slow. Useful, but not learning. The advantage goes to teams who build the data asset as they sell.

Board-ready takeaway:

The question is not “Does your CPQ have AI?” The question is “Is your quoting process building a data asset we can learn from?”

The goal isn’t a better quoting tool. It’s a smarter sales engine.