Your CPQ feels slow. That’s not the diagnosis.

Sales complains. Quotes take too long. Pricing debates never end. You start pricing a replacement project while everyone nods at the obvious culprit - the tool.

I have bad news and good news. The tool is rarely the root cause. Which means buying a new one won’t fix the real problem. That is good news if you care about outcomes, not logos.

What you are feeling is a decision problem. Not a software problem.

The friction shows up at quote time because that is where promises become real. If the commercial logic, product truth, and data signals do not agree, your CPQ becomes the messenger. We shoot the messenger too often.

A quote is a forecast. Treat it like one.

The real issue: your CPQ isn’t wired as a decision engine

Most executives I meet say the same thing: the CPQ is old, clunky, or missing features. Sometimes that is true. More often, the system is faithfully amplifying issues elsewhere - unclear portfolio rules, inconsistent pricing logic, and data that looks fine until systems have to agree.

When I dig into the numbers, I look for three things: margin leakage, wasted time, and lost learning. If you can’t explain where those come from, the brand of CPQ won’t save you. It will just make bad decisions faster.

The system is only successful if it changes daily behavior.

A proper CPQ analysis is not a feature audit. It is a strategic diagnosis of how your company decides under pressure. The goal is simple: turn CPQ from a transactional tool into a sensor network that tells you how you actually make money.

I run this diagnosis in three passes. Each one asks a decision question, not a tooling question.

The diagnosis: three questions about decision quality

1. Commercial health - are we quoting our strategy or our habits?

Most teams analyze what they sell. I analyze what they quote. There is a big difference. The quote log tells you what configuration paths sales actually takes, which prices stick, and where approvals pile up.

Start with simple metrics. Discount variance by segment. Revision rate per quote. Approval latency by deal size. If those drift up, margin is leaking. If revisions exceed 2 per quote on average, your rules are unclear. If approval latency crosses 48 hours routinely, you are training customers to expect concessions.

I remember a mid-sized manufacturer with “strict” discount policy on paper. The data showed 70 percent of won quotes had no discount at all, and most heavy discounts were on deals that still lost. That is not discipline. That is noise. We changed the conversation: which quotes deserve a discount based on win probability, and which do not.

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

Set the box first. Then decide what fits. Define which configurations are strategic, which are negotiable, and which are custom by exception. Your quote data already reveals the boundaries. You just need to read it.

Machine learning earns its place here by routing effort. Score quotes using 24-36 months of wins and losses. If probability is low, don’t spend a week polishing revisions. If probability is high, protect margin and speed approvals. Value shows up when behavior changes.

Remember the principle: if a metric cannot change a decision, it is noise. Favor the few signals that reliably shift behavior - predicted win rate, expected margin after discount, and escalation risk.

2. Ecosystem health - do the three clocks tell the same time?

This is the single source of truth problem. ERP, PLM, and CRM often look correct in isolation. Then sales tries to configure a real product with a real price and a real promise.

That is when the cracks appear. CPQ did not create them. CPQ revealed them.

Good in each system. Wrong when combined.

I use the three clocks metaphor: three clocks, all correct, all showing different time. PLM knows the latest engineering rules. ERP knows costs and availability. CRM knows the customer and terms. CPQ forces these clocks to agree at the quote. If they don’t, you get delays, overrides, or worse - promises you cannot keep.

Look for handoff failures. Manual attribute mapping. Cost updates lagging 2-4 weeks behind. Customer-specific pricing stored in spreadsheets. These are not IT sins. They are physics. People retype data because systems don’t align. Automation is removing manual error and freeing time for decisions that matter.

One warning sign is channel inconsistency. If e-commerce says one thing and sales quotes another, customers will exploit it. If one channel lags, customers will find it.

The fix is not more training. It is a decision pipeline. Which system is authoritative for each truth, how often it updates, and what happens when there is a conflict. Put that into CPQ rules so the same decision shows up everywhere, every time.

3. Data readiness - do we have a feedback loop or just dashboards?

Most companies have dashboards. Few have decisions that move because of them. If your win rate, cycle time, or discount discipline does not change month to month, your BI is just wallpaper.

Start by treating your quotes like flight data recorders. Every quote logs configuration choices, pricing, discounts, lead times, approvals, and outcome. Ignoring that set is like flying with the instruments off.

Use 12-48 months of wins and losses. Clean up the bare minimum - segment, product family, price band, approval path, time-to-quote. Data quality is not a debate. It is a measurable operational risk. Fix the fields that move decisions. Leave the rest for later.

Then connect transactional patterns to strategic outcomes. Which configurations win more often at higher margin. Which regions need approvals routed to pricing earlier. Where discount sensitivity is real and where it is wasted. You are not trying to predict everything. You are trying to route effort with better odds.

I like machine learning when it changes a decision. Route high-probability quotes to fast-track. Flag low-probability quotes for minimal revisions or a clear walk-away. Use a simple threshold - say 70 percent probability - to protect margin and time. A model nobody uses is just math with a salary.

Expect a counterargument here: our tool is too old. Sometimes that is true. But even then, this diagnosis pays for itself by preventing a migration of bad logic. Replace the tool when you know which decisions you want it to enforce, not before.

Data quality is not a debate. It is a measurable operational risk.

What this changes on Monday

Define one weekly loop. Review the three signals: predicted win rate distribution, discount variance by segment, and approval latency. Decide one change. Test it for two weeks. Measure. Roll it in or roll it back. The cadence matters more than the dashboard.

When smaller manufacturers do this, they get the best advantage they already have - speed. I love the moment when the CEO says yes in the meeting and we change the rule the same day. Speed only matters when the data is correct. Quick decisions with wrong data just create faster mistakes.

I shifted my own practice after a decade of implementations where learning stopped at order intake. Orders went through. Nobody complained. That silence is a signal. In CPQ, predictable beats impressive. Predictable is what scales.

If you take nothing else from this, take this: a CPQ assessment is an audit of decision quality. Not a software review. Run it before you buy anything.

A CPQ replacement without a diagnosis is an expensive guess.

A real analysis gives you a blueprint for a more predictable sales engine - what to quote, what to automate, and what to stop doing. Set the box first. Then decide what fits.

Last line for the board: fix decisions before tools.