You Can't Enrich Your Way Out of Bad Operations

AI is quietly becoming part of every GTM function. We use it to find prospects, enrich customer records, personalize outreach, make calls, take meeting notes, update CRM fields, analyze conversations and forecast revenue. And honestly, I find that exciting: we're getting closer to a world (some companies are already living in it) where commercial teams spend far less time on admin and far more time with their customers.

But before we get carried away, we need to address the elephant in the room, the question I left you with in my last article: what happens when we unleash all these capabilities on an organization whose processes and data haven't been particularly reliable for the last few years?

Some of you already know that things in your GTM organization aren't all sunshine and rainbows, and you're nervous about letting a bunch of rogue AI agents loose in that environment. You're right to be.

AI can capture more information, but does it understand what it means for your business?

Let's take a familiar example. Your organization rolls out AI-powered call intelligence with automated CRM updates. The system records conversations, summarizes customer requirements, picks out next steps and updates the opportunity, which alone is a big improvement over relying on salespeople to type everything in themselves.

Now imagine your sales process hasn't been looked at in years:

• Your qualification criteria are inconsistent.
• Your opportunity stages don't reflect how customers actually buy.
• And nobody has clearly defined which buying signals should move a deal forward or change its forecast category.

The AI will happily capture what the customer said. Deciding what that means for the deal is a different matter, and it depends entirely on the commercial methodology your organization has established.

What AI can capture

• The customer says they're interested
• A potential budget, timeline and decision-maker
• A recommended next action

→

What your methodology has to decide

• Is this opportunity now qualified?
• Should it move to the next stage?
• Should its forecast probability change?

If that methodology is fuzzy, the AI has no choice but to find its way through the fog. It may apply one interpretation consistently, infer rules from historical patterns, or produce recommendations that look perfectly reasonable but don't match how your business wants to operate. And because those recommendations can now be applied to thousands of records at once, a small inconsistency can quickly snowball across your entire pipeline.

AI can dramatically improve how we collect and process information. But we still need to define what that information means and how the business should act on it.

So what do we do about years of operational debt?

As an operator, I know it isn't realistic to expect perfectly maintained CRM systems and processes. Businesses grow and evolve, and especially in the startup and scale-up world, the only constant is change. RevOps and Sales Operations teams are busy keeping the business running while getting buried under ever-growing operational debt.

You know the drill: an outdated field here, an undocumented workflow there. A qualification methodology that was introduced but never consistently adopted. A handover process that worked when the company had ten customers but doesn't quite work with a thousand. On its own, each of these seems harmless. Combined, they make the information flowing through your GTM systems increasingly difficult to interpret, and now we're asking AI to make sense of it.

The good news is that AI also gives us new ways to pay down some of that debt. But I think we need to be deliberate about the order in which we tackle it.

1. 1Understand what you're fixingWhich decisions suffer from unreliable data?
2. 2Recover what you canAI and enrichment, without inventing anything
3. 3Fix the processSo the data stops drifting again
4. 4Introduce AI you can evaluateJudge the output from day one
5. 5Make it lastOwnership, controls and capacity

Step 1: Understand what you're actually trying to fix

Before launching a large-scale CRM cleanup or yet another AI implementation, I'd want to understand where poor operational data is actually hurting the business. I'd start with a few questions:

• Which commercial decisions are we struggling to make because our information is unreliable?
• Where are people spending unnecessary time collecting, correcting, or interpreting data?
• Which processes are creating the most inconsistencies?
• What information is genuinely required to run the business effectively?
• Which issues have the greatest impact on revenue, efficiency, customer experience, or risk?

The objective is to understand which operational weaknesses matter and why. Trying to clean every single field in your CRM is a classic way to boil the ocean.

Step 2: Use AI and enrichment to recover what you can

Once we understand the problem, I would absolutely use the technology available to us. Enrichment can help complete company and contact information. AI can pull relevant details out of emails, meeting notes and call transcripts (where the right access and permissions exist), analyze historical records for inconsistencies, gaps, duplicates and unusual patterns, and suggest corrections or flag the records that need a human eye. This is where I see enormous potential.

But beware: we need to separate information we can recover with reasonable confidence from information that simply doesn't exist. If a customer never mentioned a budget, we shouldn't invent one. And if the definition of a qualified opportunity changed three times over the years, we need to understand those changes before treating all historical records as comparable.

Enrichment can help us recover information. It cannot automatically recover the business logic that was missing when the information was created.

Step 3: Fix the process generating the data

This is probably the most important step. Let's say you've enriched your accounts, cleaned your pipeline, standardized some fields and corrected historical inconsistencies. If salespeople are still qualifying opportunities differently, your pipeline data will start drifting again. If handovers between Sales and Customer Success remain unclear, customer information will keep slipping through the cracks. And if nobody owns the definition of a critical KPI, the reporting inconsistencies will return.

That's why data quality can't be a one-off cleanup project. You need to look at the processes that generate information in your GTM, document them, enable your teams and give your AI agents clear guidance. In practice, that means you need to:

• Define your qualification criteria.
• Define the mandatory checks for deal progression.
• Assign responsibility to teams or people (or AI agents) for capturing and documenting information at the right time.
• Bring in automation to help you maintain the new setup.
• Be prepared to handle exceptions.

Importantly, this shouldn't mean introducing another twenty mandatory CRM fields. Quite the opposite: good AI-driven operational design makes reliable data a natural output of doing the work, rather than an additional administrative task.

Step 4: Introduce AI into a process you can actually evaluate

Once the underlying process is clear, AI can take on significantly more responsibility. I still wouldn't hand every AI-powered workflow the keys to your business-critical records on day one. Instead, I'd define what success looks like and start evaluating the output immediately:

Depending on the risk, some actions can be fully automated, while others need approval, sampling or exception-based review. The goal is to increase automation as confidence grows, without wrapping permanent manual controls around every action. Ultimately, AI in operations should reduce unnecessary work while improving the quality and consistency of execution.

Step 5: Make sure the improvements actually last

This is the danger zone, and where I think most operational initiatives fall short. We clean the data, redesign the process, introduce automation and move on to the next project. Six months later, the qualification criteria have changed, an integration has quietly stopped working, or an AI workflow is still following outdated rules, and we're back to square one.

That's why operational improvement needs ongoing ownership, a few meaningful controls and enough capacity to maintain what we've built. Someone has to make sure our processes and automations keep reflecting how the business actually operates.

We have an opportunity to rebuild operations differently

I don't think the answer is to put AI adoption on hold until every CRM field is clean and every process is perfectly documented. We'd be waiting forever. Instead, we can use AI to help pay down the operational debt we've accumulated while redesigning the processes that created it.

Start with the areas that matter most to the business. Recover and enrich what you can, fix the underlying process, introduce automation and measure whether it's actually improving outcomes. None of this needs to become a massive transformation program.

The real opportunity is to use AI as a reason to rethink how our GTM operations should work in the first place.

To help you get started, I've turned the five steps into a free AI Readiness Checklist for GTM Operations. It's an Excel template your team can work through in about an hour, from finding where unreliable data hurts the business to giving every process and automation an owner.

Free resource · Excel checklistAI Readiness Checklist for GTM Operations →

Once that groundwork is done, the next challenge is already waiting around the corner: keeping everything working while the business continues to evolve. Maintenance is a muscle, and like any muscle, it only gets stronger with regular training.

In my next article, Operational Excellence Without Bureaucracy, I'll get practical about how to build that muscle: the routines, documentation and ownership that keep your operations reliable long after the cleanup, without turning RevOps into a bureaucratic bottleneck.