The age of artificial intelligence (AI) and large language models (LLMs) has ushered in a new era of data-driven decision-making like never before. From predicting churn rates to forecasting revenue growth, digital tools like ChatGPT and Google Bard have accelerated the speed and depth of analysis to a whole new level.

This development is especially true in Revenue Operations (RevOps) teams, which are at the forefront of leveraging AI-powered insights for planning and optimization. The power of LLMs in generating language-processed data for planning cycles and sales performance definition is hard to deny. After all, these tools offer capabilities humans can only dream of when working with data:

Speed. Scale. Raw computing power.

Patterns at a glance

But with these levels of power, we also see RevOps teams over relying on LLM data when it comes to defining and determining key performance metrics. The risk in this is clear. LLMs can only present information that is fed to them—missing context, real-time data, and gray areas in sales interactions and plans.

From our experience working with RevOps teams across verticals, we’ve seen that data overreliance happens most when selecting key performance drivers. Without the “voice of sales rep” or real-world field dynamics, a purely quantitative approach to deciding what drives sales performance can lead to blind spots, lack of alignment, and disengaged field teams.

In this article, we argue that Revenue Operations teams must not be data overdependent when it comes to deciding what to measure for performance—especially when it comes to LLMs and language models in general. Let’s explore why, as well as what a more field-centric approach to performance definition can look like.

Why LLMs Are So Appealing in Revenue Operations

AI-powered tools like ChatGPT, Claude, or Gemini are only as smart as the information they have available to process. These LLMs have a neural network architecture that allows them to analyze information from several trillions of data points, including online content. In a revenue planning and optimization setting, these numbers are especially useful for:

a. Analyzing historical and cross-industry trends.
b. Generating industry-standard KPIs and baselines.
c. Predicting seller performance based on historical activities.
d. Offering competitive benchmarks and comparison for quota and territory design.

It’s no wonder these tools have become such an appealing point of entry for RevOps teams looking to up their game on speed and modernization in their planning cycles. For example, inputting the objective of identifying performance drivers for reps into an LLM can yield helpful observations, such as that top performers in the previous year have similar activity patterns. These might include:

a. Contacting 10+ new leads per day
b. Creating 5+ new opportunities per week
c. Booking at least 3 product demos per week
d. These insights are easy to measure, track, and model.

However, there is a catch. All of these outputs from LLMs are almost always pattern-based and lacking in context (or static in time)—missing the qualitative and situational insights of real-world sales conversations and team activities.

The Danger of Relying on LLM Data for Performance Definition

LLMs are trained on information and data that have been recorded and “perfected” in one way or another. But real-world activities and market conditions change all the time. If a revenue team defines important performance drivers by relying too heavily on language models, they may miss key variables that truly drive sales rep performance.

Here are three reasons why LLM dependency for performance definition can leave a sales strategy short.

1. Missing Field Realities and Market Conditions

Large language models only see what has been reported or fed to them through data pipelines and external sources. This means they can miss real-time market shifts and situational field realities that impact sales performance, such as:

a. New product launches or features from competitors
b. Changes in go-to-market strategies, such as new pricing models
c. Regional economic trends or shifts (key for global orgs)
d. Industry-specific regulatory changes

Imagine a revenue team setting compensation and performance metrics for medical device reps in the third quarter of the year. Relying solely on LLM-backed past performance data, the model may prioritize metrics such as “sales velocity” or “deal conversion rate”.

However, say that a new regulation has just been passed that causes hospitals to experience a delay in procurement. Reps are at the mercy of a longer-than-average sales cycle, one that is entirely out of their control.

LLMs would still suggest these activity indicators, as they can only rely on historical performance baselines or third-party data.

2. One-size-fits-all Metrics and Benchmarks Aren’t Motivating

LLMs are only as helpful as the information being inputted into them. There is a danger that these models may be overgeneralizing sales performance metrics for reps.

In other words, outputs from the model may be key performance indicators (KPIs) that sound generic but still apply across the board. Some LLM-backed sales performance drivers could look something like this:

a. Number of meetings booked
b. Close rate on submitted quotes
c. New demos signed per month

Lumped into a sales plan compensation, these metrics would demotivate reps in the field, especially if they are atypical for the vertical, account size, region, or sales cycle length. Say you have a sales rep in an emerging market like Africa, where buyer maturity and digital infrastructure may not be as high. The number of demo meetings will be lower simply due to the ground realities of educating more than closing deals.

Penalizing reps for lower conversion will demoralize them and damage morale. This is one reason large language models aren’t enough in sales for performance definition.

3. Field Reps Feel Like Robots, Not Agents

Motivation to reach KPIs for high performers is not just about offering incentives or bonuses. Sales reps are smart—they want to feel like agents of their own work, not “robots” being micro-managed by digital tools.

Sales plan and quota metrics that are overdesigned from data (without field collaboration) will feel robotic to reps in the field. Morale will drop.

At one Fortune 500 tech company, when LLM-derived performance metrics were rolled out with limited context or explanation to the field sales team, there was significant pushback. The metrics were based on benchmarks for productivity and sales velocity seen across the industry and historical data.

The European and Middle Eastern (EMEA) sales team felt these KPIs were unfair because they failed to account for longer sales cycles and the cultural differences in their buying patterns. After several feedback sessions with the field reps and managers, the company redefined the sales plan to include a new metric called “sales-qualified leads generated through events”. This made more sense to reps on the ground.

Field Dynamics: The Missing Puzzle Piece to Performance Definition

LLMs are excellent for finding patterns at scale and generating insights based on historical activities and external research. However, the missing piece in truly effective performance driver selection is ground truth from the field.

Sales reps are the lifeblood of a revenue organization and are uniquely qualified to offer feedback and ideas on what makes a sale move. While LLMs can suggest these, additional context and a push towards field collaboration can make or break a high-impact, holistic sales plan and quota compensation.

There are several sources of data points for metrics definition that are only available from the field, such as:

a. Sales reps themselves: What’s your pain point in closing? What drives conversion in your segment?
b. Frontline sales managers: What are your direct reports’ biggest obstacles and motivators?
c. Sales enablement: Are your playbooks and best practice guides being used effectively?

Market intelligence: Are there current trends (or shifts in trends) in your buyer personas due to competitor moves, changes in the regulatory landscape, or seasonality in demand?

We recently worked with a large retail company. After inputting sales rep activities into an LLM, the model found that one of the key drivers of rep performance was number of in-store visits per day. However, after speaking with the sales and field ops team, we found that store associates in a few regions were being penalized for lower sales volume simply because of traffic patterns.

Stores in certain regions of a country had huge traffic spikes on weekends. However, weekend hours were split across associates, leading to major swings in store traffic-to-sale conversion ratios daily. A manager proposed a hybrid metric: conversion by time slot (weekday vs. weekend) + average ticket size growth.

This was more representative of associate performance.

Balancing LLM Insights with Field Collaboration

Performance definition and measurement in the field sales context is only as accurate as the data being reported (and re-measured over time). If a RevOps team is interested in creating effective, trusted sales plan and quota metrics, LLMs will only get you so far.

LLMs excel at finding patterns at a macro level. However, if you are overdependent on LLMs without iterating and including field feedback, you may design an incentivized sales plan that does not make sense in practice, demotivating high performers.

To build a more balanced approach between LLM data and field-collaborated metrics for performance, follow this 4-step approach.

1. Use LLMs to Generate Hypotheses, Not Conclusions

LLMs should be used to generate hypotheses, not conclusions. Use large language models at the start of your planning cycles to ask questions.

2. Validate with the Field Sales Team and Managers

Take LLM-generated sales performance drivers and run them by the field reps and frontline managers. Conduct workshops, interviews, and feedback sessions with these key stakeholders to see what works, what doesn’t, and if there are any nuances to be considered.

3. Add Field-level Adjustments

Add unique variables based on:

a. Buyer journey length differences between regions, territories, and customer verticals.
b. Buyer maturity or purchasing preferences
c. Territory challenges (urban vs. rural, sales complexity)

4. Make Metrics Adaptive and Built to Evolve

Design metrics that can be adjusted over time based on feedback from reps and managers. Have a quarterly review cycle of all metrics where reps can propose changes based on new things that are happening in the field.

Real-world Example: Leveraging AI and Field Intel for Performance Metrics

Let’s look at an example of how this approach can work in a practical setting. In the case of a SaaS firm entering Southeast Asia for the first time, we started the sales plan and quota design process with LLMs to identify common sales performance drivers by inputting key business variables.

The baseline LLM suggestions for sales plan performance drivers for the region were as follows:

a. Number of product demos
b. Customer acquisition cost (CAC)
c. Time to close

However, several reps in Vietnam and the Philippines began flagging that these metrics did not account for their local buyers’ preferences in making a buying decision. Buyers were more relationship-driven in these countries, and community outreach and non-digital touchpoints (radio, local events) were important. This translated to higher touchpoint numbers and longer sales cycles.

RevOps worked with the field reps to add in new key performance drivers that better aligned with these buyer preferences in the new regions:

a. Number of community events hosted
b. Number of leads sourced from referral networks
c. Sales cycle quality score (based on buyer feedback)

This blended model was met with much more excitement from the field teams. In a short 6 months, the new metrics led to a 22% increase in regional conversion rate.

Conclusion

LLMs are great tools for Revenue Operations teams interested in unearthing patterns and benchmarking. However, these large language models are no substitute for field intelligence.

LLMs cannot be relied on to dictate sales plan design. High-impact metrics should include ground truth from field reps in the form of direct feedback, engagement metrics, and custom indicators that are specific to that organization or unique to certain regions or verticals.

For example, a “sales cycle quality score” for mid-market rep performance would be customized to the organization, while “number of meetings booked” would be a common metric found in most sales enablement data.

LLMs can only do so much when it comes to metrics definition. The next step in making your sales compensation more effective is incorporating field data, as well as real-world insights into your planning process.