Signal over Noise: Measuring AI’s Impact at Hyper-Growth Speed

Signal over Noise: Measuring AI’s Impact at Hyper-Growth Speed

Leaders from Ramp, Lovable and Harvey share how to turn rapid AI experimentation into measurable customer value.

Share

Based on a session at FRONTIER by Planhat, on 10 November 2026.

A customer spends months evaluating your product, completes a detailed pilot and finally signs the contract. Then customer success schedules a 45-minute kickoff that repeats everything they’ve already discussed.

Harvey encountered exactly this problem as it scaled its post-sales organization. Customers wanted to start making progress.

Harvey responded by removing unnecessary material from its kickoffs and improving the handover between sales and customer success. The lesson extends to AI adoption: the ability to build and automate workflows quickly makes it even more important to identify which work actually creates value.

Build quickly, measure what matters and be prepared to remove what doesn't work.”

“

Based on a session at FRONTIER by Planhat, on 10 November 2026.

A customer spends months evaluating your product, completes a detailed pilot and finally signs the contract. Then customer success schedules a 45-minute kickoff that repeats everything they’ve already discussed.

Harvey encountered exactly this problem as it scaled its post-sales organization. Customers wanted to start making progress.

Harvey responded by removing unnecessary material from its kickoffs and improving the handover between sales and customer success. The lesson extends to AI adoption: the ability to build and automate workflows quickly makes it even more important to identify which work actually creates value.

Build quickly, measure what matters and be prepared to remove what doesn't work.”
“

Based on a session at FRONTIER by Planhat, on 10 November 2026.

A customer spends months evaluating your product, completes a detailed pilot and finally signs the contract. Then customer success schedules a 45-minute kickoff that repeats everything they’ve already discussed.

Harvey encountered exactly this problem as it scaled its post-sales organization. Customers wanted to start making progress.

Harvey responded by removing unnecessary material from its kickoffs and improving the handover between sales and customer success. The lesson extends to AI adoption: the ability to build and automate workflows quickly makes it even more important to identify which work actually creates value.

“Build quickly, measure what matters and be prepared to remove what doesn't work.”

Growth requires a different approach to prioritization

At Lovable, headcount is expected to grow from approximately 100 employees at the beginning of 2026 to 500 by year-end. Ramp has grown to more than 2,000 employees, with over 200 people in post-sales, while Harvey serves some of the world’s largest law firms and Fortune 500 companies.

Lovable serves a substantial long tail, Harvey works extensively with high-touch enterprise relationships and Ramp covers a broad range of segments. All three face the pressure of supporting rapid growth while improving their operating models.

At Harvey, the initial instinct was to standardize customer success activities through templates and presentations. The team quickly realized that creating more materials was consuming time that could be better spent elsewhere. Instead, it began using Harvey itself to make information such as product roadmaps accessible to customers, reducing the need for CSMs to prepare and distribute the same information repeatedly.

Ramp applies a similar principle to its operating cadence. Its post-sales teams are encouraged to develop and release improvements quickly, then evaluate whether those changes are affecting the metrics that matter. When an approach fails to produce the intended result, the team adjusts rather than continuing to invest in the original plan.

Data quality also affects measurement.”

“

Growth requires a different approach to prioritization

At Lovable, headcount is expected to grow from approximately 100 employees at the beginning of 2026 to 500 by year-end. Ramp has grown to more than 2,000 employees, with over 200 people in post-sales, while Harvey serves some of the world’s largest law firms and Fortune 500 companies.

Lovable serves a substantial long tail, Harvey works extensively with high-touch enterprise relationships and Ramp covers a broad range of segments. All three face the pressure of supporting rapid growth while improving their operating models.

At Harvey, the initial instinct was to standardize customer success activities through templates and presentations. The team quickly realized that creating more materials was consuming time that could be better spent elsewhere. Instead, it began using Harvey itself to make information such as product roadmaps accessible to customers, reducing the need for CSMs to prepare and distribute the same information repeatedly.

Ramp applies a similar principle to its operating cadence. Its post-sales teams are encouraged to develop and release improvements quickly, then evaluate whether those changes are affecting the metrics that matter. When an approach fails to produce the intended result, the team adjusts rather than continuing to invest in the original plan.

Data quality also affects measurement.”
“

Growth requires a different approach to prioritization

At Lovable, headcount is expected to grow from approximately 100 employees at the beginning of 2026 to 500 by year-end. Ramp has grown to more than 2,000 employees, with over 200 people in post-sales, while Harvey serves some of the world’s largest law firms and Fortune 500 companies.

Lovable serves a substantial long tail, Harvey works extensively with high-touch enterprise relationships and Ramp covers a broad range of segments. All three face the pressure of supporting rapid growth while improving their operating models.

At Harvey, the initial instinct was to standardize customer success activities through templates and presentations. The team quickly realized that creating more materials was consuming time that could be better spent elsewhere. Instead, it began using Harvey itself to make information such as product roadmaps accessible to customers, reducing the need for CSMs to prepare and distribute the same information repeatedly.

Ramp applies a similar principle to its operating cadence. Its post-sales teams are encouraged to develop and release improvements quickly, then evaluate whether those changes are affecting the metrics that matter. When an approach fails to produce the intended result, the team adjusts rather than continuing to invest in the original plan.

“Data quality also affects measurement.”

Experiment freely, then scale what works

AI has made it possible for more employees to build applications and workflows without relying on dedicated engineering teams. That increases the volume of experimentation, but it also creates the risk of investing in too many solutions without a clear understanding of their value.

At Lovable, the emphasis is on allowing people to experiment before identifying which applications should become established organizational capabilities. Once a project demonstrates value, it needs an owner, a defined objective and a way to measure its contribution to the business.

The company’s internal customer success platform, Catalyst, emerged through this approach. An employee developed an application that proved useful, and its adoption grew until it became a tool used across the approximately 60-person CS team. The transition from individual experiment to shared infrastructure happened because the application addressed a recurring operational need.

Ramp encourages employees to build alternatives even when an existing tool addresses the problem. This creates duplication but also familiarity with AI and ideas that may outperform current systems. Internal hackathons surface promising projects, and the organization is prepared to replace established tools with better employee-built alternatives.

Experiment freely, then scale what works

AI has made it possible for more employees to build applications and workflows without relying on dedicated engineering teams. That increases the volume of experimentation, but it also creates the risk of investing in too many solutions without a clear understanding of their value.

At Lovable, the emphasis is on allowing people to experiment before identifying which applications should become established organizational capabilities. Once a project demonstrates value, it needs an owner, a defined objective and a way to measure its contribution to the business.

The company’s internal customer success platform, Catalyst, emerged through this approach. An employee developed an application that proved useful, and its adoption grew until it became a tool used across the approximately 60-person CS team. The transition from individual experiment to shared infrastructure happened because the application addressed a recurring operational need.

Ramp encourages employees to build alternatives even when an existing tool addresses the problem. This creates duplication but also familiarity with AI and ideas that may outperform current systems. Internal hackathons surface promising projects, and the organization is prepared to replace established tools with better employee-built alternatives.

Experiment freely, then scale what works

AI has made it possible for more employees to build applications and workflows without relying on dedicated engineering teams. That increases the volume of experimentation, but it also creates the risk of investing in too many solutions without a clear understanding of their value.

At Lovable, the emphasis is on allowing people to experiment before identifying which applications should become established organizational capabilities. Once a project demonstrates value, it needs an owner, a defined objective and a way to measure its contribution to the business.

The company’s internal customer success platform, Catalyst, emerged through this approach. An employee developed an application that proved useful, and its adoption grew until it became a tool used across the approximately 60-person CS team. The transition from individual experiment to shared infrastructure happened because the application addressed a recurring operational need.

Ramp encourages employees to build alternatives even when an existing tool addresses the problem. This creates duplication but also familiarity with AI and ideas that may outperform current systems. Internal hackathons surface promising projects, and the organization is prepared to replace established tools with better employee-built alternatives.

Measure the business impact, not the volume of AI activity

The abundance of AI applications makes measurement particularly important. Companies can track how many employees use a tool, how many agents they’ve deployed or how much time a workflow appears to save. But these indicators don’t necessarily establish whether the business or its customers are better off.

Ramp has developed a framework centered on contribution profit, which is connected to the economics of its consumption-based business. The company calculates how much contribution profit its post-sales organization generates per minute or hour of employee time, giving teams a way to evaluate whether their activities are producing greater commercial output.

New tools must improve efficiency and contribution profit per unit of time without reducing customer satisfaction. Ramp uses this to evaluate the economics of customer success, rather than as an individual performance measure.

Harvey measures success through a different framework, which it calls the bar. Customers are assessed against adoption, embedded value and the outcomes achieved through the product. Embedded value considers how agents are used within the customer’s workflows and connected to other systems, while outcomes focus on the work those capabilities allow customers to accomplish.

Harvey establishes a minimum level customers should reach in their first 90 days, then works toward the broader bar. For a law firm, an outcome might be a junior associate completing a complex assignment before a deadline.

Lovable focuses on what it calls load-bearing apps: applications that have progressed beyond experimentation and become part of an organization’s real operations. A prototype built with dummy data demonstrates what the technology can do, but an application connected to business systems and used in daily work provides stronger evidence of value.

Customer requests for additional workflows, seats or resources also demonstrate demand for the capabilities they’ve experienced.

Measure the business impact, not the volume of AI activity

The abundance of AI applications makes measurement particularly important. Companies can track how many employees use a tool, how many agents they’ve deployed or how much time a workflow appears to save. But these indicators don’t necessarily establish whether the business or its customers are better off.

Ramp has developed a framework centered on contribution profit, which is connected to the economics of its consumption-based business. The company calculates how much contribution profit its post-sales organization generates per minute or hour of employee time, giving teams a way to evaluate whether their activities are producing greater commercial output.

New tools must improve efficiency and contribution profit per unit of time without reducing customer satisfaction. Ramp uses this to evaluate the economics of customer success, rather than as an individual performance measure.

Harvey measures success through a different framework, which it calls the bar. Customers are assessed against adoption, embedded value and the outcomes achieved through the product. Embedded value considers how agents are used within the customer’s workflows and connected to other systems, while outcomes focus on the work those capabilities allow customers to accomplish.

Harvey establishes a minimum level customers should reach in their first 90 days, then works toward the broader bar. For a law firm, an outcome might be a junior associate completing a complex assignment before a deadline.

Lovable focuses on what it calls load-bearing apps: applications that have progressed beyond experimentation and become part of an organization’s real operations. A prototype built with dummy data demonstrates what the technology can do, but an application connected to business systems and used in daily work provides stronger evidence of value.

Customer requests for additional workflows, seats or resources also demonstrate demand for the capabilities they’ve experienced.

Measure the business impact, not the volume of AI activity

The abundance of AI applications makes measurement particularly important. Companies can track how many employees use a tool, how many agents they’ve deployed or how much time a workflow appears to save. But these indicators don’t necessarily establish whether the business or its customers are better off.

Ramp has developed a framework centered on contribution profit, which is connected to the economics of its consumption-based business. The company calculates how much contribution profit its post-sales organization generates per minute or hour of employee time, giving teams a way to evaluate whether their activities are producing greater commercial output.

New tools must improve efficiency and contribution profit per unit of time without reducing customer satisfaction. Ramp uses this to evaluate the economics of customer success, rather than as an individual performance measure.

Harvey measures success through a different framework, which it calls the bar. Customers are assessed against adoption, embedded value and the outcomes achieved through the product. Embedded value considers how agents are used within the customer’s workflows and connected to other systems, while outcomes focus on the work those capabilities allow customers to accomplish.

Harvey establishes a minimum level customers should reach in their first 90 days, then works toward the broader bar. For a law firm, an outcome might be a junior associate completing a complex assignment before a deadline.

Lovable focuses on what it calls load-bearing apps: applications that have progressed beyond experimentation and become part of an organization’s real operations. A prototype built with dummy data demonstrates what the technology can do, but an application connected to business systems and used in daily work provides stronger evidence of value.

Customer requests for additional workflows, seats or resources also demonstrate demand for the capabilities they’ve experienced.

Move quickly, but measure deliberately

Rapid experimentation creates another challenge: determining which change produced a particular improvement.

When several tools and workflows are introduced simultaneously, their effects become difficult to separate. Ramp addresses this by trying to release one meaningful change at a time, then reviewing performance after approximately 15 and 30 days.

If early evidence suggests an experiment isn’t working, the team can investigate, stop it and redirect resources. Two to four weeks of positive data may justify further investment, although it won’t establish the long-term effect with certainty.

In regulated industries, iteration must account for the consequences of errors. Harvey emphasizes reliable sources and citations that allow users to verify output. Lovable advocates introducing AI into defined processes before expanding its scope.

Move quickly, but measure deliberately

Rapid experimentation creates another challenge: determining which change produced a particular improvement.

When several tools and workflows are introduced simultaneously, their effects become difficult to separate. Ramp addresses this by trying to release one meaningful change at a time, then reviewing performance after approximately 15 and 30 days.

If early evidence suggests an experiment isn’t working, the team can investigate, stop it and redirect resources. Two to four weeks of positive data may justify further investment, although it won’t establish the long-term effect with certainty.

In regulated industries, iteration must account for the consequences of errors. Harvey emphasizes reliable sources and citations that allow users to verify output. Lovable advocates introducing AI into defined processes before expanding its scope.

Move quickly, but measure deliberately

Rapid experimentation creates another challenge: determining which change produced a particular improvement.

When several tools and workflows are introduced simultaneously, their effects become difficult to separate. Ramp addresses this by trying to release one meaningful change at a time, then reviewing performance after approximately 15 and 30 days.

If early evidence suggests an experiment isn’t working, the team can investigate, stop it and redirect resources. Two to four weeks of positive data may justify further investment, although it won’t establish the long-term effect with certainty.

In regulated industries, iteration must account for the consequences of errors. Harvey emphasizes reliable sources and citations that allow users to verify output. Lovable advocates introducing AI into defined processes before expanding its scope.

Reliable data underpins every AI initiative

AI applications also depend on the quality and accessibility of the information they use. For established businesses, data may be distributed across decades of systems, with inconsistent definitions and incomplete records. These problems become more visible when agents attempt to execute workflows across departments.

Organizations generally need either a shared data environment with a consistent layer for interpreting information or reliable connections between their existing systems of record. The appropriate architecture depends on the business, but agents need dependable access to the information required for their work.

Data quality also affects measurement. If teams can’t connect AI activity to customer behavior, operational outcomes or financial performance, they’ll struggle to distinguish genuine improvements from increased activity.

Reliable data underpins every AI initiative

AI applications also depend on the quality and accessibility of the information they use. For established businesses, data may be distributed across decades of systems, with inconsistent definitions and incomplete records. These problems become more visible when agents attempt to execute workflows across departments.

Organizations generally need either a shared data environment with a consistent layer for interpreting information or reliable connections between their existing systems of record. The appropriate architecture depends on the business, but agents need dependable access to the information required for their work.

Data quality also affects measurement. If teams can’t connect AI activity to customer behavior, operational outcomes or financial performance, they’ll struggle to distinguish genuine improvements from increased activity.

Reliable data underpins every AI initiative

AI applications also depend on the quality and accessibility of the information they use. For established businesses, data may be distributed across decades of systems, with inconsistent definitions and incomplete records. These problems become more visible when agents attempt to execute workflows across departments.

Organizations generally need either a shared data environment with a consistent layer for interpreting information or reliable connections between their existing systems of record. The appropriate architecture depends on the business, but agents need dependable access to the information required for their work.

Data quality also affects measurement. If teams can’t connect AI activity to customer behavior, operational outcomes or financial performance, they’ll struggle to distinguish genuine improvements from increased activity.

The strongest signal is sustained customer value

The strongest AI initiatives will be those that become dependable parts of how a business operates, producing outcomes that customers recognize and value. Build quickly, measure what matters and be prepared to remove what doesn’t work.

The strongest signal is sustained customer value

The strongest AI initiatives will be those that become dependable parts of how a business operates, producing outcomes that customers recognize and value. Build quickly, measure what matters and be prepared to remove what doesn’t work.

The strongest signal is sustained customer value

The strongest AI initiatives will be those that become dependable parts of how a business operates, producing outcomes that customers recognize and value. Build quickly, measure what matters and be prepared to remove what doesn’t work.

Don't miss a beat.

By submitting this form I agree that Planhat may collect, process and retain my data pursuant to its Privacy Policy.

Don't miss a beat.

By submitting this form I agree that Planhat may collect, process and retain my data pursuant to its Privacy Policy.

Don't miss a beat.

By submitting this form I agree that Planhat may collect, process and retain my data pursuant to its Privacy Policy.