How cost per query, model routing, and adoption patterns shape AI ROI inside modern CRM platforms.
Steve, as Chief Executive Officer of Insightly, how has your journey in leading a CRM platform shaped your perspective on the evolving economics of AI copilots?
Quite simply, I knew we had to figure out the economics of delivering an AI copilot in the CRM application in order to stay competitive in the CRM space. From day one of ideation, we prioritized query cost innovations alongside core development, ensuring seamless integration into Insightly CRM without margin pressure.
The market conversation around AI copilots often centers on subscription pricing. Why do you believe the more consequential metric is the underlying cost per query rather than the headline monthly fee?
I just think about the days where hotels would charge for wifi…that’s a crazy thought now. Why? Well first, cellular networks have become so powerful that you may not even need wifi during your stay. And second, it’s become as necessary to the hotel experience as a bed and a shower. In my mind, we had to get the cost per query down low enough so that use of the AI copilot could be included in our CRM plans without an additional fee. I realize that some CRM vendors have taken the ‘additional fee’ approach. We didn’t…we’re in it for the long game. As such, while we were developing the copilot, training it, and refining the user experience, we were simultaneously coming up with creative ways to reduce the cost per query so that we could include it in our offering without eroding our gross margins.
Model pricing can range widely depending on complexity and usage. How should executive teams think about this variability when evaluating long-term ROI?
We approached cost optimization by intentionally using a combination of LLM models based on the query type and complexity rather than defaulting to a single model for all workloads. Lightweight, high-volume, or low-risk tasks are routed to lower cost models, while more complex reasoning, multi-step generation, and customer-facing outputs are handled by higher performance models where quality and speed matter most. This tiered routing strategy allows us to balance performance and reliability with cost efficiency, ensuring we pay for premium intelligence only when it materially improves outcomes while keeping overall unit economics under control.
In CRM environments, usage patterns differ across sales, marketing, and customer success. What distinctions have you observed in how these teams interact with AI inside their daily workflows?
We look at usage patterns in many ways in an effort to understand how users are engaging with the Insightly CRM copilot, how it is delivering value, and if there are any quality gaps we need to address. DAU (daily active users), number of prompts per user per day, and average session duration are some of the ways we track usage. The CRM is dominated by sales users, so clearly the most activity is coming from them. With Insightly licensing, marketers will typically have a login to Insightly Marketing for marketing automation and customer success team members will have a login to Insightly Service. Both of these applications allow for a view of CRM data for cross-team alignment, but these users mostly stay within those applications rather than logging in to the CRM itself.
Two organizations may pay the same per-seat price yet experience very different economic outcomes. What factors create that divergence in value realization?
CRM is like any other application in that users get out of it what they put into it. A fully-adopted CRM is where the gold is. A recent research report that Insightly conducted with Ascend2 research shows that sales teams that fully adopt their CRM are 2x more likely to report a major boost in sales efficiency (59% vs. 23%). For so long, we heard that CRM adoption was hampered by complexity. Now that users have access to a conversational AI-copilot built into their existing subscription, we anticipate this objection to lessen – and even disappear. No longer hampered by keystrokes, mouse clicks and menus, CRM users can have the copilot do busy work for them and allow the CRM to fulfill its promise as the single source of truth for the business.
Insightly recently introduced a conversational AI Copilot within its CRM. What guiding principles shaped the product’s architecture to ensure query-level efficiency from day one?
We implemented an architecture which routes AI queries that users make (e.g. requests) based on task characteristics such as complexity, required reasoning depth, and customer visibility. Simpler, high-volume tasks are handled by cost-efficient models, while complex or high-stakes queries are escalated to more performant models.
Prompt optimization and intelligent model routing play a major role in controlling costs. How did your team approach these technical decisions to balance performance with predictability?
During the development process, we invested in prompt optimization — refining instructions for efficiency, implementing response caching where appropriate, and reducing unnecessary token usage while preserving critical context. In parallel, we establish a clear evaluation framework to determine model fit over time. We benchmarked models against defined quality thresholds, latency requirements, and cost-per-task targets, running structured comparisons before expanding usage. Moreover, selection is treated as an ongoing optimization exercise rather than a one-time decision, enabling continuous recalibration as pricing, model capabilities, and workload patterns evolve.
Features such as pipeline summaries, data hygiene checks, and follow-up recommendations are positioned as high-impact workflows. How did you determine which use cases truly justify incremental AI queries?
The result of this approach is improved unit economics and strategic flexibility. By avoiding reliance on a single model, we reduce cost volatility and mitigate platform risk. At the same time, the architecture allows rapid adoption of new models as they emerge, ensuring the product benefits from market innovation while maintaining disciplined cost management.
Mid-market companies often seek enterprise-grade automation without financial volatility. How does Insightly address the risk of unexpected AI expenses while maintaining a robust conversational experience?
If you are a software company building AI into your platform, you’ll need to be aware of the costs associated with it and ensure that you don’t erode your gross margin. Teams can mitigate this risk in several ways. First, plan for a prolonged beta period where you can look at usage patterns and therefore make predictions and budgets based on actual usage. Second, place limits (as generous as possible) on usage and plan to charge for overages. We go the extra mile with Insightly CRM as group account overages vs. charging overages per individual, allowing organizations to get the most out of their query limit across all users. And third, spend engineering time looking for ways to use multiple AI providers for queries, sending queries of varying complexity to different providers to keep costs down. When we first began developing the AI copilot in Insightly CRM, queries were costing us pennies; after finding efficiencies in refining query types and using multiple providers, we were able to reduce the cost per query to fractions of a penny.
What developments do you anticipate that will elevate “cost per query” to a board-level discussion, and how should organizations prepare today?
While AI is exciting and enticing to use in today’s business environment, there is a cost to the technology. If you are a software company building AI into your platform, you’ll need to be aware of the costs associated with it and ensure that you don’t erode your gross margin. Some companies will make an AI assistant or AI copilot as an add-on for an additional fee; that’s a sound strategy that ensures you aren’t in danger of potential losses in profitability. Building the AI feature into your existing offering without a price increase is a bit more risky. Teams can mitigate this risk in several ways. First, plan for a prolonged beta period where you can look at usage patterns and therefore make predictions and budgets based on actual usage. Second, place limits on usage and plan to charge for overages. And third, spend engineering time looking for ways to use multiple AI providers for queries, sending queries of varying complexity to different providers to keep costs down.
Steve Oriola
Chief Executive Officer, Insightly
Steve Oriola is the CEO of Unbounce Go-to-Market Solutions. He is a tenured CEO with more than two decades of experience scaling dynamic B2B SaaS platforms, including Act!, Constant Contact, Pipedrive, and Julius. He recently led Unbounce through the acquisition of Insightly CRM where the two companies effectively merged. He served as Executive in Residence at Bessemer Venture Partners where he participated in partner meetings and evaluated investment opportunities while providing advice and counsel to portfolio companies. Steve Oriola attended Boston University Questrom School of Business.