Operationalizing AI: A Practical Guide for Growth Companies
Insights from Summit’s Cathy Tanimura on how teams can build and execute their AI strategy
AI is top of mind for many company leaders, whether driven by desire to gain an edge or concern about falling behind. While the hype is still substantial, real use cases and capabilities have emerged and continue to emerge rapidly. From our perspective, today’s challenge is not whether to engage with AI, but how to move from broad mandates to operational reality.
For most companies, this starts with internal applications—using AI to improve productivity and automate processes. But for software companies in particular, the imperative goes further: embedding AI directly into core product experiences. Done thoughtfully, this can unlock new forms of customer value and create lasting competitive advantage.
What follows is a practical AI guide designed to help growth-stage companies begin and build their AI strategy. We focus on three high-leverage areas: individual productivity, automation and product experience.
1. Productivity: Practical Applications for Daily Work
One of the easiest and most immediate ways for companies to realize value from AI is through personal productivity tools that take human input and return content in text or another form, such as code, image, audio or video. The goal is simple: make individuals (and their teams) more effective with minimal disruption. Here are the most widespread “individual” use cases we see today:
General-Purpose Chatbots
Chatbots like ChatGPT (OpenAI), Claude (Anthropic) and Gemini (Google) are the most common types of personal productivity AI tools. Many organizations are finding that signing enterprise licenses providing access to most or all employees helps to drive broad usage. Each provider / model has strengths and trade-offs. I recommend piloting a few options and choosing the one that best fits your team’s needs and workflows, which will increase the likelihood of adoption for daily work.
Developer Tools
AI coding assistants have become nearly ubiquitous on development teams, helping developers write code faster, find bugs and refactor codebases. (Read more on AI-Accelerated Software Development here.) A variety of standalone tools and IDE plugins meet developers where they want to work. The ecosystem is evolving quickly, and vendors are increasingly focused on extending functionality beyond code generation to broader tasks within the software development lifecycle, including creating pull requests and performing code reviews.
Marketing and Creative Work
Marketing and creative teams are using AI to write and edit copy and generate creative assets, such as images or videos. As foundational model capabilities improve, some of the specialized functionality offered by software vendors early on has become commoditized and is now available in general purpose offerings. That said, some teams may still benefit from niche products with domain-specific capabilities or integrations.
Meeting Summarization and Call Transcription
AI summarization and call transcription tools are increasingly popular among sales and client-facing teams. They help reduce time spent logging notes, updating CRMs and capturing insights. Again, consider whether a general-purpose solution can handle your needs—or whether a verticalized tool offers material advantages.
Implementation Considerations
To help address security and data leakage concerns, we recommend that companies consider enterprise agreements with vendors to access enhanced data privacy and security features. Most of the major vendors will not train models on the data of customers on their enterprise plans. Tread carefully with respect to contract duration, however; avoid long-term contracts (anything over one year), as model performance is improving rapidly while costs are declining. The tools available a year from now will almost certainly surpass those available today. While vendors may offer some privacy assurances, highly sensitive or regulated data such as patient data should not be passed to general-purpose chats. For those cases, consider domain-specific vendors or model wrappers that can detect and remove private data with techniques like tokenization.
2. Automation: Unlocking New Capabilities with AI Agents
Automation has been around since long before computers but has historically relied on standardized steps and rule-based logic to guide work through defined processes. However, exceptions to standard processes are hardly exceptional and, in our experience, as business environments evolve, rigid rule systems quickly become too complex to manage. We believe AI has the potential to change this paradigm through semantic reasoning and agentic workflows.
From Rigid Rules to Semantic Understanding
Large language models (LLMs) are more adaptable than traditional rules-based systems because they can parse the meaning of input, rather than relying on exact word matches. This enables them to handle tasks like customer service triage, content categorization and dynamic routing, even if users describe issues in unfamiliar ways. For example, an LLM may be used to read a support ticket and determine the type of issue, even if the description differs from what it has seen before, or if it's in another language entirely. Relevant help content can be returned based on the user’s intent, rather than relying on exact keyword matches. As products and user behavior evolve, AI systems should be able to adapt, without the rule updates or reconfiguration associated with traditional software.
Introducing Agents and Agentic Workflows
Models are increasingly tuned to reason and interact with external tools—capabilities that enable them to function as "agents." Reasoning models can consider options, check work and generate a plan for how to execute a request. Tools, in this context, are resources or actions outside the model itself, such as web search, database queries or API calls.
"Agentic" systems combine multiple such agents in a sequence, enabling them to collaborate across steps to complete a task. For example, an agentic workflow might begin with a user request to help plan upcoming travel. The first agent might reason that the user needs to know the weather in order to decide what to pack, activities for that type of weather and any risks of transit delays. This plan is then handed off to one agent that retrieves the weather forecast for a destination, with the result passed to other agents that suggest suitable activities, recommend items to pack and look up transportation options. Unlike traditional workflows, where each step is predefined and rule-based, agentic workflows use an AI model to determine the path dynamically, adjusting the work based on the nature of the input and the execution plan created on the fly.
Build, Buy or Both?
Organizations looking to implement agentic automation have several paths:
- Build from scratch using open-source components
- Use offerings from hyperscalers like AWS, GCP, or Azure
- Choose from a growing set of vendors in the agent platform category, with tools ranging from no-code and low-code interfaces to SDKs for developers
Regardless of the approach, we believe success depends on understanding the process that you want to automate—inputs, decision points and outputs. Good outcomes depend on high quality data inputs, access to effective tools, safety guardrails (such as keeping humans in the loop) and evaluations (evals) that tell you whether the AI system has accomplished the task accurately and effectively.
Growth Timeline
3. Product Experiences: Building with the Rapidly Evolving AI Stack
For software companies, AI can deliver more than operational efficiency—it can become a competitive advantage. Companies building AI into product experiences should consider the evolving landscape of models and tools so that they can focus on delivering customer value.
Model Optionality and Cost Dynamics
Frontier model labs are in a race to develop ever more capable general models and to help drive down the cost for any given level of performance. This is good news for developers incorporating models into products. Model optionality is also increasingly important: we believe companies that can easily switch between models and providers will be able to take advantage of improved capabilities and pricing. Fortunately, this theme has gained traction. Most agent platforms now support multiple models, and a new category of developer tools is emerging to help address the tactical challenges of model orchestration and rotation.
Evaluation and Observability
Evolving capabilities, swapping between models, and importantly, the non-deterministic nature of AI model outputs creates new challenges for QA testing during product development and for ongoing monitoring. In traditional software, a given input always yields the same output. That’s not the case with LLMs. Until recently, most evaluation of an LLM's response to a prompt was based on "vibes"—whether an answer felt right to a human. However, developers and vendors are increasingly working to quantify the quality of LLM responses and add rigor to the testing process as a requirement to put AI products into production. At the same time, observability tools are evolving to help teams track model behavior, identify hallucinations and send alerts when outputs deviate from expected norms. These tools are becoming essential for deploying AI systems in production with confidence.
Designing With the User in Mind
As AI moves from novelty to production-ready, we believe "bolt-on" features will give way to experiences where AI is woven into the core product flow. Whether you're building something new or upgrading an existing product, in our experience, the best work starts with a deep understanding of customer needs and expectations.
We believe AI is capable of opening up new possibilities to deliver better, faster and more personalized solutions, but product management fundamentals still apply. Embedding AI thoughtfully means treating it as a tool in service of delivering real value, not as a feature for its own sake.
Don't delete this element! Use it to style the player! :)
.png)
AI is, increasingly, part of our daily lives—from consumer to enterprise applications. In time, it will feel less like a marvel and more like just another tool that helps us get work done.
For companies working to operationalize AI, there’s no single right answer. From our perspective, the best way to make progress is to try tools in real-world settings, empower employees to use them in their day-to-day work and be prepared to evolve. The landscape is shifting quickly, capabilities are improving, and costs are declining.
Rather than waiting for the perfect strategy or solution, we recommend leaders focus on building familiarity and momentum. Let teams experiment, learn what’s useful and stay open to what comes next. We believe the organizations that benefit most from AI will be those that are curious, adaptable and grounded in solving real problems.
Related Experience
About the Authors
Related Content
The content herein reflects the views and opinions of Summit Partners and is intended for executives and operators considering partnering with Summit Partners. The information herein has not been independently verified. In recent years, technological advances have fueled the rapid growth of artificial intelligence (“AI”), and accordingly, the use of AI is becoming increasingly prevalent in a number of sectors. Due to the rapid pace of AI innovation, the broadening scope of potential applications, and any current and forthcoming AI-related regulations, the depth and breadth of AI’s impact - including potential opportunities – remains unclear at this time. The information herein is as of July 2025.
Definitions:
“IDE” represents Integrated Development Environment, unless otherwise noted
“QA” represents Quality Assurance, unless otherwise noted
“SDK” represents Software Development Kit, unless otherwise noted
Get the Latest from Summit Partners
Subscribe to our newsletter to stay up to date on our partners, portfolio, and more.