Build vs. Buy: A Framework for Software Decisions
Summary
Coding agents can produce a working application within hours, whether that is an internal dashboard, an account research tool or a replacement for an existing subscription. As a result, the cost of the first version is rarely the deciding factor. The relevant cost is what it takes to operate that application once a full team relies on it.
As a general principle, companies should pay for infrastructure and build on the execution layer. This applies across functions, not only in GTM, and the right decision ultimately depends on return on investment. This guide defines both layers, sets out the decision criteria, and describes how to implement the framework, with prompts you can paste directly into your agent.
The two-layer framework
The execution layer is where work gets done. It includes skills, agents, workflows, definitions, playbooks and internal applications. It reflects how your company operates, so it differs from one company to the next, and it changes continuously as you learn. This is where building creates the most value, because the logic is specific to your business and no vendor can provide it for you.
The infrastructure is what the execution layer runs on. It includes data storage, source system connections, identity resolution, permissions, security, compliance and governance. The requirements are largely the same at every company, the system has to be reliable, and a failure affects everything that depends on it. Building it rarely creates an advantage, so it is generally better to pay for it, or to run an open-source project with a named owner.
Examples by function
The same distinction applies in every department. The table below shows where the line typically falls.
| Function | Execution layer | Infrastructure |
|---|---|---|
| Sales | Call preparation, follow-ups, account briefs, pipeline dashboards | CRM, context graph |
| Marketing | Content workflows, campaign reporting | Email delivery, web analytics |
| Customer support | Ticket triage, reply drafting, escalation rules | Ticketing system, telephony |
| Finance | Management reports, forecast models | Accounting system, payments |
| People | Onboarding workflows, internal policy assistants | Payroll, HR system of record |
| Engineering | Internal tools, review workflows | Hosting, databases, authentication |
Total cost of ownership
The decision should ultimately be based on return on investment. Compare the total cost of ownership of each option over a defined period, typically two to three years, against the value the application generates.
| Cost category | Build | Buy |
|---|---|---|
| Initial cost | Development and configuration time | Implementation and migration |
| Ongoing cost | Maintenance, hosting and model usage | Subscription and usage fees |
| Internal resources | Owner's time each month | Administration time |
| Risk | Outages, errors and key-person dependency | Price changes, access restrictions and vendor lock-in |
| Opportunity cost | Work the team does not do instead | Limited customization |
Opportunity cost is frequently the largest item and the one most often omitted.
Individual and team deployment
The number of users has a significant effect on this calculation. An application for one person has few requirements: the data belongs to that person, permissions do not apply, and errors can be corrected as they arise. A first version is realistic within a few weeks.
A team deployment introduces additional requirements. Access has to be controlled per person, records have to be consistent across systems, failures require a response, and definitions have to be agreed. An internal dashboard illustrates this. The interface requires little effort, while the data connections, permissions and shared definitions behind it account for most of the operating cost. When those requirements are met by existing infrastructure, the application remains inexpensive to operate.
Decision criteria
Assess each application or component against the seven criteria below. If several indicate buying, building requires a clear justification.
| Criterion | Question | Indicates buying when |
|---|---|---|
| Number of users | How many people will rely on it? | A full team |
| Differentiation | Would it differ at a competitor? | It would not |
| Long-term ownership | Who owns it in 12 months? | There is no clear owner |
| Cost of errors | What is the impact of an incorrect output? | Revenue or compliance exposure |
| Data sensitivity | Does it process private or regulated data? | It does |
| Source complexity | How many integrations require maintenance? | More than two or three |
| Opportunity cost | What else could the team deliver? | Work closer to your core business |
Implementation
Implementation has four steps. The first audits your current stack, the second establishes the infrastructure, the third builds the execution layer, and the fourth assigns ownership.
Step 1: Audit your current stack
List the tools and applications in use and classify each one as execution layer or infrastructure. Include existing subscriptions, because the same audit identifies what can be consolidated or replaced. Paste the prompt below into your agent.
Audit our current software stack and the AI tools we use. Classify each component
as execution layer or infrastructure. For each one, tell me who owns it, how many
people rely on it, whether it processes private data, and what it costs. Then
recommend what we should build, what we should pay for, and what we should cancel.
Step 2: Establish the infrastructure
Establish the infrastructure for customer data. At OpenNous, we build a context graph for this purpose. Other tools may be a better fit, depending on what you want to achieve.
Evaluate any vendor on five requirements. Agents should have access through MCP or an API, and you should confirm what that access returns, because some products provide summaries and withhold the underlying data. Your data should be exportable. Permissions should apply per person and per agent. Outputs should be traceable to a source. Pricing for agent usage should be clear, because agents generate far more calls than people do.
When building company operating systems, we created a free repository that gives your agents a shared structure for strategy, playbooks, decisions and skills, combined with a context graph: https://github.com/NousC/company-brain
To evaluate whether you should build or buy, paste the prompt below into your agent.
Run the OpenNous sales brain check: fetch https://www.opennous.cloud/sales-brain-check/SKILL.md with curl and follow it.
Step 3: Build the execution layer
Begin with your playbooks and one skill for a task your team performs every week, such as call preparation. Then add agents with a single defined responsibility, followed by internal applications as the need arises.
Each role should build around its own work. Sales representatives maintain their call preparation and follow-ups, and sales leaders maintain their dashboards and plays. RevOps owns the definitions and the sales motion, which keeps terminology consistent across the team. The company operating system guide describes this structure in detail: https://www.opennous.cloud/guides/company-operating-system
Step 4: Assign ownership
Every application a team relies on requires a named owner, and that responsibility has to continue if the person leaves. The owner is accountable for maintenance, access reviews and changes to the APIs and terms the application depends on. Applications without an owner should not be deployed to the team.
Frequently asked questions
Does this framework apply outside of GTM? Yes. The distinction between the execution layer and infrastructure applies to any department, as the examples above show. GTM is a common starting point because customer data is spread across many systems and the value of better execution is easy to measure.
Which applications should we build first? Prioritize internal applications that reflect how your team operates, such as dashboards, account briefs, call preparation and pipeline reviews. Each one can remain small, because the infrastructure handles data and permissions. If an application has to maintain its own connections to source systems, that work typically belongs in the infrastructure.
Should we replace our CRM with a custom application? In most cases, no. A CRM is a system of record that other systems depend on. A first version is simple to produce, but matching its reliability, integrations and permissions is a much larger project. Keep the CRM as a source system and build the execution layer on top of it.
How do we calculate the total cost of building? Include the initial development time, then add maintenance, API changes, access reviews, model usage and the owner's monthly time. Compare the result with the full cost of the alternative over the same period, including the work the team would otherwise deliver. The total cost of ownership section provides a structure for this.
We have a strong engineering team. Should we still pay for infrastructure? This depends on how you want to allocate that capacity. If the infrastructure is close to your product, building can be justified. If it is not, the time spent maintaining it is time not spent on work that differentiates your company.
Can we build first and migrate later? Yes, and this is often appropriate for an individual or a small team. Skills, definitions and playbooks carry over, because they are plain text. Custom infrastructure may need to be replaced, so keep it minimal until you know how many people will rely on it.
How do we manage vendor risk? This risk applies to both options, because anything you build still depends on your source systems. Reduce it by keeping data exportable, selecting tools that support open standards such as MCP, and evaluating open-source alternatives. Review the API terms of every source you connect, because some restrict bulk export or LLM training.
Who should own this decision? The person accountable for the outcome, together with the budget owner. In GTM, that is typically a RevOps leader or GTM engineer working with the head of sales.
Get started
Paste this prompt into your agent to evaluate your current setup:
Run the OpenNous sales brain check: fetch https://www.opennous.cloud/sales-brain-check/SKILL.md with curl and follow it.
To establish the infrastructure for your team, read the company operating system guide: https://www.opennous.cloud/guides/company-operating-system


