The Question is Back on the Table
Four years ago, the build-versus-buy debate for customer experience tooling was mostly settled. Pay Gainsight, Planhat, or Totango. Wire in Salesforce or HubSpot. Train your CSMs. Move on. The buy path won, not because it was fashionable, but because the real cost of building something as layered as a CS platform — data model, automation engine, SLA-backed uptime, and a community of operators who already knew the tool — was roughly a decade of engineering you didn’t have.
In the last six months, every other founder I speak with has quietly re-opened that decision. Agentic AI has changed two things at once: the cost of building has dropped sharply, and the incremental advantage of buying has narrowed. A ten-person SaaS company with one sharp engineer and a clean customer data model can now stand up an internal “customer agent” that drafts QBR notes, forecasts churn, writes renewal emails, and summarizes support threads — the same outcomes the $180k-a-year platform was sold to deliver.
That does not mean you should build. It means anyone still making this decision on 2022 assumptions is making a 2022 decision.
What You're Actually Buying (and What You're Not)
When you buy Gainsight, Salesforce Service Cloud, HubSpot, or Planhat, you are not primarily paying for features. Features commoditize every quarter. You are paying for four durable things:
1. A data model that someone else has already validated on hundreds of other SaaS companies.
2. A library of best-practice workflows — health scores, onboarding plays, risk alerts — that took a decade of customer research to encode.
3. A community of operators on LinkedIn, Reddit, and in your own hiring pool who already know the tool.
4. A vendor with a roadmap, an SLA, and someone to escalate to at 2 a.m.
For most B2B SaaS companies between $5M and $100M ARR, those four things beat anything you would build in year one. The founders who tell me “we can stand this up in a weekend” are invariably pricing the MVP, not the full lifecycle.
But agentic AI is slowly unbundling each of those four. The data model can be scaffolded. The playbooks live inside every model’s training set. The community is one Slack channel away. The roadmap — the real one, the one about what matters to your motion — is now something a well-briefed internal team can actually move on.
What "Build" Actually Costs
Most build pitches do not price year two. Year two is when sales asks for territory assignment logic, the CSMs want Slack integration, finance needs a GAAP-friendly revenue retention report,
the first enterprise security review arrives, and the product owner you assigned has rotated to a more urgent project. A buy decision includes ten years of that maintenance you do not have to fund. A build decision silently creates a new permanent department inside your company.
Agentic AI does not fix this. If anything, agents make the maintenance tax worse when your data is messy. An agent with clean inputs is a force multiplier. An agent with inputs spread across six tools, three spreadsheets, and the heads of four CSMs is a confidently wrong junior employee you cannot fire.
This is the part of the conversation I find most founders under-weight. “We’ll just plug an agent in” assumes a unified customer object. If you do not have one, the honest first step is not building or buying — it is cleaning. Which, ironically, is exactly the kind of structured data work the major CX platforms have already done for you?
The Hybrid Middle (Where Most Winners Land)
The most interesting pattern I am watching right now is this: buy the durable system of record, build the agentic layer on top.
Use Gainsight, Salesforce, or Planhat as the place your customer data lives. Let the vendor carry the compliance burden, the SSO complexity, the audit trail, the tedious permissioning work. Those things are table stakes for anyone selling into the enterprise, and they get harder, not easier, to backfill once you have built around them.
Then build — or buy narrow — the agentic layer that reads from and writes to that system of record. Renewal-drafting agents. Health-score explainers. QBR preparers. Sentiment analyzers.
These are relatively cheap to build now, differentiating when tuned to your specific motion, and disposable when something better comes along in six months.
The winner in this cycle is not the company that builds everything. Nor is it the one that buys everything. It is the company that knows which parts should be durable (and therefore bought) and which parts should be disposable (and therefore built). Get that split right and you will outpace competitors on both sides of the debate.
A Five-Question Decision Framework
When a founder asks me build or buy, I walk them through five questions. The answers cluster tightly.
5. Do you have 50 or more customers with consistent signal? If no — buy. You do not have the data density to justify the build.
6. Is customer retention a strategic moat or a cost center for your business? If a moat, buy the platform and build the agents on top. If a cost center, buy and stop.
7. Do you have one operator who owns CS ops full-time? If no — buy. A tool without an owner becomes debt regardless of origin.
8. Is your customer data sitting in fewer than three systems, or more? If more — buy, and use the implementation project to consolidate.
9. Will you regret not being able to change this in eighteen months? If yes, build the agentic layer on top of a bought platform. If no, buy the whole thing and get back to your real work.
I have yet to meet a founder who ran those five questions honestly and still ended up building a full CX platform from scratch. Most of them end up in the hybrid middle — and those are the ones I see moving their retention numbers the fastest.
The Retention and NPS Lens
Let me close where this conversation probably should have started: retention and NPS. Neither improves because of a tool choice. They improve because of a system, and the system has three parts — clean data, a clear motion, and an operator who runs it.
A build decision usually fails on part one: data cleanliness takes at least twice as long as anyone budgets for. A buy decision usually fails on part two: teams inherit the vendor’s motion instead of adapting it to their own. Agentic AI amplifies whichever of the three parts you actually invested in, and it amplifies the gaps just as loudly. An AI layer on top of a broken motion simply executes the broken motion faster.
The founders I have seen lift NPS fifteen points in a year and move NRR into the 120s did the same thing consistently: they treated the CX tool as perhaps twenty percent of the decision, and the motion, the data, and the ownership as the other eighty. Build-versus-buy was downstream of that.
The Real Question, Two Years Out
Build versus buy is not the interesting question anymore. The interesting question is: what will the agentic layer on top of this platform look like in twenty-four months, and are you investing in a stack that can carry that layer cleanly?
If you are in the middle of this decision right now, my practical advice is this. Buy the platform that gives you the cleanest data model for your specific motion. Use the engineering hours you saved to build the two or three agents that move the retention numbers you actually get paid on. Revisit the split every twelve months. That is what I would do — and it is what I am doing, alongside the founders I advise.
