Build vs Buy: Should You Buy a Customer-Service AI Agent or Build Your Own?

The customer-service AI market just sent a loud signal. In June 2026, Salesforce reportedly acquired Fin, the AI support agent that grew out of Intercom, for about $3.6 billion, folding a category leader into its Agentforce platform. When a vendor pays billions to absorb a support-AI product, two things are confirmed at once: AI support agents are now a serious, mainstream category, and that category is consolidating into a handful of big platforms. For any business weighing how to automate support, that reframes the central question, not "should we use an AI agent?" but "should we buy one or build our own?"

What the consolidation actually means for buyers

Consolidation cuts both ways. If you already live inside a platform's ecosystem, a bundled support agent can be a genuine win, tighter integration, one vendor, less glue code. If you do not, the same bundling is how lock-in starts: your support data, your knowledge base, and your workflows get tied to a suite whose pricing and roadmap you do not control. The Salesforce/Fin deal is part of the broader shift we describe in who owns the agent layer, where major platforms race to own the agent inside their own walls.

The practical takeaway is not to avoid platforms, they are often the right answer, but to go in with eyes open about portability and cost, and to make the build-vs-buy call deliberately rather than by default.

When to buy

Buying an off-the-shelf agent is the right starting point for the large majority of businesses. Mature products already handle the common support load, deflecting frequent questions, looking up order or account status, triaging tickets, and handing off cleanly to a human, and they do it without you running any infrastructure. You get to value in weeks, not quarters, and the vendor absorbs the ongoing work of keeping the model current.

The classic mistake is the opposite: a team builds a generic support bot from scratch, spends months on it, and ends up with something a commercial product does better and cheaper. If your support needs look like everyone else's, that is a strong sign to buy. For a current view of the Canadian options, see our guides to the best AI receptionists for Canadian business and AI automation playbook.

When to build

Building, or commissioning a custom build, earns its cost in specific situations: when the agent must do something genuinely particular to your business that no product supports, when it needs to integrate deeply with proprietary or legacy systems, when data sensitivity or Canadian data-residency rules rule out the available SaaS tools, or when avoiding vendor lock-in is itself a strategic priority. In those cases the control and fit justify the engineering and the ongoing ownership.

Lean buy when…	Lean build when…
Your support looks like everyone else's	Workflows are specific to your business
You want value in weeks	Deep proprietary/legacy integration is required
You have no team to run infrastructure	Data residency / privacy rules out SaaS
A platform you use already offers it	Avoiding lock-in is a strategic priority

The hybrid most businesses actually need

In practice the answer is rarely pure. The pattern that works for most mid-sized businesses is to buy the platform for the common cases and build (or customize) for the parts that are genuinely yours. Let an off-the-shelf agent deflect the repetitive 80%, and add custom logic, integrations, or a separate model only where your business is different. The architectural key is to keep your knowledge base and conversation data portable, and to route AI calls through your own interface where you can, so you are never fully captive to one vendor's pricing or roadmap, the same operating-layer discipline we cover in the shift to the operating layer.

Watch the economics, not just the demo

Platform agents are quick to start and can get expensive at scale, per-resolution or per-seat pricing, setup services, and the cost of leaving later. Before you commit, model total cost of ownership over two years and put a number on switching cost, the same discipline that keeps any AI business case honest, as we argue in why most AI ROI models are wrong. Buy for speed, but buy portable.

The bottom line

Salesforce paying billions for Fin tells you the customer-service AI category has arrived and is consolidating fast. For your business, that is good news, the tools are mature, so you rarely need to build from scratch. Buy the platform that fits, pilot it narrowly, keep your data portable, and reserve custom build for the parts that are truly yours. Decide build-vs-buy on purpose, and you get the speed of buying without the lock-in that consolidation is quietly selling.

Frequently Asked Questions

What does Salesforce buying Fin mean for customer-service AI?

In June 2026, Salesforce reportedly acquired Fin (the AI support agent formerly associated with Intercom) for about $3.6 billion, folding a category leader into its Agentforce platform. The signal for buyers is consolidation: customer-service AI is increasingly bundled into large suites rather than bought as a standalone tool. That can mean tighter integration if you are already on the platform, and more lock-in if you are not. It also confirms that AI support agents are now a mainstream, enterprise-grade category, not an experiment.

Should I buy an off-the-shelf AI support agent or build my own?

For most businesses, buy first. Off-the-shelf agents from established vendors handle the common 80% (FAQ deflection, order status, triage, handoff) quickly and reliably, with no infrastructure to run. Build, or commission a custom build, only when your workflows are unusual, your data integrations are deep, you need to avoid platform lock-in, or the off-the-shelf options cannot meet your privacy and data-residency requirements. The most common mistake is building a generic support bot from scratch that a mature product already does better and cheaper.

When does building a custom AI agent actually make sense?

When the agent needs to do something specific to your business that no product supports, when it must integrate deeply with proprietary or legacy systems, when data sensitivity or Canadian data-residency rules rule out the available SaaS options, or when you want to avoid being locked into a single vendor whose pricing and roadmap you do not control. A hybrid is common: buy the platform for the common cases and build custom components for the parts that are genuinely yours.

What are the hidden costs of buying a platform agent?

Per-resolution or per-seat pricing that scales with your volume, professional-services fees for setup, the cost of keeping the knowledge base current, and lock-in that makes leaving expensive later. Platform agents are fast to start but their economics can bite at scale, which is exactly why you should model total cost of ownership over a couple of years and keep your data and content portable, rather than judging on the sticker price alone.

How should a Canadian business get started without overcommitting?

Start with a narrow, measurable pilot on one channel, deflecting your top 20 repetitive questions, with a clean handoff to a human. Pick an off-the-shelf agent that meets your PIPEDA and data-residency needs, measure resolution rate and escalation-error rate against a human baseline, and only then decide whether to expand, switch, or build custom around it. That keeps the first decision cheap and reversible.