The CRM-as-a-service category was built for a different world. A world where integrating two business systems took $50,000 and six months. A world where AI was a feature you bought, not a layer you composed. A world where your data lived with the vendor because the vendor was the only one who could host it reliably.
None of those things are true anymore. And the moment all three stopped being true, the math underneath the $300-per-seat-per-month CRM started to break.
This is not a hot take. We have watched five mid-sized businesses in Southern California in the last 12 months walk away from Salesforce, HubSpot, or both, in favor of a custom dashboard built around their actual workflow. One of them was 85 percent migrated to Salesforce when they decided to stop. They are now running on a Heed-built platform. Their detailed story is on the structural engineering case study page.
What Actually Changed
Four things. Each of them, on its own, would not have been enough. Together, they collapsed the moat.
Cloudflare Workers and Hyperdrive made custom backends ten times cheaper. Five years ago, building a custom multi-tenant business app meant standing up infrastructure: load balancers, database clusters, queue workers, identity providers, monitoring. A serious engineering team and six figures of monthly cloud spend. Today, a single developer can stand up the same infrastructure on Cloudflare in days, at single-digit dollars per month per workload, with global edge presence and zero ops overhead. The unit economics flipped.
The Anthropic and OpenAI APIs gave any team enterprise-grade AI for under $1,000 a month. The CRM vendors built their AI features as paid add-ons priced as if AI were still scarce. It is not. A custom dashboard can call Claude or GPT directly for a fraction of the per-seat AI uplift the major CRMs charge, and the calls can be tuned to the specific workflow rather than generic suggestions.
AI-fluent developers ship in days what used to take quarters. A developer working with Claude Code, MCP servers, and modern frameworks ships features in a fraction of the time they took two years ago. The bottleneck is no longer how fast a person can write code, it is how fast a person can describe what should be built. That changes who can afford custom software. It is not just F500 anymore.
ISO 42001 and the EU AI Act made data ownership a legal advantage, not a technical liability. When the regulator asks where your customer data lives, "Salesforce" is a fine answer until the day it is not. Owning your data plane in your own Cloudflare account, with your own audit trail, with role-based access you control, is now a board-level governance posture. It used to be a technical hassle.
The bundle that the CRM-as-a-service category sold, integrations plus AI plus hosting plus security, has been unbundled by the cloud, the model providers, and the regulators. What is left is a UI you could have built and a per-seat price you would not have agreed to.
What CRMs Are Still Good At
We are not categorically anti-CRM. Salesforce did not get to where it is by being bad at its job. The major platforms are still good at three specific things, and if those three things describe your situation, do not move.
One, they are good at supporting very large sales orgs with mature, formal sales processes. If you have 200 reps with three layers of management, opportunity stages defined down to the second decimal, and a forecasting cadence the board signs off on, the CRM is doing real work. Replacing it with custom is a multi-year project most companies would not survive.
Two, they are good at ecosystem fit. If your business runs on a marketplace of certified integrators, third-party apps, and consulting partners, you are buying that ecosystem more than the software itself. That ecosystem does not exist around custom builds.
Three, they are good at being the de facto answer when leadership has no opinion. The CFO who has signed off on Salesforce in three previous companies is not going to greenlight a custom build at the fourth. That is a perfectly rational risk posture.
What Custom Wins
Almost everything else. Specifically, almost any business under 100 seats with a workflow that does not fit the standard schema. Which is most businesses.
Real estate firms with deal pipelines that need to track entitlements and inspection windows. Law firms with matter intakes that branch on practice area. Field service companies that need photo evidence and GPS metadata on every job. Engineering firms that need to relate projects to permits to inspections to invoices, none of which the CRM understands. Healthcare practices with HIPAA requirements that turn every standard CRM workflow into a compliance question.
For all of those, the standard CRM schema is a tax. You either pay developers to bend the schema (which costs more than the seats), pay a consultant to live inside your instance (which costs more than the seats), or accept that 30 percent of your team's time is spent forcing the workflow into a box that does not fit it.
A custom dashboard is built around the workflow. The workflow is not bent to fit the dashboard. That is the whole pitch.
The Math
Run the numbers on a typical 30-seat SMB stack. The list looks something like this in 2026: Salesforce or HubSpot Professional at $90 to $180 per seat per month, plus an email marketing tool at $300 to $600 per month, plus a forms or landing page builder at $150 per month, plus an e-signature tool at $50 per seat per month, plus an AI add-on at $30 per seat per month. Realistically you are paying somewhere between $1,800 and $4,200 per month, plus the implementation partner you hired to keep the schema honest.
Heed's Operations Platform Growth tier is $649 per month. Custom dashboard, agentic workflows on Anthropic and OpenAI, role-based access, your data in your Cloudflare account, quarterly upgrades included. The build cost is amortized into the platform fee. Most clients hit payback in months, not years.
That is not a fair fight. The CRM stack has to defend a price that was set when none of the underlying costs had collapsed yet. The custom dashboard has to defend a price that reflects the new cost structure. Customers who run the numbers tend to move.
What the Migration Actually Looks Like
This is the part that scares people, and the part the CRM vendors lean on. "You will be in implementation hell for six months." Sometimes true. Often not, especially for SMBs who never used most of what they were paying for.
A typical migration we run takes six to ten weeks. Week one and two are workflow mapping and Proof of Concept on the highest-friction process. Week three and four are connectors: the live read or read-write into M365, QuickBooks, the existing email tool, the existing e-signature tool, whichever of those the team will keep. Week five and six are the role-based UI and the agentic workflows. Week seven through ten are pilot, training, and cut-over.
The data migration is usually the easy part. CRMs have export APIs. The hard part is deciding which of the 47 custom fields, three sales processes, and 12 workflow rules anyone has actually been using. Most teams are surprised by how few survive the audit.
Where to Start
If you are on a CRM today and the per-seat math is bothering you, the first move is not to rip and replace. The first move is to map the five workflows your team uses every day and see how many of them the CRM actually supports cleanly versus how many require workarounds. If the workaround count is high, custom is on the table. If it is low, your CRM is doing its job.
If you are evaluating a CRM purchase right now and the seat count is under 50, run a parallel evaluation against a custom dashboard. Even if you choose the CRM, the comparison will sharpen your thinking about which features you actually need. If you choose custom, you will know exactly why.
The CRM-as-a-service category is not going away tomorrow. It is going to slowly compress into the niches where it still wins. Everywhere else, the dashboard your team actually wants is now affordable to build.