You keep hearing that AI is coming for enterprise SaaS. How much of that narrative is actually true? Recently, The Information’s Laura Bratton wrote an article titled “How Small Firms Use Claude to Quit Salesforce”. In it, a property management firm in Atlanta with 55 employees cut roughly $100,000 a year by swapping Salesforce for a custom app it built with AI tools. A pro rugby club replaced its CRM and ticketing system with four engineers and four months of work. A 24-person medical software firm walked away from a $40,000 Salesforce contract for a homegrown system that costs about $500 a year to run. This approach has a name: agentic coding. You describe the software you want in plain English, and AI writes it for you.
What you don’t hear as often is what happens when the firm building that software provides trustee services, and the data inside it belongs to beneficiaries.
That distinction matters, because trust companies are feeling the exact pain that’s driving everyone else to build. Most fiduciary practices run on two or three systems that were never designed to talk to each other: a CRM, a trust accounting system, a data aggregator, and a compliance application, for example. The connective tissue between them is usually a person. Someone exports from one system, reformats in a spreadsheet, and keys into another. Every handoff is a chance for an error, and in this business an error isn’t a typo. It shows up in litigation documents with your name on it.
So when a trust organization sees a 45-person business replace its CRM for $1,200 a year, the question is obvious. Why can’t we do that?
You could. The question is whether you should.
The case for building your own
The appeal is real, and it’s worth being honest about it.
You only pay for what you use. The most common complaint from companies leaving Salesforce and HubSpot isn’t that the software is bad. It’s that they were paying for an enterprise platform and using a sliver of it, or that it wasn’t applicable to their specific domain. Small firms that made the switch report cutting software costs 40 to 80 percent.
The software fits your workflow instead of the reverse. Generic CRMs don’t come out of the box with concepts of a grantor, a remainder beneficiary, or a principal and income allocation. Every trust company that runs on Salesforce has paid, in configuration time, consulting fees, or in-house FTEs, to teach a monolithic tool how trusts work.
You control the roadmap. No waiting on a vendor’s release cycle. If you need a field, a report, or an automation, you build it this week.
For a company selling tickets to fans, that math may genuinely work. But a fiduciary isn’t selling a product and the risk side of this ledger looks very different when you hold legal duties to the people whose data is in the system.
The case against, if you’re a fiduciary
Who audits the auditor? AI-generated internal software works until it doesn’t, and the failure modes are quiet. A rounding behavior nobody specified. A permissions gap nobody tested. A backup routine that was never actually running. In most businesses those are embarrassing. In a fiduciary practice, they’re potential breaches of duty. DIY works until a fire needs to be put out. DIY software works until the examiner arrives. You don’t get to pick which day they happen.
AI access to client data is not a casual decision. Building your own AI tooling means deciding, on your own, which models see client data, where that data is processed, whether it’s retained, and whether it trains someone else’s system. Those are questions security teams at large vendors spend careers on. A small or mid-sized fiduciary firm building an agent this month may not have asked them all.
Maintenance is forever. The build takes four months. The ownership lasts as long as your firm does. And if the person who understands your homegrown system leaves, your entire operating platform becomes an unreadable black box overnight.
Your regulator doesn’t grade on effort. Court accountings, exams, and audits demand defensibility: a clean audit trail, separation of duties, records that show who did what and when. Purpose-built systems earn that trail over years of scrutiny. A homegrown app has to prove it from scratch, and you’re the one proving it.
The option to modernize operating models
The reason firms want to build is not that they love writing software. It is that their current stack is fragmented, and AI finally makes unification feel possible. That is the real lesson of the agentic coding era.
Firms are tired of paying for systems that do part of the job, then relying on people, spreadsheets, exports, and manual checks to connect the rest.
But for fiduciaries, the answer is not simply to build a custom AI tool and hope it holds up under real scrutiny. The better question is whether the operating model itself needs to change.
A purpose-built fiduciary platform should close the gaps that DIY projects are trying to solve. Administration, accounting, compliance, records, reporting, and AI should not all live in separate places with a person acting as the connective tissue.
They should live in one system, with the permissioning, audit trail, security, and human review that fiduciary work requires.
That is the model we have built ProTrustee around.
Not AI layered casually on top of client data. Not a generic CRM forced to understand trusts. Not three disconnected systems held together by exports.
A single fiduciary operating platform where AI is designed around the obligations trustees already carry.
What this means for standalone CRMs
The second question worth asking is what happens to Salesforce, HubSpot, and ServiceNow as standalone vendors, because the technology shift here has a name: the Model Context Protocol, or MCP.
MCP is an open standard that lets AI agents connect directly to data sources and tools. In practical terms, it means an AI agent no longer needs a CRM’s interface to work with customer data. One AI startup in The Information’s article stopped using HubSpot not over cost, but because it wanted its AI agents working directly against its own database, without a CRM app in the middle.
That’s the real threat to standalone platforms. Not that companies will stop needing customer data, but that the interface layer, the screens and dashboards that justified per-seat pricing, stops being where the work happens. When agents can read and write your data directly, a system whose main value is being the place you look at data has a problem.
Mission-critical systems of record with domain expertise natively built into the operating layers are in a different position. The lesson of the emerging MCP era isn’t that all software gets replaced, it’s that generic software does and the mission-critical applications are what AI interacts with.
The questions to ask before you decide
If your firm is weighing any version of this, these are the questions that matter:
- Where does client data go when your AI tool processes it, and is it used to train anyone else’s model?
- Who maintains the system in year three, and what happens if they leave?
- Can the system produce an audit trail a court or regulator will accept?
- Is a human reviewing what the AI does before it becomes part of the record?
- And the biggest one: is building software the best use of your time, or is administering trusts?
The firms leaving Salesforce aren’t wrong that the old model is broken. Paying enterprise prices for a fraction of the functionality, then paying again in headcount to bridge systems that don’t talk, is a bad deal. But the fix for a fiduciary isn’t becoming a software company. It’s demanding software that was built for fiduciaries in the first place.
Reference: How Small Firms Use Claude to Quit Salesforce — The Information