Every vendor in your stack now has an AI button. Your PSA has one. Your RMM has one. Your documentation platform has one. Your CRM has had one for two years. Each is sold as a productivity leap, each carries a per-seat uplift, and each does something genuinely useful — inside its own four walls.
That last part is the problem.
The question your stack can’t answer
Ask any MSP owner what they most want to know, and it’s rarely something a single application can tell them. It’s usually a version of this:
Which clients actually make us money?
Answering that properly means pulling contract values and billing from your PSA, time entries and effort from your service desk, device counts and alert volume from your RMM, deal and renewal context from your CRM, and probably a bit of documentation quality data to explain why one engineer takes twice as long on the same job. Five systems. Five schemas. Five different ideas of what a “client” is.
The in-app AI in your PSA cannot answer it, because the data it needs sits in four other products. It can only ever answer PSA questions — well, but narrowly. And so the actual answer arrives the way it always has: someone exports to Excel, spends a fortnight reconciling identifiers, and produces a number that’s out of date by the time it’s presented.
This isn’t a criticism of vendor AI. It’s a structural limit. An AI feature built into an application can only see the application it’s built into. No amount of model quality changes that.
What changes when the AI sits above the stack
Put the model outside the applications instead of inside them, and connect it to all of them, and the shape of what you can ask changes entirely. You stop asking application questions and start asking business questions.
CEO:
“Show me current profitability across all active contracts.” “Schedule that as a report every Monday at 9am.”
Finance:
“Which customers represent the highest support cost?” “Show me all unbilled time entries.”
Account Manager:
“Prepare the monthly ABC client report showing ticket volume, SLA performance, and identify service gaps where we could upsell.”
Service Desk:
“What is blocking SLA attainment today?” “Extend the expiry date on all active contracts at ABC client by 2 months.”
Read those again and notice what’s missing. Nobody named a system. Nobody said “in Autotask” or “in HubSpot.” The person asking doesn’t know or care which application holds which field — and shouldn’t have to. That’s the whole point. The profitability question spans contracts, time, and cost. The upsell question spans tickets, SLAs, and commercial context. The account manager’s request isn’t a query; it’s a small piece of analysis that would previously have been a half-day job and a PowerPoint.
Notice also that two of them aren’t questions at all. “Schedule that as a report.” “Extend the expiry date on all active contracts.” Reading is useful. Acting is where the time actually goes.
Five reasons the cross-system approach wins
- The valuable questions are inherently cross-silo. Ticket volume alone tells you nothing. Ticket volume against contract value tells you whether a client is worth having. Alert noise alone tells you nothing. Alert noise against engineer hours tells you which RMM policy is quietly costing you £40k a year. Every insight that changes a decision is a join across two or more systems.
- One interface, not seven. In-app AI means learning a different assistant in every product, each with its own quirks, prompt style, and limits. Your team already uses Claude, ChatGPT, Gemini, or Copilot every day. Meeting people in a tool they already know is worth more than any feature comparison.
- Access without licences. Your CFO doesn’t have a PSA licence and doesn’t want one. Neither does your sales lead. In-app AI is gated behind the app’s seat model, so the people who most need the answers are exactly the people who can’t get them. Move the interface outside the application and anyone you nominate can ask — with permissions still enforced by the underlying systems, so they see only what they’d be entitled to see anyway.
- Cost and lock-in. Per-seat AI uplift across five vendors, multiplied by headcount, adds up quietly and fast. It also deepens your dependence on each vendor’s roadmap. A stack-level layer is one commercial relationship and one integration surface, and it survives you switching RMM next year.
- Model choice stays yours. In-app AI ties you to whatever model the vendor picked and whenever they get round to upgrading it. The frontier moves every few months. An LLM-agnostic connection means you point at whichever model is best today and change your mind later without touching the plumbing.
The bit that’s actually hard
None of this is a prompting problem. It’s a data problem.
Your PSA calls it an account. Your RMM calls it an organisation. Your CRM calls it a company. Your documentation platform has its own name for the same client — spelt slightly differently, with a stray “Ltd” on the end. Time is recorded in one place, billed in another, and forecast in a third. Nothing shares a primary key.
Handing raw API access to a language model and hoping it works it out produces confident nonsense. The model will happily reconcile “Acme Ltd” and “ACME Limited” as separate customers and give you a profitability figure that is wrong in a way nobody catches for a quarter.
The work that matters is the normalisation layer underneath: resolving identities across systems, reconciling schemas, enforcing permissions, and presenting the model with clean, consistent, well-described data it can reason over reliably. Get that right and the AI part looks easy. Skip it and the AI part looks like magic right up until someone checks the numbers.
Where Recursyv AI fits
This is exactly what we built. Recursyv AI connects your ITSM platform — Autotask, ConnectWise, HaloITSM, ServiceNow, or Freshworks — directly to your preferred AI assistant through a secure, fully managed connection. Connectors for CRM, finance, and other business systems are in development.
The principles behind it:
- Use the AI you already know. Claude, ChatGPT, Gemini, Copilot. One integration serves all of them, and updates benefit every platform at once.
- Nothing to install. No agents, no infrastructure, no code. We build and manage the connection.
- Permissions enforced at every level. Users authenticate through their AI platform and see only what their existing ITSM permissions allow. All access is logged.
- Azure hosted, ISO certified. Multi-region, high availability, fully managed.
Read and write, both. Ask what’s blocking SLA today, then reassign the queue. Find the unbilled time, then approve it. As one MSP owner put it: identifying their unbilled WIP paid for the whole thing immediately.
Decouple the AI from the app
In-app AI isn’t bad. It’s just bounded. It will keep getting better at the questions its own application can answer, and those are real questions worth answering.
But the questions that decide whether your business is healthy — which clients to keep, where margin is leaking, which contract to renegotiate, which service gap is an upsell — live in the gaps between your systems. No amount of in-app AI reaches across those gaps, because it was never designed to.
The MSPs pulling ahead aren’t the ones who turned on the most AI features. They’re the ones who put a single, well-connected assistant above the stack and started asking better questions.
Ready to see it? Request a demonstration or get in touch: info@recursyv.com · +44 118 380 0142 (UK) · +1 833 749 3781 (US)
