For thirty years, business software has followed one shape: a company builds a tool, your people learn the tool, your people do the work with the tool. Software as a Service changed how that tool was hosted and billed — but the work never moved. It stayed with your team.
Service as Software is the inversion. The software does the work. You don’t licence a tool and staff it; you buy an outcome — invoices reconciled, reports produced, documents processed — and the system that delivers it happens to be an AI agent rather than a person at a desk.
The definition
Service as Software: a service performed by AI agents, delivered and managed by a provider, bought and measured by output — not by seats, licences or usage.
The unit of purchase is the giveaway. SaaS sells you access (per seat, per month). Service as Software sells you work done (the month-end pack delivered, the day’s invoices matched, the inbox triaged) — with accuracy and turnaround agreed up front, like any service contract you’d sign with a firm.
SaaS vs Service as Software
| SaaS | Service as Software | |
|---|---|---|
| Who does the work | Your team, using the tool | The agent |
| What you pay for | Seats and tiers | The service, monthly |
| What you manage | Licences, training, adoption | Nothing — you review output |
| What you measure | Usage | Output, accuracy, turnaround |
| When volume grows | More seats | Same service, agreed price |
Why this is possible now
Two things converged. First, AI models became capable of genuine knowledge work — reading a contract, matching an invoice to a purchase order, compiling a report from six systems — not just chatting about it. Second, the engineering around agents matured: supervision, audit trails, exception handling, human checkpoints. The first without the second is a demo. Together, they’re a service you can put a name and a number on.
The second part is where the “managed” matters. An unsupervised agent is an intern with admin rights. A managed agent has every action logged, every exception routed to a human, every month’s output measured against the agreed standard — and an engineering team accountable for it. That’s the difference between buying AI and buying work.
What to put an agent on first
The best first candidates share a profile: high volume, clear rules, mostly judgement-free, soul-destroying for the humans currently doing it. Invoice matching. Report assembly. Document intake. Order-status chasing. Data-quality checks.
Not the work your experts should do — the work that stops them doing it.
What it looks like month to month
A defined service. A simple subscription — ours start at £1,500 a month on annual terms. A monthly report: volumes handled, accuracy, exceptions raised, time saved. A standing review where the service gets extended, tightened or improved. Your data stays in your systems; the agent comes to it.
If that sounds less like software procurement and more like hiring a very fast, very consistent department — that’s exactly the point.