Designed, built, and run for firms where the work has to stand up. Onboarding, AML, reconciliation, reporting. You only pay when the agent delivers. Failed outputs aren't billed.
The people you hired for judgement, client work, and growth are spending their afternoons on data entry, document chasing, and the same engagement letters they wrote last year. The work gets done, but it gets done by the wrong people.
1.2 hours a day, lost per accountant. Silverfin's All Accounted For 2025 survey of 350 partners and staff at UK firms turning over £1m to £1bn found that accountants lose an average of 1.2 hours every working day to low-value tasks. Manual data entry, chasing colleagues or clients, repetitive compliance work. Across the sector, that is an estimated £5.3 billion in foregone revenue every year. [1]
1.5 hours of admin for every hour of advisory. The same study found that for every hour a partner spends on client advisory, they spend 1.5 hours on internal admin. Only 14% of partner time goes to strategic planning, while 51% of firms list growth as their primary goal. The arithmetic doesn't work. [1]
The work has to defend itself. Accountancy work has to stand up to ICAEW, ACCA, FRC, HMRC, and increasingly to AI-specific scrutiny. The July 2025 update to the ICAEW Code of Ethics introduced new obligations on members for work produced with AI assistance. Most general-purpose AI tools were never built to leave the audit trail your work needs.
Clients are bringing in their own mess. Public AI tools have arrived in your inbox. Clients turning up with ChatGPT-generated tax advice and incorrect figures is now a recurring cost. Practitioners are spending hours each week unwinding errors that confident-sounding chatbots produced on the other side. The answer isn't no AI. It's AI you can stand behind.
Trigger-based agents that work through a defined sequence of logical steps and produce a structured output, every time. Each agent handles one process end-to-end. You set the rules and the standards. The agent executes against them, and we run it.
The workflow areas below are illustrative. The exact steps and outputs are designed around your firm's specific process at Discovery.
One way to build this: new-client intake from first contact through to a signed-off onboarding pack.
A completed onboarding pack ready for partner review.
How we might approach this: engagement letter generation and refresh from a single brief.
A signed-ready letter on your letterhead, in your template, with the right clauses for the engagement type.
Could be scoped as an agent that reads inbound invoices as they arrive.
A coded transaction ready for posting, with an audit trail of every decision the agent made and why.
One possible version is an agent that matches transactions across bank feeds, ledgers, and supporting documents.
A reconciliation pack with exceptions ranked by materiality, ready for a reviewer to act on rather than investigate.
This could look like an agent that produces a management accounts pack in your house format.
A draft pack with variance narrative, ready for partner review and client release.
We might build this as an agent that takes year-end preparation from source documents to a draft return for senior review.
A working file your team can sign off on, not start from.
These are the workflow areas most likely to come up in scoping for an accountancy firm. The shape gets written from your process.
We don't hand you a tool and walk away. We design the agent around your process, build it with senior engineers, and run it on managed infrastructure.
A free 30-minute Discovery session, then a paid scoping engagement if the fit is right. We map your process, identify the trigger and the output, and produce a specification. Many firms come in with a process to automate. Some come in knowing only that they want to do something with AI. Both are fine.
Custom development by senior engineers, over four to eight weeks. No juniors learning on your project. No offshore handoff. We engineer the prompts, define the evaluation criteria, and test against real cases from your firm. Even a 1% change in prompt accuracy compounds at document scale. Getting it right before go-live is the job.
Once live, we monitor outputs, retrain as your process evolves, and maintain quality. Most AI fails when tested seriously. We do that testing for you on isolated Azure infrastructure dedicated to your firm.
Most AI vendors charge for a build and walk away. We get paid when the agent produces an output to your firm's agreed standard. Failed outputs aren't billed. The commercial structure is the differentiator.
A single refundable commitment per agent, agreed at scoping. It goes towards the work, not into a deposit account. If we don't deliver to the spec we agreed, you get it back. That is the skin in the game.
Once the agent is live, you pay per workflow completed. Unit price scales with workflow complexity. No seats. No infrastructure markups. No usage minimums dressed up as a subscription.
If an output doesn't meet the standards we agreed at Discovery, you don't pay for it. We fix it, we improve the agent, and we absorb the cost. The output is what you bought.
Specific figures live on /pricing. Pricing scopes to your processes, not a rate card. The Discovery session is free, with no commercial commitment to proceed.
A hypothetical, framed for clarity rather than as a case study. The shape is drawn from how mid-tier accountancy firms typically structure intake.
A 22-partner firm onboarding 600 new clients a year
Six steps before the case handler can begin: ID and proof of address, sanctions and PEP screening, risk rating, engagement letter, conflicts check, and practice management setup
90 to 180 minutes of partner or senior time per client, before any chargeable work starts
Triggered by the new client form submission
ID data pulled through your verification provider, screened against your AML watchlists, rated against your risk matrix
Engagement letter drafted from your standard templates, conflicts checked across the firm's client list, case created in your practice management system
A complete onboarding pack on the case handler's desk, ready for sign-off, with every decision logged
Before, your senior staff spend roughly three working weeks a month on intake admin. After, they spend that time on advisory work, planning, and the conversations only a partner can have. Sign-off ready the same day, with a full audit trail. The agent isn't replacing the judgement. It's removing everything around it.
This is an illustration of the model we'd scope for a typical mid-tier firm. It isn't a named client engagement.
AICPA and CIMA's December 2025 Future-Ready Finance survey of 1,446 senior finance leaders found that 88% believe AI will be the most transformative technology of the next two years, while only 8% feel their organisation is very well prepared to manage it. That gap is what governance-led firms are quietly worried about. It's the gap we build for. [2]
Accountancy AI doesn't get deployed in a vacuum. It sits inside a clear governance frame. The ICAEW Code of Ethics has been updated for technology use. The FRC has published its first AI in Audit guidance and follow-up agentic AI guidance. The seven PCRT bodies have issued joint guidance for AI in tax. AML work sits under the Money Laundering Regulations and your professional body's supervisory framework.
Any agent worth running in a UK accountancy firm has to be built with this frame in mind, from spec through to live operation. Ours are designed to be ready for it. Every agent runs on isolated Azure tenancy. Client data stays inside infrastructure dedicated to your firm. Agent decisions can be logged with the inputs, the reasoning, and the output, so the audit trail is built in rather than retrofitted.
We are ISO 27001 certified, ICO registered, and UK GDPR aligned. Synextra, our parent company, has been running governance-grade Azure infrastructure for UK firms for many years.
These come up when we scope agents with accountancy firms. If yours isn't here, bring it to the call.
We build agents with both in mind. The July 2025 ICAEW Code update and the January 2026 PCRT joint guidance both shape how AI in accountancy work needs to be scoped and documented. Transparent documentation of where AI is used, human review of agent outputs against defined criteria, and named partner responsibility for the final work. The governance gets designed into the spec, not bolted on later.
Yes. Agent runs can be logged with the inputs, the agent's reasoning at each step, the output, and the human review record. The audit trail is exportable in the format your professional body expects. We design this in from the start because we know the work needs to stand up to FRC, ICAEW, and HMRC scrutiny.
That's a scoping question. We can build integrations into the systems most accountancy firms use, including IRIS, CCH, Xero, Sage, and the major document management platforms. If your stack is bespoke, we build to the spec at Design phase. The agent works inside your existing tools, not alongside them.
An AML or KYC agent would run against your firm's risk matrix and your chosen verification providers. The agent runs the screening, applies your risk rating logic, and produces a documented decision the MLRO can sign off. We design to the Money Laundering Regulations 2017 and your professional body's supervisory framework. Final responsibility stays with your MLRO, where it belongs.
This is a real and recurring issue in the sector. Dext's December 2025 survey of 500 UK accountants found 50% had clients suffering financial losses from public-AI tax advice. The answer isn't to ban AI in client communication. It's to put a governed agent at the point where errors arrive, screening client submissions against your standards before they cost you billable hours. We can scope an inbound-handling agent specifically for this. [3]
Failed outputs aren't billed. If an output doesn't meet the standard we agreed at Discovery, you don't pay for it. We fix the agent, run the workflow again, and absorb the cost. That is the operational commitment behind pay per output.
No. The aim is for agents to take on the structured, repeatable work your senior people are doing today and shouldn't be. The senior people get back to client advisory, planning, and the work they trained for. The goal is more output from the same headcount, not less headcount.
30 minutes. We'll map the process, scope an agent, and tell you honestly whether we can help. If we don't think an agent is the right answer for your firm, we say so.