Karbon AI is a set of AI features built into Karbon's practice management software for accountants. The main capabilities are email triage and reply drafting, document and client summarisation, suggested next actions on triage items, and chat-with-your-client-data style queries against the Karbon database. It runs on top of OpenAI's models and operates only on data already inside Karbon. It does not integrate with Xero, QuickBooks, AML providers, or any system outside Karbon's own boundary.

How much does Karbon AI cost in the UK?

Karbon AI is bundled into Karbon's mid-tier and Enterprise subscriptions in 2026 and varies by region and contract. UK pricing typically lands around £15-£35 per user per month for the AI add-on on top of the core Karbon Practice subscription, which itself runs roughly £45-£75 per user per month. A 10-user practice on Karbon Practice plus AI is therefore looking at £600-£1,100 per month all in for Karbon, before any other tools.

What can Karbon AI not do?

Anything outside the Karbon database. The big gaps for UK practices: it does not pull data from Xero or QuickBooks for working-paper preparation, it does not integrate with AML providers like Smartsearch or Veriphy, it does not handle client onboarding documents or DocuSign sends, it does not chase missing receipts or bank reconciliation exceptions in the bookkeeping software, and it does not generate engagement letters with custom matter-specific paragraphs. For those workflows, you need either separate tools or a custom build.

Is Karbon AI worth it for a small UK practice?

Yes if your bottleneck is email triage, internal communications, and Karbon-internal task management. The product is genuinely good at those. No if your bottleneck is the cross-system work (client onboarding, AML, year-end working papers, missing-records chase, bookkeeping data hygiene), because Karbon AI doesn't reach those workflows by design. Most small UK practices have both kinds of bottleneck, and the right answer is usually to keep Karbon AI for what it does well and add a custom layer for the rest.

Can a custom build replace Karbon AI?

Mostly yes for the data and workflow side, but the email triage and templated reply features are non-trivial to replicate and rarely worth replacing. Custom code can read and write to Karbon over its REST API, can call Claude or GPT for the same drafting and summarisation tasks, and can integrate with anything outside Karbon. The pieces that make Karbon AI pleasant to use (the inline email assistant, the suggested-action button in the Karbon UI) require building inside the Karbon UI and that gets expensive fast. Most custom builds I do for accountancy clients keep Karbon AI for in-Karbon work and add custom for cross-system work.

When does a custom build pay back vs Karbon AI?

When a single workflow outside Karbon's reach is costing the practice meaningful billable hours. If your bookkeepers spend 6 hours a week chasing missing receipts in Xero, that's £200-£300 a week of cost that no Karbon feature touches. A £15,000 custom workflow that automates the chase pays back in roughly 12-18 months on hours saved, and continues paying back at zero marginal cost as the practice grows. If the practice has multiple cross-system bottlenecks, the payback shortens further.

Karbon AI vs custom build for UK accountancy practices in 2026

You bought Karbon two years ago because the practice management was a step up from whatever you were on before. Six months ago you turned on Karbon AI and it's been useful. Email triage is faster. Drafting client replies is faster. The suggested-action stuff occasionally surfaces something useful.

Then you sit down on a Friday afternoon and notice that the actual painful workflows haven't moved. Bookkeepers are still chasing receipts in Xero. The year-end prep junior is still copying numbers from QuickBooks into a working-paper template. Onboarding a new client still takes 90 minutes spread across two weeks of email chasing. Karbon AI doesn't do any of that.

That's the question this post answers. What Karbon AI is genuinely good at, where the line is, and when a custom build alongside Karbon starts making sense for a small UK practice. I'm Robin, I run Laires Labs, a solo AI engineering studio in London. Most of my accountancy work involves Karbon, AccountancyManager, IRIS, or CCH at the practice management layer.

The short answer

Karbon AI is good at email triage, drafting client replies inside the Karbon UI, summarising documents and conversations, and generating suggested next actions inside the Karbon database. It is not designed to do anything outside Karbon. The high-cost workflows in most small UK practices (client onboarding plus AML, year-end working-paper preparation, missing-records and bank-rec chase, cross-system reporting) all live outside Karbon's data boundary. For those, Karbon AI doesn't help, and a custom build that runs alongside Karbon is usually the right answer. Most practices end up keeping Karbon AI for what it does well and adding a custom layer for the rest.

What Karbon AI actually does well

Stripped of marketing language, Karbon AI is a small handful of features built into the Karbon UI that call OpenAI models against data already in Karbon. The features that earn their keep:

Email triage and reply drafting in Triage. When inbound email lands in Karbon Triage, the AI suggests a draft reply, pulls relevant context from the client record, and lets the user accept, edit, or rewrite. For practices that handle a lot of routine client correspondence (where's my tax return, please confirm the deadline, can you send me a copy of last year's accounts), this saves real time. Drafting quality is good for routine queries and noticeably weaker for anything that requires technical advice.

Document and conversation summarisation. Long email chains, attached documents, and meeting notes can be summarised on demand. Useful for briefing yourself on a client matter you haven't touched in three months.

Suggested next actions. Karbon AI surfaces "what should happen next" on a job or a client, based on the activity history and any work templates configured. Hit-rate is mixed but the feature reduces the amount of "wait, where did I leave this" friction.

Chat-with-your-data inside Karbon. A chat interface lets you ask questions about clients, jobs, deadlines, and tasks in natural language. Usefulness depends on how well the practice has structured its Karbon data. A messy Karbon instance returns mediocre answers; a well-organised one is genuinely productive.

These are real features and the product is well-executed for the use case it targets. The use case is "make working inside Karbon more efficient." Which is also the limit.

Karbon AI product page showing the bundled AI assistant features being compared against a custom build for UK accountancy.

Where Karbon AI stops

The features above operate on Karbon's own data: clients, jobs, tasks, emails, documents stored inside Karbon, and the work template configuration. Karbon AI cannot reach outside that boundary by design. The things it does not do:

It does not read from Xero, QuickBooks, FreeAgent, or Sage Business Cloud. Year-end working-paper preparation, the highest-cost workflow in most accountancy practices, requires pulling trial balances, journal data, and ledger detail from the bookkeeping software. Karbon AI cannot see any of it.

It does not integrate with AML providers. Smartsearch, Veriphy, AMLCC, ID-Pal, Credas. Client onboarding workflows that touch MLR sit outside Karbon's data model and outside what Karbon AI can automate.

It does not chase missing receipts or bank-rec exceptions in bookkeeping. This is the chronic admin pain in most small practices. The data lives in Xero or QuickBooks. Karbon AI doesn't reach it.

It does not draft engagement letters with matter-specific paragraphs. Karbon has document templates, but they are template merges, not LLM-drafted text generation. A custom layer can use Claude or GPT to write the matter-specific paragraph that explains the scope of an unusual engagement; Karbon AI cannot.

It does not run cross-system reporting. A dashboard pulling capacity from Karbon, billable utilisation from your time tracker, and year-end status from Xero requires reaching all three systems. Karbon AI sees only Karbon.

It does not run scheduled batch jobs against your data. Weekly capacity reports, monthly client revenue analysis, quarterly write-down monitoring. These are jobs that read Karbon data, transform it, and produce outputs that go somewhere else (a partner email, a Slack channel, a spreadsheet). Karbon AI does not have a scheduler.

None of this is a criticism of Karbon. It's a vendor scoping its product. The point is that the marketing language ("AI for accountants", "AI-powered practice management") tends to leave buyers thinking Karbon AI covers more ground than it does.

Five highest-ROI accountancy automations: how each one fares

In my accountancy buyer's guide I ranked the five workflows that pay back fastest in a 3-30 staff UK practice. Here's how Karbon AI handles each.

1. Client onboarding and MLR document collection

Karbon AI: partial. Karbon has work templates that create the right tasks on a new-client trigger and AI can draft the welcome email. The actual MLR check, ID verification, engagement letter drafting with matter-specific paragraphs, and DocuSign signature flow are not in Karbon AI's scope.

Custom build needed for: AML provider integration, LLM-drafted engagement letters, signature flow, automatic Xero/QuickBooks setup on signature.

2. Year-end working-paper preparation

Karbon AI: none. The trial balance and ledger data lives in Xero or QuickBooks, outside Karbon's reach.

Custom build needed for: Xero/QuickBooks API integration, working-paper template population, anomaly flagging, queries-for-client list generation.

3. Missing-records and bank-reconciliation chase

Karbon AI: none. The missing data is in the bookkeeping software, not in Karbon.

Custom build needed for: bookkeeping software watch, gap identification, drafted client email chase, escalation rules, status update back to Karbon.

4. Client email triage and query routing

Karbon AI: strong. This is the use case Karbon AI was built for. Drafting, classification, suggested replies, and triage assistance work well.

Custom build needed for: if Karbon AI handles this well in your practice, no custom work is needed. If you have specialised triage rules (route VAT queries to one team, payroll queries to another, anything sensitive to a partner), custom routing logic on top can extend Karbon's defaults.

5. Capacity and billable utilisation tracking

Karbon AI: partial. Karbon shows job status and time recorded; the AI can summarise but does not model forward capacity or surface over-allocation across the team.

Custom build needed for: capacity modelling that reads from Karbon, projects forward, and emails partners weekly. Possibly cross-system if you use a separate time tracker.

The pattern: Karbon AI handles workflow 4 well, partly handles workflow 1, and does nothing for workflows 2, 3, and 5. The workflows it doesn't handle tend to be the highest-cost ones in most practices.

When custom code is the right answer

Three signals that say a practice has outgrown Karbon AI alone.

The painful work happens in Xero or QuickBooks, not in Karbon. If your bookkeepers spend more time in the bookkeeping software than in Karbon, the automation that matters has to reach the bookkeeping software. Karbon AI doesn't.

You're hiring or thinking about hiring to handle workflow volume. If the next hire is to handle onboarding throughput, year-end prep capacity, or missing-records chase, that hire's full-year cost (£28,000-£45,000 in 2026 for a UK accounts assistant or junior) compares directly with a custom build that automates the same workflow.

You have a process that's specific to how your practice runs. Every practice has 1-3 workflows that are unique to them: a particular engagement letter format, a specific compliance pack for a vertical client base, a custom dashboard for the partners. Karbon AI is general-purpose by design and won't bend to specifics. Custom code does.

How a Karbon-plus-custom stack actually works

The practical pattern I ship for accountancy clients running on Karbon.

Karbon stays as the system of record. All clients, jobs, tasks, deadlines, and recorded time live in Karbon. The custom layer reads from and writes to Karbon over the REST API.

Karbon AI keeps doing what it does well. Email triage, reply drafting, in-Karbon summarisation. No replacement attempted, no parallel system built for that.

The custom layer handles cross-system workflows. Client onboarding plus AML. Year-end working-paper prep against Xero or QuickBooks. Missing-records chase. Cross-system reporting and partner dashboards. Anything that needs to read from or write to a system outside Karbon.

Triggers are coordinated. A new client created in Karbon triggers the custom onboarding workflow. A new year-end job in Karbon triggers the custom working-paper prep. A weekly schedule fires the missing-records chase across all clients. The trigger lives wherever it makes sense; the work runs in the right place.

Outputs land in the right tool. Drafted documents land in Karbon as attachments. Tasks land as Karbon tasks. Email goes via your existing mail provider, logged back to Karbon as activity. Data dashboards live in Looker, Metabase, or a custom UI, depending on the practice.

The custom layer is small. Most builds I do are one or two workflows for a 5-15 staff practice, total under £25,000 to ship. Larger practices commission more workflows; the architecture is the same.

Cost over 2 years: Karbon AI alone vs Karbon plus custom

For a 9-staff UK practice. Both paths keep Karbon as the practice management system. The custom path adds a workflow layer for the three workflows Karbon AI doesn't reach (onboarding plus AML, year-end working-paper prep, missing-records chase).

Karbon AI alone path

Karbon Practice (9 users): £69/seat/month × 9 × 24 = £14,904
Karbon AI add-on: £25/seat/month × 9 × 24 = £5,400
Dext Premium (3 users): £40/seat/month × 3 × 24 = £2,880
AML provider (Smartsearch): £80/month × 24 = £1,920
Ignition for engagement letters: £75/month × 24 = £1,800

2-year total: £26,904. The cross-system workflows (year-end prep, missing-records chase) are still done manually by junior or bookkeeper time. Estimate: 6-10 hours per week of unrecovered admin across the practice, costing roughly £18,000-£30,000 per year in fully-loaded staff time depending on who absorbs it.

Karbon plus custom path

Karbon Practice (9 users): £69/seat/month × 9 × 24 = £14,904 (same)
Karbon AI add-on: £25/seat/month × 9 × 24 = £5,400 (same; kept for what it's good at)
Dext Premium (3 users): £40/seat/month × 3 × 24 = £2,880 (same)
AML provider (Smartsearch): £80/month × 24 = £1,920 (same)
Ignition for engagement letters: replaced by custom = £0 (was £1,800)
Render or AWS hosting for custom layer: £40/month × 24 = £960
LLM API costs: £150/month × 24 = £3,600
Custom build: £18,000 one-off
Retained maintenance: £3,000-£6,000 across 2 years

Karbon AI alone versus Karbon plus custom build 2-year cost comparison for a 9-staff UK accountancy practice with unrecovered admin time included.

2-year total: £50,664-£53,664. The custom path costs roughly £24,000-£27,000 more over 2 years. In return, the cross-system admin time (estimated £36,000-£60,000 of staff cost across 2 years) drops materially. Net: the custom path is positive on year one if the freed staff time gets redeployed into billable work, and clearly positive over 2 years on most realistic redeployment assumptions.

The argument for keeping Karbon AI inside the custom path: the email triage and in-Karbon drafting are real time-savers and replicating them in custom code costs more than the Karbon AI subscription. Pay the Karbon AI fee, build custom for what Karbon AI doesn't reach.

When to skip the custom build

Three cases where Karbon AI alone is enough.

Practices under five staff. The cost of a £15,000-£25,000 custom build doesn't pay back fast enough at this scale. Use Karbon AI for what it does well, accept the cross-system admin friction, and revisit when the practice grows.

Practices where the bottleneck genuinely is internal communication. If your team's biggest pain is internal email volume and Karbon-internal task management, Karbon AI handles that. A custom build wouldn't help.

Practices in transition or restructuring. Mid-pivot, mid-merger, mid-systems-change. Wait for the new shape of operations to settle before committing to custom work. Automation built against the wrong workflow is a sunk cost.

Book a call

If you're on Karbon, you've turned on Karbon AI, and you can tell that the actual painful workflows aren't getting touched, that's the call to book. Twenty minutes, no pitch. Tell me your practice size, the two workflows your team complains about most, and how much of your work happens in Xero or QuickBooks vs Karbon. I'll come back with a one-page write-up of what a custom layer alongside Karbon would build, what it would cost, and which workflows I'd leave in Karbon AI's hands. Yours to keep whether you hire me or not.

Don't book if your practice is under three staff, or if you're mid-systems-change, or if the work is genuinely all inside Karbon already. In those cases Karbon AI alone is the right answer.

Book a 20-minute call or read the full accountancy automation buyer's guide if you're earlier in the decision.