Early access · UK limited-company year-ends

Year-end working papers, done in minutes.

Upload the Xero, QuickBooks or Sage export and the evidence. Get back a review-ready pack in your own Excel template, one exceptions list and the year-end journal. Client names never reach the model.

Early access for UK practices and the outsourcers that serve them. No licence, no minimum, no migration.

Works from the exports you already have

XeroQuickBooksSage
→ Lumpsum →
IRISTaxCalcBTCSoftwareExcel

No API connection, no migration. File exports in, files out.

How it works

One job in. Three things out.

A proven trial balance for IRIS or TaxCalc, the working-paper pack with its exceptions list, and the journal back to the ledger.

Inputs flow through Lumpsum's three lanes, code, model and person, to three outputsINPUTSTB + nominal ledger exportXero · QuickBooks · SagePrior-year trial balanceor last year's packEvidence PDFsbank · ageing · statements · returnsYour Excel templatetagged once at setupLUMPSUM · INSIDE THE FIRM'S BOUNDARYCODEEvery number· parse and cast the TB· ledger = TB per account· opening = prior closing· match to bank, ageing, statements· recalculate accruals, interest, VATMODELReading and drafting, masked· reads awkward PDFs into tables· suggests mapping + confidence· drafts variance notes from tokens· never sees a name or a balancePERSONJudgement and signature· confirms new mappings· clears the exceptions list· edits the drafts· signs the packOUTPUTSWorking-paper packyour Excel layout + PDF, every figure linkedExceptions listthe handful a person must decideMapped, proven TBready for IRIS · TaxCalc · BTCYear-end journalback into Xero, QuickBooks or Sage
Code does every calculation. The model reads, maps and drafts from masked payloads. A person confirms and signs. No number in the pack is generated by a model.
  1. 01

    Upload the exports and the evidence

    TB and ledger from Xero, QuickBooks or Sage, last year's TB, and the PDFs you already chase.

  2. 02

    Lumpsum builds and proves the pack

    Code casts, maps and matches every balance to its document. The model reads and drafts, from masked data only.

  3. 03

    You clear the exceptions and sign

    A short list of what needs a person. Sign in your own Excel; the journal goes back to the ledger.

The review surface

The Excel pack is the deliverable. This is where you clear it.

A reviewer's view of each engagement: what needs a person, which checks passed in code, and a download of the pack in your template at any point.

Open the live demo sample files, nothing to install

Lumpsum dashboard showing active engagements, open flags, evidence gaps and mappings to confirm
Flags, with the reason
Flags, with the reasonDuplicate, round sum, out of period, personal, unusual amount. Each with the Xero account, the IRIS account, the payee and what looked wrong.
Evidence per head
Evidence per headEvery balance-sheet head shows tied, differs or missing, with the document behind it. Drop the statement in and it is matched in code.
One engagement, seven stages
One engagement, seven stagesSource data, map, working papers, flags, evidence, import, back to the ledger. What needs you is always at the top.

The job today

A preparer spends the day moving numbers. A reviewer spends theirs finding what was missed.

Four to twelve hours a file, most of it re-keying and tying out. The reviewer's hours go on errors code would never make.

StepWhat goes wrongTime
Export TB, ledger and prior yearFormats differ by client; wrong date; subtotals misread30 min
Map new codes to lead schedulesFrom memory; last year's mapping not reused30–60 min
Tie each balance-sheet head to evidenceBy hand in its own tab; unmatched items buried2–4 hrs
Write the P&L variance notes"Up due to increased activity"; ledger never opened1–2 hrs
Build the pack and cross-referencesInconsistent between preparers; references break1 hr
Reviewer clears pointsMostly re-keying errors, not judgement1–2 hrs
Post adjustments back to the ledgerOften skipped; next year opens out of balance30 min, or never
With LumpsumEverything above except the judgement callsminutes ✓

Data protection

The AI never knows whose books these are.

Other tools send the raw ledger to a model and promise not to keep it. Here it is architectural: tokens leave the firm, names never do.

Inside the firm boundary, identifiers are detected, tokenised and stored in a vault; an outbound guard checks the payload; the model sees only tokens; the draft is rehydrated before the reviewer sees itINSIDE THE FIRM'S ENVIRONMENTOUTSIDELedger lineFenwick Timber Ltd£4,860.00 · 31 Mar20-45-67 11223344Detect, in coderegex: sort code, IBAN,UTR, VAT, NI, postcode…+ local name recogniserStable tokens (HMAC)ORG_017£4,860.00 · 31 MarBANK_ACCT_002🔒 Encrypted vaultper engagement, purgedat sign-offOutboundguardre-scans, blocksLanguage model seesORG_017 4860.00 PYhead: creditorstraining off · short retention · DPAdraft: "ORG_017 balance settled in April"Rehydratetokens → names, locally;unknown token = rejectedReviewer reads"Fenwick Timber balancesettled in April"Audit log: every call, its masked payload and the guard's verdict, hash-chained.
Model taskSentNever sent
Account mappingNominal code, account name, your lead-schedule listBalances, client name
Variance commentaryVariance table, top ledger entries with counterparties as ORG_nnnClient name, addresses, bank details, employee names
Supplier and lender statementsExtracted text with every identifier tokenisedThe raw PDF
Exception matchingUnmatched bank lines vs ledger lines, tokenisedAccount numbers

Where it sits

Balance's business model, applied to Silverfin's customer. For the job neither touches.

The UK year-end pack for small firms is still Excel and a person. Nobody sells the finished pack to the firm that keeps the client.

SilverfinBalance and the AI-native firmsOffshore outsourcersLumpsum
Sells toAccounting firms, mid-market and top 100The end businessSmall and mid UK practicesSmall UK practices and the outsourcers that serve them
You buyA platform to do the work inA finished monthly close; they become the accountantA person's hours, per fileA finished year-end pack, per file
Who keeps the clientThe firmBalanceThe firmThe firm, always
PricingPer file per year, from a 50-file minimum, 12-month contract, modules on topFrom £350 a month per clientPer file, 40–60% below UK costPer completed file, priced against the outsourcer, no licence or minimum
MigrationInto Silverfin templates and standard chartYour books move to themNoneNone: output lands in your own Excel
Client dataLives in their cloudLives with them; raw ledger to the model under contractOffshore, security variesMasked before any model call; single-tenant; audit-logged

From each company's published pricing and positioning, October 2026. Approximate.

Pricing

Per completed file.

Priced against what an outsourcer charges for the same pack. No licence, no seats, no minimum, no annual contract. If the pack is not usable, you do not pay.

The exact figure is being set with pilot firms.

Join the pilot waitlist
One file includesPilot pricing
  • ✓Cast and checked trial balance, mapped to your lead schedules
  • ✓Every balance-sheet head reconciled to its evidence
  • ✓P&L variance notes drafted from the entries behind each movement
  • ✓One exceptions list, signed off in your own Excel template
  • ✓Mapped TB for IRIS or TaxCalc, and the journal back to the ledger
Outsourcers: volume pricing and a white-label option. Ask on the waitlist form.

Questions

The ones partners ask first.

Does Lumpsum replace IRIS, TaxCalc or Silverfin?

No. It sits between the bookkeeping export and the accounts-production software. It produces the proven, mapped trial balance those tools start from and the working-paper pack that supports it. If you use Silverfin you probably do not need Lumpsum; most small practices do not use Silverfin.

Is client data sent to an AI model?

Only masked. Identifiers (names, account numbers, sort codes, UTRs, VAT and company numbers, postcodes, emails) are detected in code inside your environment and replaced with stable tokens. An outbound guard re-scans every payload and blocks anything that still looks like an identifier. The model never receives a client name or, by default, a real balance.

Does this breach UK GDPR or the EU AI Act?

The EU AI Act regulates use cases; matching invoices, mapping codes and drafting notes is minimal-risk and carries no high-risk obligations. UK GDPR and ICAEW/ACCA confidentiality rules do apply, and masking is what turns them from a contractual promise into an architectural fact: a data-processing agreement, training off and short retention on the model contract, plus a payload that contains no personal data. Not legal advice; the assessment is documented and available to pilot firms.

Do we have to change our working-paper template?

No. You tag your Excel template once at setup. Every pack comes out in that layout with your sign-off boxes and cross-references, and the totals are live formulas, not pasted values.

Which bookkeeping and accounts systems are supported?

Exports from Xero, QuickBooks Online and Sage are read today, including sectioned ledger reports and PDFs with a text layer. Output is an import TB for IRIS, with TaxCalc and BTCSoftware layouts to follow, and journals in Xero, QuickBooks and Sage 50 import formats.

Who signs the accounts?

You do. Lumpsum prepares; a person in the firm clears the exceptions list, edits the drafts and signs. Nothing is filed and nothing is written to a client ledger without that sign-off.

Early access

Stop paying for the hours a program should be doing.

Pilot places for UK practices and outsourcers. Bring one messy client; we will hand back the pack.

Early access for UK practices and the outsourcers that serve them. No licence, no minimum, no migration.