Find the break in the week it happened, not at year end.
Two systems hold the same figures and they stop agreeing quietly. This agent matches them line by line every day, closes off everything that ties out, and hands your team the small remainder with an explanation already attached.
What the reconciliation agent does
Match. Flag. Explain.
Reconciliation is the clearest example of work that is essential and almost entirely mechanical. Somebody exports two sets of records, sorts them, and spends a morning agreeing the ninety per cent that were always going to agree in order to arrive at the ten per cent that need thinking about.
The agent does the first part and hands over the second. It applies your matching rules, including the tolerances and the reference handling that make your data particular, and closes off everything that ties out. What is left is a break report where each item carries a proposed reason. A timing difference, a partial payment, an entry missing on one side, a duplicate. Your team confirms or overrides.
Running it daily is what changes the economics. A discrepancy found the following morning can be traced back to the transaction that caused it while everyone still remembers the week. The same discrepancy found at the next quarter end means reopening everything that moved in between, and that is where the fortnight goes.
This agent works on records that already exist. Where those records still arrive as documents, the data extraction agent gets them into structured form first. The management reporting agent then turns reconciled figures into the pack, and the data quality agent deals with the duplicates and inconsistencies that cause a good share of breaks in the first place.
Capabilities
What it handles.
Line level matching across two sources
Compares the records in one system against the records in another, item by item. Bank against ledger, invoicing against the client system, supplier statement against the purchase ledger. What matches is closed off without anyone looking at it.
Matching rules you define
Real records rarely match exactly. Rounding, timing differences, references entered three ways and payments that arrive as one lump for four invoices. The tolerances and the fuzzy matching rules are set to how your data actually behaves, not to a textbook.
Breaks explained, not just listed
A list of unmatched items is the start of the work, not the end of it. The agent proposes what each break is. A timing difference, a partial payment, a missing entry on one side, a duplicate. Your team confirms or overrides rather than investigating from nothing.
Ageing on anything unresolved
Items that stay open are tracked from the day they first appeared. The break that has been rolling forward for five months is visible as exactly that, rather than reappearing each period as if it were new.
Runs daily rather than once a month
The cost of reconciliation is not the matching, it is the delay. A discrepancy found the day after it happens can be traced. The same one found seven weeks later means going back through everything that moved in between.
Traceable to both source records
Every match and every break records both underlying entries and the rule that produced the decision. When somebody asks why an item was closed off, the answer is on the record rather than in the head of whoever ran it.
Who it is for
Built for teams who carry the cost of a late discovery.
Finance teams
Bank reconciliations, control accounts, intercompany balances and supplier statements. The work that has to be right, has to be done every period, and consumes the days when the same people are also closing the month.
Operations and client accounting
Client money, fee ledgers and transaction volumes that have to agree across an internal system and an external one. High volume, low judgement, and expensive when a break is found late.
Firms running two systems that should agree
A CRM and an invoicing system. An operational platform and the accounts. A broker statement and an internal record. Anywhere the same fact is held twice, the two versions drift, and somebody checks quarterly at best.
Managing directors
The number you are being shown is only as good as the reconciliation behind it. A daily match means the figures in the management pack have already been checked rather than checked afterwards if something looks wrong.
What is included
What you get.
Matching across the two sources you nominate, connected to how each one currently exports or exposes its records.
Matching rules, tolerances and reference handling agreed during the build and tuned against your own historic data.
A break report showing what did not match, the proposed reason for each item and how long it has been open.
Ageing on unresolved items so long-standing breaks are visible as long-standing rather than new.
A daily run on the schedule you set, with an alert when a break crosses the value or age threshold you define.
A full audit trail on every decision, recording both source records and the rule that matched or flagged them.
How it gets built
How it gets built.
01
Agree the two sides
We pick the single reconciliation that costs your team the most time and identify how each side makes its records available. One pairing is scoped, built and proven before anything else is added.
02
Learn how your data actually behaves
We work through your historic records to see how references are entered, where timing differences arise, how payments are grouped and what the recurring categories of break are. The matching rules come out of that rather than out of an assumption.
03
Run it against periods you have already closed
The agent is tested on months your team reconciled by hand and the two results are compared. Did it match what you matched, did it flag what you flagged, and did it call anything a match that was not one. Tolerances are tuned here rather than after go live.
04
Set the schedule and the thresholds
You decide how often it runs, what value or age of break raises an alert rather than waiting for the next report, and who receives it. Write back to a source system, if you want it at all, is configured deliberately with approval steps.
Why Delancy
Against a spreadsheet and a fortnight at month end.
Built around your data, not a standard template
Reconciliation fails on the exceptions, and the exceptions are specific to how your systems are used. The rules are built against your real records, including the messy ones, rather than assuming a clean reference on every line.
The explanation is the output
Producing a list of unmatched items is easy and does not save anybody time. The work your team currently does is working out why each one is unmatched. That is the part the agent is built to take on.
Frequency is the saving
Moving reconciliation from monthly to daily changes what a break costs to resolve. It is the same task done at a point where the surrounding context still exists, which is why it stops taking a fortnight.
No ongoing dependency
The agent runs in your environment against your systems, with the rules documented and the price agreed before build. You are not renting seats on a reconciliation platform to keep your own numbers.
Frequently Asked Questions
Common Questions.
Anything where the same fact is recorded in two places and the two should agree. Bank to ledger, invoicing to client system, supplier statement to purchase ledger, control account to subledger, broker or platform statement to internal records. The pattern is the same. What changes between them is the matching rules, which is what gets built.
That is the normal starting position and it is the reason exact matching alone is not enough. The build works through your historic data to learn how references are actually entered, how payments are grouped and where timing differences arise, then sets the rules against that. Anything the rules cannot resolve confidently becomes a break rather than a guess.
Not unless you decide it should. The default is read only. The agent produces the matched set and the break report, and your team posts the adjustments. Where a firm wants write back on the clean matches, that is configured deliberately with approval steps, not switched on because it is technically possible.
The data extraction agent turns a document into a record. This agent works on records that already exist and checks two sets of them against each other. They sit next to each other in the same process. Extraction gets the supplier statement into a structured form, reconciliation is what then matches it to your ledger.
They go to a person, which is the correct outcome. The agent handles the volume of straightforward matching and the routine categories of difference. What is left is the small set that needs judgement, and your team gets it with both source records attached and the ageing already on it rather than having to assemble that first.
Pilots start from £2,500. The price is agreed before any build starts. Where the agent relies on third-party services to run, those operational costs sit with you and are made clear before build begins.
FREE ASSESSMENT
Is an AI agent the right shape for this?
Ten questions on the task, the volume and the output you need. You get a diagnosis of whether an agent fits, not a score.
Take the assessmentWork out what your last reconciliation actually cost.
Book a discovery call. We will take the reconciliation that eats the most time, look at how the two sides behave, and show you what a daily run would have caught.
Book a discovery call →