Clean records, and rules that keep them that way.
Duplicates, formats entered six ways, empty fields and contacts who left two years ago. This agent finds them, proposes the fix, and then applies the same rules to every new record so the problem does not rebuild itself.
What the data quality agent does
Find it. Fix it. Keep it fixed.
Every system that people type into ends up in roughly the same state. The same company under three spellings, a contact who left in 2023 still sitting on the active list, phone numbers in five formats and a required field that half the records never had filled in. Nobody caused it. It is what a decade of ordinary use looks like.
The agent works through what you hold and reports the actual position, which is usually worse and more specific than expected. Duplicate groups come back with a merge proposed and a surviving value chosen for each field. Formats are standardised to your conventions. Completeness is measured against the fields you say a record needs, and stale records are flagged as no longer current rather than left to look live.
Then it keeps going, which is the part that matters. Most firms have already run a cleanse and watched the position degrade, because the clean fixed the records without changing how records get created. The same rules run against new and edited entries on arrival, so the work holds instead of becoming something to redo before the next audit.
This is usually the agent that makes the others worth having. The reconciliation agent stops chasing breaks that were only ever duplicate entries, and the management reporting agent produces a pack nobody spends the meeting arguing with.
Capabilities
What it handles.
Duplicate detection with a merge proposed
The same company under three spellings, the same person against two records, the same supplier entered twice by two people. The agent identifies the group, works out which record holds the best version of each field and proposes the merge for approval.
Format standardisation
Phone numbers, postcodes, company suffixes, country names, dates and job titles entered a dozen ways over a decade. Standardised to the conventions you set, so a search or a segment returns what is actually there.
Missing fields against your required schema
You define which fields have to be present for a record to be usable and for what purpose. The agent reports what is missing, where the gap blocks a process, and which gaps can be filled from another record you already hold.
Stale records surfaced
Contacts who left the company, accounts nobody has touched in two years, addresses that no longer exist. Records that are not wrong so much as no longer true, which is the harder category to see and the one that quietly degrades every list.
Rules applied on arrival
The reason a cleanse project decays is that nothing changed about how records get created. The same rules run against new and edited records as they enter the system, so the position holds instead of degrading from the day the project finished.
A person approves anything destructive
Merges and deletions are proposed, not performed. Standardisation and enrichment of empty fields can run automatically where you want them to. Nothing that loses information happens without somebody agreeing to it.
Who it is for
Built for teams working from records they do not quite trust.
Sales teams and CRM owners
A pipeline report is a summary of the records underneath it. Duplicate accounts, contacts who moved on and missing fields mean the forecast, the segmentation and the outreach are all built on something nobody quite trusts.
Operations and client data teams
Client and supplier records that feed billing, delivery and service. A duplicate here is not an inconvenience, it is an invoice sent twice or a job routed to an address that changed eighteen months ago.
Marketing teams
Every campaign is limited by the list. Duplicates cost money and look careless to the recipient, missing fields break personalisation, and stale contacts damage deliverability for everyone else on the send.
Anyone whose reporting is questioned
When the meeting turns into an argument about whether the number is right, the underlying issue is almost always the records rather than the report. Clean data is what makes the reporting conversation about the business again.
What is included
What you get.
A first pass profile of the records you hold, showing where duplicates, gaps and inconsistencies actually sit and at what volume.
Duplicate detection with a proposed merge and a chosen surviving value for each field.
Standardisation of formats and naming conventions to the rules you set.
Completeness checks against the fields you define as required, with the blocking gaps identified.
Stale record flagging so contacts and accounts that are no longer current are visible.
The same rules applied to new and edited records on arrival, so the improvement holds rather than decays.
How it gets built
How it gets built.
01
Profile what you actually hold
Before any rules are written we measure the current position. How many likely duplicate groups, which fields are empty and how often, where formats diverge and what proportion of records look stale. That profile is useful on its own and it sets the priorities.
02
Define what good looks like
You decide which fields are required and for what purpose, what the conventions are, and which signals make two records the same entity. This is a business decision rather than a technical one, and it is what the whole agent runs on.
03
Work the backlog under review
The existing records are processed in batches, with merges and deletions proposed rather than applied. Your team reviews the first batches closely, which is also how the matching confidence gets tuned before the volume goes through.
04
Put the rules on the front door
The same rules are applied to new and edited records as they arrive, so a duplicate is caught at creation rather than found next year. This step is what separates this from a cleanse project, and it is the reason the position holds.
Why Delancy
Against a cleanse project that gets done once and decays.
The point is that it keeps holding
Most firms have already run a cleanse. The records were fixed, nothing changed about how they get created, and eighteen months later the position was the same. Applying the rules continuously is the difference between a project and a fix.
Judgement on the hard cases
A matching script finds identical records, which were never the problem. The difficulty is deciding whether two similar records are the same entity, and which of two conflicting values is the current one. That is what the agent is configured for.
It makes the other work worth doing
Reconciliation and reporting both sit on the same records. Cleaning the underlying data is what stops a share of the breaks appearing and what stops the pack being queried, so it tends to pay twice.
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 data quality platform to keep your own records usable.
Frequently Asked Questions
Common Questions.
No. Anything that loses information is proposed rather than performed, and a person approves it. Merges and deletions always go through review. Reversible corrections such as format standardisation and filling an empty field can run automatically where you choose to allow it, and every change is logged either way.
On the signals that are reliable in your data rather than on name similarity alone. Registration numbers, domains, addresses, account references and relationship history all carry weight. Confident matches are grouped for merge and borderline cases are presented as a question rather than resolved quietly.
Because a cleanse fixes the records and changes nothing about how records are created, which is why the position degrades from the day it finishes. This agent applies the same rules to new and edited records as they arrive. The one-off clean is the starting point rather than the deliverable.
Wherever your records live. CRM, practice management, operational platforms, finance systems and the spreadsheets that sit alongside them. The build scopes the systems in and out explicitly, and it is normally better to prove the agent on the single system causing the most trouble before extending it.
The data extraction agent creates records from documents. This agent works on the records already sitting in your systems, however they got there. In practice they pair well, because extraction with no quality rules behind it simply adds well-structured duplicates at a faster rate.
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 assessmentFind out how many duplicates are sitting in your system today.
Book a discovery call. We will profile the records in the system causing the most trouble and show you the real position before anything gets built.
Book a discovery call →