Back to Articles

Data Quality for Business Reporting - A UK Guide to Fixing Bad Data Before It Reaches Your Dashboards

Data Quality for Business Reporting - A UK Guide to Fixing Bad Data Before It Reaches Your Dashboards

Every UK business that reports from a database eventually hits the same wall: the dashboard looks polished, but nobody trusts the numbers. Data quality for business reporting is the unglamorous discipline that decides whether your sales figures, stock counts and cash forecasts can be acted on — or whether every board meeting starts with twenty minutes of arguing about whose spreadsheet is right. Fix it at source and reporting becomes a tool; ignore it and even the most expensive Power BI licence becomes an elaborate way of displaying errors.

This how-to guide walks through the practical process we use with UK SMEs and mid-market firms: how to measure data quality, which data quality checks to automate first, how to set up lightweight master data management without an enterprise budget, how data cleansing fits into the reporting pipeline, and how to keep a single source of truth once you have one. You will find realistic timelines, cost ranges in pounds, a scoring framework, a checklist you can hand to your team on Monday, and the mistakes that quietly undo most clean-up projects within six months.

What is data quality in business reporting? Definition and context

Data quality is the degree to which the records in your systems are fit for the decisions you make from them. In a reporting context that sounds abstract, so it helps to break it into the six dimensions most data teams use, which line up closely with the guidance in the UK Government Data Quality Framework published by the Government Data Quality Hub:

  • Completeness — are the fields you report on actually populated? A customer table where 30% of records have no region cannot produce a reliable regional sales report.
  • Accuracy — does the value reflect reality? A product cost of £0.00 or a delivery date in 1900 is complete but wrong.
  • Consistency — does the same fact match across systems? If your CRM says a client is “Acme Ltd” and Xero says “ACME Limited”, any report joining the two will under-count.
  • Timeliness — is the data current enough for the decision? Yesterday’s stock levels may be fine for a weekly review and useless for a same-day despatch decision.
  • Uniqueness — is each real-world thing recorded once? Duplicate customers inflate counts and split lifetime value across records.
  • Validity — does the value conform to the rules? UK postcodes, VAT numbers, Companies House numbers and email addresses all have formats you can test automatically.

The important shift is to treat data quality as a property of the pipeline, not of the report. Most businesses try to fix quality in the last mile — a finance analyst who “tidies up” an export in Excel every Monday morning. That works until the analyst is on holiday, and it means the clean version of the truth never makes it back to the source system. A durable approach moves the checks upstream: validation at the point of entry, automated tests as data lands in the reporting database, and clear ownership of each core dataset.

There is also a compliance angle that UK firms cannot ignore. Under UK GDPR Article 5(1)(d), personal data must be “accurate and, where necessary, kept up to date”, and the ICO expects you to take reasonable steps to correct or erase inaccurate records. Duplicate and stale customer records are therefore not just a reporting nuisance; they are a data protection risk, particularly when they drive marketing or credit decisions.

Pro Tip

Before buying any data quality tool, pick the three reports your leadership team relies on most and trace each figure back to its source tables. In most SMEs, 80% of the trust problems come from fewer than ten fields — typically customer identity, product codes, dates and status values. Fix those first.

Data quality scoring — where most UK businesses sit today

When we run an initial data quality assessment for a new client, we score each of their core datasets against the six dimensions above. The pattern is remarkably consistent across sectors. Transactional data — invoices, orders, payments — is usually in decent shape because accounting software enforces structure. Reference and master data — customers, suppliers, products, staff — is where the damage lives, because it is typed in by many people, across many systems, over many years.

The score cards below summarise the typical risk profile we see in a 20–250 person UK business that has grown through a mix of off-the-shelf SaaS tools and spreadsheets.

Customer and supplier master data
Duplicate records across CRM and financeHigh risk
Missing or free-text region and sector fieldsHigh risk
Stale contact details (UK GDPR accuracy)Medium risk
Companies House / VAT numbers unvalidatedMedium risk
Consistent customer ID shared between systemsHigh risk
Product and stock data
SKU naming inconsistent between channelsHigh risk
Cost prices missing or out of dateMedium risk
Discontinued items still activeMedium risk
Unit-of-measure mismatches (each vs case)Medium risk
Stock movements recorded in real timeLow risk
Financial and transactional data
Invoices posted to correct nominal codesLow risk
Manual journals without supporting notesMedium risk
Currency and VAT treatment consistentLow risk
Sales orders linked to invoicesMedium risk
Cost centre / department tagging completeHigh risk

Notice where the high-risk badges cluster: on the attributes you slice by rather than the numbers you add up. Revenue totals are usually right; revenue by region, by sector, by account manager or by product family is usually wrong. That is exactly why leadership teams lose faith — the headline matches the accounts, but the breakdowns they want to act on do not reconcile.

The cost of poor data quality — the numbers that matter

Poor data quality rarely appears as a line on the P&L, which is why it gets deprioritised. It shows up instead as analyst hours, mis-targeted campaigns, write-offs and slower decisions. Gartner has long estimated that poor data quality costs organisations an average of around $12.9 million a year at enterprise scale; for a UK SME the absolute figure is smaller, but the proportion of wasted effort is often higher because there is no dedicated data team absorbing it.

The figures below are typical ranges we see during discovery with UK businesses of 20–250 staff. They are illustrative rather than guaranteed, but they give you a sense of where to look in your own organisation.

6–12 hrs
Analyst time per week spent reconciling and correcting report exports in a typical SME finance team
8–15%
Typical share of duplicate customer records when CRM and accounting systems are not linked by a shared ID
£8k–£35k
Usual range for a first-phase data quality and cleansing project for a UK SME
6–10 wks
Realistic timeline from assessment to automated checks on the core reporting datasets

The analyst-hours figure is the easiest to quantify. At a fully loaded cost of around £35–£45 an hour for a finance or operations analyst, eight hours a week of manual reconciliation costs £14,500–£18,500 a year — before you count the decisions delayed while people wait for “the corrected version”. In many cases a first-phase project pays for itself on analyst time alone within 12–24 months, although the more significant value is usually in decisions made faster and with more confidence.

Fix in the report or fix at source? Comparing the two approaches

There are two broad ways to tackle data quality for reporting. The first is to clean data downstream, inside the reporting layer — Power Query steps, SQL views that patch bad values, Excel lookups that map “Mcr” to “Manchester”. The second is to fix data at source and validate it as it flows into a governed reporting database. Most mature setups use both, but the balance matters enormously for long-term cost.

Downstream patching

Clean it in the report layer

Time to first result Days
Upfront cost Low (£1k–£5k)
Source systems corrected No
Rules visible to the business Hidden in queries
Fragility as data grows High
UK GDPR accuracy obligation Not addressed
Best for Quick fixes, one-off analyses

Source-first quality pipeline

Validate on entry, test on load

Time to first result 6–10 weeks
Upfront cost Medium (£8k–£35k)
Source systems corrected Yes
Rules visible to the business Documented and owned
Fragility as data grows Low
UK GDPR accuracy obligation Supported
Best for Recurring management reporting

Downstream patching is not wrong — it is often the right first move to get a credible number in front of the board this quarter. The problem is that every patch is a silent rule. When the person who wrote the Power Query step leaves, nobody knows why “Region = blank” was being reassigned to “South East”. A source-first pipeline makes those rules explicit, testable and owned, which is what lets a reporting estate scale beyond one person’s head. If you are still at the spreadsheet stage, our guide to moving from spreadsheets to database-driven dashboards covers the groundwork that comes before this step.

Where data quality problems come from — root causes ranked

Fixing symptoms without understanding causes guarantees that the problems return. When we log the root cause of every data quality issue found during an assessment, the distribution looks broadly like the chart below. The percentages show how often each cause contributes to the issues found in a typical SME engagement (many issues have more than one cause, so they do not sum to 100).

Free-text fields where a picklist belongs
71%
No shared ID between systems
64%
Manual re-keying between tools
58%
No named owner for the dataset
52%
Legacy data migrated without cleansing
43%
Integration failures not alerted
37%
Conflicting metric definitions
33%

Two things stand out. First, most causes are design problems rather than people problems. Staff do not enter “Bham”, “B’ham” and “Birmingham” out of carelessness; they do it because the form gives them a free-text box. Replacing that box with a validated dropdown removes an entire class of errors permanently. Second, silent integration failures are more common than most teams realise. A Zapier or Make scenario that has been quietly erroring since March is a classic cause of “missing” orders in a sales report — and nobody notices until a customer chases an invoice.

The last item, conflicting metric definitions, is technically a semantic problem rather than a data problem, but it produces exactly the same symptom: two reports disagree. If sales count an order on signature and finance count it on invoice, both numbers are “right” and the business still cannot agree on revenue. We return to this in the single-source-of-truth step below.

How much of your data is actually report-ready?

A useful headline metric for leadership is the share of records in your core reporting tables that pass every automated quality rule — sometimes called the clean record rate. It is deliberately strict: a customer record with a valid name and email but a missing region still fails, because it cannot be used in a regional report.

In a typical first assessment of a UK SME’s customer master data, we see a clean record rate somewhere around 55–65%. The figure below shows a representative midpoint.

60%
Typical clean record rate for SME customer master data at first assessment

After a focused cleansing and validation project, most clients reach 90–95% on the same rules within three months, and then hold it there because new records are validated at the point of entry. The remaining 5–10% is usually legitimate edge cases — overseas customers without UK postcodes, sole traders without Companies House numbers — which should be handled with explicit exception rules rather than forced into the standard pattern.

How to fix data quality for reporting — a step-by-step timeline

The following timeline is the sequence we follow for a first-phase data quality project covering two to four core datasets (typically customers, products, orders and invoices). It assumes a business with a reporting database or warehouse already in place, or being built alongside — if you are still choosing that architecture, read our comparison of data warehouses versus reporting databases first.

Week 1 — Pick the decisions, not the data
Agree with leadership which three to five reports matter most and which decisions they drive. Trace every figure on those reports back to source tables and fields. This bounds the project to the data that actually affects decisions.
Week 2 — Profile the source data
Run profiling queries on each in-scope table: null rates per column, distinct value counts, min/max dates, pattern frequency for postcodes and phone numbers, and candidate duplicates using fuzzy matching on names and addresses. Record a baseline clean record rate.
Week 2–3 — Assign owners and write the rules
Name a business owner for each dataset (sales ops for customers, purchasing for products, finance for invoices). With them, write plain-English quality rules: “every active customer has a region from the approved list”, “every SKU has a cost price updated in the last 12 months”.
Week 3–4 — Design the master record and shared IDs
Decide which system is the system of record for each entity and introduce a shared identifier (for example, the CRM account ID stored in Xero’s contact reference). Build crosswalk tables to link legacy records that lack the shared ID.
Week 4–6 — Cleanse and deduplicate
Bulk-correct standardisable values (county names, sector codes, title case), validate postcodes against Royal Mail PAF or an equivalent address service, check company numbers against the Companies House API, and work through duplicate clusters with the data owner deciding the surviving record.
Week 5–7 — Prevent at the point of entry
Replace free-text fields with picklists, make reporting-critical fields mandatory, add format validation to web forms and CRM layouts, and switch manual re-keying to proper integrations wherever the volume justifies it.
Week 6–8 — Automate quality tests on load
Implement the agreed rules as automated tests that run every time data lands in the reporting database — SQL assertions, dbt tests or scheduled stored procedures. Failures go to a quarantine table and an alert to the dataset owner, not silently into the dashboard.
Week 8–10 — Publish a data quality scorecard
Add a small data quality dashboard alongside the business reports showing clean record rate, open exceptions and freshness per dataset. Review it monthly with owners so quality becomes a managed KPI rather than a one-off project.

Step-by-step: the data quality checks to automate first

Automated data quality checks are the heart of a durable fix. Without them, a cleansing project is a spring clean: satisfying for a month, then the dust returns. With them, every load into the reporting database is tested and failures are caught before they reach a dashboard. You do not need an enterprise platform to start. Most of the checks below are a few lines of SQL each, and they can run in SQL Server Agent, Azure Data Factory, a scheduled Cloudflare Worker, dbt or any orchestration tool you already use.

1. Freshness checks

Test that each source has delivered data within its expected window — for example, “the latest order in the reporting table is less than two hours old during trading hours”. Freshness failures are the earliest signal that an integration has broken, and they are the check that most often prevents the “why is yesterday missing?” conversation. Our guide to real-time versus batch reporting explains how to set sensible freshness targets for each dataset.

2. Row count and volume checks

Compare today’s load against a rolling average. A sudden 60% drop in orders on a Tuesday is either a genuine business event or, far more often, a partial load. Flag anything outside an agreed band (commonly ±30–50% of the trailing four-week same-weekday average) for a human to look at.

3. Not-null and mandatory field checks

For every field a key report slices by — region, sector, account manager, product family, cost centre — assert that it is populated on active records. Express it as a rate rather than a hard pass/fail if historical data is incomplete: “at least 98% of orders created this month have a cost centre”.

4. Referential integrity checks

Every order should reference a customer that exists; every invoice line should reference a valid product and nominal code. Orphaned records are a classic cause of totals that do not reconcile between a detailed report and a summary, because one joins and the other does not.

5. Uniqueness and duplicate checks

Assert that primary business keys are unique (one row per invoice number, one active record per customer ID) and run a weekly fuzzy-match job on names, emails and postcodes to surface probable duplicates for review. Exact-key checks are cheap and catch load errors; fuzzy matching catches the human-entry duplicates.

6. Validity and format checks

Test UK postcodes against the standard format, VAT numbers against the GB nine- or twelve-digit pattern, email addresses for basic structure, and dates for plausible ranges (no orders dated before the company was incorporated, no delivery dates three years in the future).

7. Business rule and reconciliation checks

Finally, test the rules that only your business knows: gross margin on a sales line should sit between −10% and 80%; the sum of invoice lines must equal the invoice total; the month-end revenue in the reporting database must match the figure in your accounting system to the penny. Reconciliation against the ledger is the single most powerful trust-builder for a finance audience.

Data quality maturity — a readiness gauge

To help clients see where they stand, we score data quality maturity out of 100 across five areas: ownership (are datasets owned?), rules (are quality rules written down?), prevention (is bad data stopped at entry?), detection (are automated tests running?), and remediation (is there a process to fix failures?). Each area is worth 20 points.

The typical UK SME we assess for the first time lands in the low-to-mid forties: some prevention through accounting software, a little informal ownership, but almost no automated detection and no remediation process.

42/100
Typical UK SME data quality maturity at first assessment (Cloudswitched benchmark)

A score above 70 is a realistic target for a 6–10 week first phase. That usually means every core dataset has a named owner, written rules, validation at entry for the most error-prone fields, automated tests on load and a monthly review. Scores above 85 tend to require cultural embedding — data quality objectives in role descriptions, quality metrics reviewed at board level — which takes longer and depends on leadership commitment more than technology.

Data quality project costs — what UK businesses typically pay

Costs vary with the number of source systems, the volume and age of the data, and how much remediation is done by your own staff versus an external partner. The ranges below reflect typical UK pricing in 2026 for SME and lower mid-market organisations.

ScopeTypical workIndicative costDuration
Data quality assessmentProfiling of 2–4 core datasets, baseline scorecard, prioritised issue log£2,500–£6,0001–2 weeks
Cleansing and deduplicationStandardisation, address and company validation, duplicate resolution with owners£4,000–£15,0002–4 weeks
Prevention at sourcePicklists, mandatory fields, form validation, replacing manual re-keying with integrations£3,000–£12,0002–3 weeks
Automated quality tests and alertingSQL/dbt tests on load, quarantine tables, owner alerts, data quality dashboard£4,000–£14,0002–3 weeks
Ongoing data quality supportMonthly review, rule maintenance, new-source onboarding£400–£1,800 per monthOngoing

A full first phase across all four lines typically lands between £8,000 and £35,000. Running costs for the tooling itself are usually modest: address validation services charge per lookup (often a few pence each, with bulk pricing for initial cleansing), the Companies House public data API is free to use with a registered key, and open-source testing frameworks such as dbt Core and Great Expectations carry no licence cost. The largest cost driver is almost always the human time to resolve ambiguous duplicates and agree definitions — which is also the part that delivers the most lasting value.

For context on the reporting layer itself, our custom reporting dashboard cost breakdown covers what UK businesses pay for the dashboards that sit on top of a clean data foundation.

Data quality benchmarks and KPIs to track

Once the automated tests are running, they produce a stream of measurements you can track over time. The benchmarks below show the pass rates we typically see on each type of check after a first-phase project has bedded in for three months. Use them as a starting point for your own targets rather than as fixed standards — a business with a heavy overseas customer base, for example, will naturally score lower on UK postcode validity.

Typical rule pass rates three months after a first-phase project

Freshness within SLA
98%
Referential integrity
99%
Mandatory reporting fields populated
96%
UK postcode validity
94%
Unique active customer records
97%
Product cost prices current
91%
Cost centre tagging on transactions
93%
Ledger reconciliation at month end
100%
Exceptions resolved within 5 working days
85%

Two KPIs deserve a place on the leadership dashboard itself: the overall clean record rate for each core dataset, and the exception resolution time. The first tells leaders how far they can trust the breakdowns in front of them; the second tells them whether the organisation is actually maintaining quality or just detecting its decline. Ledger reconciliation should be treated as binary — if the reporting database does not match the accounts at month end, the finance pack should not be published until it does.

If you are designing the wider KPI set, our KPI dashboard reporting guide for UK SMEs covers how to choose metrics that drive decisions rather than decorate a screen.

Common data quality mistakes that undo clean-up projects

Most data quality projects deliver a visible improvement. Far fewer are still delivering it a year later. The difference almost always comes down to a handful of avoidable mistakes:

  • Treating cleansing as a one-off project. A bulk clean without prevention at source and automated detection simply resets the clock. Within six to twelve months, new errors have accumulated at the same rate as before.
  • Fixing data in the reporting layer only. Corrections made in Power Query or SQL views never reach the CRM or finance system, so staff using those systems still see the wrong data — and the UK GDPR accuracy duty is not met.
  • No named business owner. When data quality is “IT’s job”, nobody with the authority to decide which of two duplicate records is correct is ever in the room. Owners must sit in the business function that creates the data.
  • Deleting duplicates instead of merging them. Deleting the “wrong” customer record orphans its orders, invoices and email history. Always merge into a surviving record and keep a crosswalk of retired IDs.
  • Alerting everyone about everything. A daily email listing 400 failed rules is ignored by week two. Route each failure to the owner of that dataset, group by rule, and set severity so only material breaks interrupt anyone.
  • Writing rules nobody agreed. A technically perfect rule that the sales team considers wrong will be worked around. Rules need business sign-off, in plain English, before they become code.
  • Ignoring metric definitions. Clean data with two competing definitions of “active customer” still produces two different numbers. Agree and document definitions in a shared glossary alongside the quality rules.
  • Silently dropping failed records. Filtering bad rows out of reports without telling anyone makes totals look tidy and quietly wrong. Quarantine them visibly and show the exception count on the dashboard.
Watch out

Be careful with bulk cleansing of personal data. Matching and merging customer records is processing under UK GDPR, so record it in your Record of Processing Activities, make sure third-party address or enrichment services have a suitable data processing agreement, and keep an audit trail of merges in case a data subject later asks why their record changed. Restricting who can see the merged data matters too — our guide to row-level security in database reporting covers the access side.

Data quality checklist — the 12-point essentials

Use this checklist to review your own reporting estate. If you cannot tick at least eight of the twelve, the trust problems in your dashboards are almost certainly data quality problems rather than reporting tool problems.

  1. The three to five reports that drive leadership decisions are identified, and every figure on them can be traced to a named source table and field.
  2. Each core dataset (customers, suppliers, products, orders, invoices, staff) has a named business owner with authority to approve corrections.
  3. There is one agreed system of record for each entity, and everyone knows which it is.
  4. A shared identifier links the same customer, supplier or product across CRM, finance, e-commerce and operational systems.
  5. Reporting-critical fields such as region, sector, product family and cost centre use picklists rather than free text.
  6. Web forms and CRM layouts validate UK postcodes, emails, phone numbers and VAT numbers at the point of entry.
  7. Data quality rules are written in plain English, signed off by the dataset owner, and stored somewhere the business can read them.
  8. Automated tests for freshness, volume, completeness, uniqueness, validity and referential integrity run on every load into the reporting database.
  9. Failed records go to a visible quarantine with an alert to the relevant owner, rather than being silently dropped or silently included.
  10. The reporting database reconciles to the accounting ledger at each month end before the management pack is issued.
  11. A data quality scorecard showing clean record rate and open exceptions is reviewed monthly with owners.
  12. Bulk cleansing and merging of personal data is documented in your Record of Processing Activities with a retained audit trail.
Note

Items 1–3 are organisational rather than technical and cost almost nothing beyond a few hours of leadership time. They are also the items most often skipped. Start there before commissioning any cleansing or tooling work — the technical steps go faster and stay fixed once ownership is clear. A virtual CIO is often well placed to drive this; see our guide to what the first 90 days with a virtual CIO should look like.

Mini case study: a Midlands distributor rebuilds trust in its sales reporting

A 64-person building-products distributor in the West Midlands came to us with a familiar complaint: the monthly sales pack took the finance team three working days to prepare, and the managing director no longer believed the regional breakdowns. Sales were recorded in a CRM, invoiced in Sage 200, and the trade counter ran a separate point-of-sale system. Each held its own customer list. A profiling exercise found just over 11,000 active customer records across the three systems representing around 7,400 real customers, with 38% of CRM accounts missing a sales region and four different spellings for most counties.

Over nine weeks we agreed Sage 200 as the system of record for customer accounts, introduced the Sage account code as the shared identifier in the CRM and the point-of-sale system, and built crosswalk tables for historical records. Address data was validated against a PAF-based service, company numbers checked against Companies House, and duplicate clusters were worked through with the sales operations manager in four two-hour sessions. Region and sector were converted to mandatory picklists. Finally, 46 automated tests were added to the nightly load into the Azure SQL reporting database, with failures routed to the owning manager each morning.

Three months later the clean record rate on customer data stood at 94%, the regional figures reconciled to the ledger, and month-end pack preparation had dropped from three days to under half a day because the reconciliation step was automated. The more significant change, according to the finance director, was that board discussions moved from questioning the numbers to acting on them.

“We had spent two years blaming the dashboard. It turned out the dashboard was perfectly accurate — it was faithfully reporting three different versions of our customer list. Once we fixed that and put the checks in, nobody has asked me ‘is this number right?’ in a board meeting since.” — Finance Director, West Midlands building-products distributor

At a glance: data quality for business reporting

TopicKey fact
Six dimensionsCompleteness, accuracy, consistency, timeliness, uniqueness, validity
Where problems clusterMaster and reference data — customers, suppliers, products — not transactional totals
Biggest root causeFree-text fields where a validated picklist belongs
Typical starting clean record rateAround 55–65% for SME customer master data
Realistic post-project target90–95% clean record rate within three months
Typical maturity at first assessmentLow-to-mid 40s out of 100
First-phase project cost£8,000–£35,000 for a UK SME
First-phase duration6–10 weeks
Checks to automate firstFreshness, volume, not-null, referential integrity, uniqueness, validity, reconciliation
UK validation sourcesRoyal Mail PAF-based address services, Companies House API, HMRC VAT number checks
Legal anchorUK GDPR Article 5(1)(d) accuracy principle; ICO guidance on rectification
OwnershipBusiness owner per dataset in the function that creates the data
Non-negotiable finance checkReporting database reconciles to the ledger before the pack is issued
Leadership KPIsClean record rate per dataset and exception resolution time

How Cloudswitched helps with data quality and database reporting

Cloudswitched’s database reporting team works with UK businesses to profile their source systems, agree ownership and rules with the people who create the data, cleanse and deduplicate core datasets, and build automated data quality checks into the pipeline that feeds their dashboards. We work across SQL Server, Azure SQL, MySQL, PostgreSQL and cloud data platforms, and integrate with common UK business systems including Xero, Sage, Microsoft Dynamics, HubSpot and Shopify. The aim is a reporting estate your leadership team can act on without a weekly reconciliation exercise.

Want reports your board can trust?

Talk to our database reporting team about a data quality assessment of the datasets behind your most important reports.

Talk to us about Database Reporting

Frequently Asked Questions

What is data quality in business reporting?

Data quality in business reporting is the degree to which the records behind your reports are complete, accurate, consistent, timely, unique and valid enough for the decisions you make from them. A report can be technically correct — summing exactly what is in the database — and still mislead if the underlying data has duplicates, missing fields or conflicting values between systems. Good data quality means the figures in your dashboards reflect what actually happened in the business, and that breakdowns by region, product or customer reconcile with your accounts.

What are the most important data quality checks to automate?

Start with seven checks: freshness (has each source delivered on time?), volume (is today’s load within an expected range?), completeness of reporting-critical fields, referential integrity between related tables, uniqueness of business keys, validity of formats such as UK postcodes and VAT numbers, and reconciliation of totals against your accounting ledger. Together these catch the large majority of issues that undermine trust in management reporting, and most can be written as simple SQL assertions running on each load.

How much does a data quality project cost for a UK SME?

A first-phase data quality project for a UK SME typically costs between £8,000 and £35,000, covering assessment, cleansing and deduplication, prevention at source and automated testing. A standalone data quality assessment usually costs £2,500–£6,000. Ongoing support for rule maintenance and monthly reviews runs from roughly £400 to £1,800 a month. The main cost driver is the number of source systems and the human time needed to resolve ambiguous duplicates with data owners.

Do we need master data management software?

Most SMEs do not need a dedicated master data management (MDM) platform. Lightweight master data management — agreeing a system of record for each entity, introducing a shared identifier across systems, maintaining crosswalk tables for legacy records and validating at entry — delivers most of the benefit for a fraction of the cost. Dedicated MDM software becomes worthwhile when you have many source systems creating the same entity, complex hierarchies, or regulatory requirements for formal data stewardship workflows.

How long does data cleansing take?

Data cleansing for two to four core datasets in an SME typically takes two to four weeks within a six-to-ten-week overall project. Standardising values and validating addresses can be largely automated and completes quickly. Resolving duplicate clusters takes longer because a business owner must decide which record survives and how to merge history. The age and size of your data matter: a business with fifteen years of migrated customer records will need more review time than one with a recent, single CRM.

Is inaccurate customer data a UK GDPR issue?

Yes. UK GDPR Article 5(1)(d) requires personal data to be accurate and, where necessary, kept up to date, and the ICO expects organisations to take reasonable steps to correct or erase inaccurate data. Duplicate and stale customer records can lead to sending communications to the wrong person, ignoring an opt-out recorded on a different duplicate, or making decisions on outdated information. A data quality programme that validates at entry and merges duplicates properly supports your compliance with the accuracy principle.

Should we fix data in Power BI or in the source systems?

Fix data in the source systems wherever possible, and use the reporting layer only for genuine transformations or temporary patches. Corrections made in Power Query or SQL views never reach the people using the CRM or finance system, they hide business rules inside queries that few people can read, and they break silently when source data changes. A short-term patch in Power BI is reasonable to get a credible board figure quickly, but it should be logged and replaced with a source fix and an automated check.

Who should own data quality in a small business?

Data quality should be owned by the business function that creates the data, not by IT. Sales operations typically own customer data, purchasing or product management own product data, finance owns invoices and nominal codes, and HR owns staff records. IT or an external partner builds the tooling, tests and alerts, but the dataset owner decides what the rules are and approves corrections. A senior sponsor — often the finance director or a virtual CIO — keeps the programme on the leadership agenda.

What is a single source of truth in reporting?

A single source of truth means that for every business entity and every key metric there is one agreed authoritative version. For entities, that is a designated system of record plus shared identifiers that link the same customer or product across all other systems. For metrics, it is a documented definition — for example, exactly when an order counts as revenue — applied consistently in one governed reporting layer. Without both parts, different reports will continue to produce different answers to the same question.

How do we keep data quality high after the clean-up?

Sustaining data quality depends on three things: prevention at the point of entry through picklists, mandatory fields and validation; automated tests on every load that route failures to the right owner; and a monthly review of a data quality scorecard showing clean record rate and open exceptions. Treat the clean record rate as a managed KPI with a target, and include data quality in onboarding for anyone who creates records. Without these, a cleansed dataset typically degrades noticeably within six to twelve months.

Can AI help with data quality?

AI tools can help with specific data quality tasks, particularly fuzzy matching of duplicate names and addresses, classifying free-text values into standard categories, and extracting structured fields from documents. They work best as assistants that propose matches or classifications for a human owner to approve, with confidence thresholds and an audit trail. They do not replace the fundamentals of ownership, agreed rules and validation at entry, and any AI processing of personal data still needs to meet UK GDPR requirements.

Build your reporting on data you can trust

Cloudswitched’s database reporting service covers data quality assessment, cleansing, automated checks and the dashboards that sit on top — scoped around the decisions your business actually makes.

Explore Database Reporting
Tags:Database Reporting
Cloudswitched

London-based managed IT services provider offering support, cloud solutions and cybersecurity for SMEs.

Cloudswitched Service

Database Reporting & Analytics

Custom dashboards, automated reports and powerful data search tools

More about this service
CloudswitchedDatabase Reporting & Analytics
020 3004 3450
Insights & guides

Latest from our blog

Data Quality for Business Reporting - A UK Guide to Fixing Bad Data Before It Reaches Your Dashboards
Database Reporting16 Nov 2026

Data Quality for Business Reporting

Every UK business that reports from a database eventually hits the same wall: the dashboard looks polished, but nobody trusts the numbers. Data quality for...

Read article
RAG for UK Businesses - How to Build an AI Knowledge Assistant on Your Own Documents
AI15 Nov 2026

RAG for UK Businesses

Retrieval-augmented generation (RAG) is the most practical way for a UK business to build an AI knowledge assistant that answers questions from its own...

Read article
Google Consent Mode v2 and Enhanced Conversions - A UK How-To Guide for Accurate Google Ads Tracking
Google Ads & PPC14 Nov 2026

Google Consent Mode v2 and Enhanced Conversions

Google Consent Mode v2 and enhanced conversions decide whether your Google Ads account can still see what your advertising is actually doing. For a UK business...

Read article
Cyber Essentials and Unsupported Software - The UK Guide to End-of-Life Devices in 2026
Cyber Security6 Nov 2026

Cyber Essentials and Unsupported Software

If you are preparing for certification this year, the single question most likely to stop you is not about firewalls or passwords it is about Cyber Essentials...

Read article
Structured Data and Schema Markup: A UK Business Guide to Winning Rich Results in Google Search in 2026
SEO5 Nov 2026

Structured Data and Schema Markup

Schema markup is structured code added to web pages that describes their content in a standard vocabulary search engines understand: this page is a product...

Read article
Website Redesign Planning: A UK Business Guide to Avoiding the Mistakes That Tank SEO Rankings During a Rebuild in 2026
Web Development4 Nov 2026

Website Redesign Planning

Treating website redesign SEO as an afterthought is the most reliable way for a UK business to lose the organic traffic it spent years building. The pattern is...

Read article
Virtual CIO Onboarding: A UK Business Guide to What the First 90 Days Should Actually Look Like in 2026
Virtual CIO3 Nov 2026

Virtual CIO Onboarding

Good virtual CIO onboarding looks, from the outside, surprisingly unglamorous. Many UK businesses sign a virtual CIO contract expecting a strategy presentation...

Read article
Network Capacity Planning: A UK Business Guide to Forecasting Growth Before Your Network Becomes the Bottleneck in 2026
Network Admin1 Nov 2026

Network Capacity Planning

Network capacity planning is the practice of knowing how much headroom each part of your network has, forecasting when growth will use it up, and acting before...

Read article
Office Downsizing and Consolidation: A UK Business Guide to Moving to a Smaller Office Without Losing IT Capability in 2026
IT Office Moves31 Oct 2026

Office Downsizing and Consolidation

Planning office downsizing IT is a different exercise from planning a normal office move, and treating it as a smaller version of one is where most problems...

Read article
Microsoft 365 Copilot Adoption: A UK Business Guide to Why Most Deployments Fail to Change How People Work in 2026
Microsoft 365 Copilot30 Oct 2026

Microsoft 365 Copilot Adoption

A workable Copilot adoption strategy starts from an uncomfortable fact: buying Microsoft 365 Copilot licences does not change how anyone works. The pattern in...

Read article
Technology stack

Built on technology you can trust.

We design, run and secure your IT on proven, industry-leading platforms, including Microsoft 365 and Azure, Cisco Meraki, Cloudflare, AWS, BitDefender and SolarWinds, so your business stays secure, scalable and reliable.

  • SolarWinds
  • Cloudflare
  • BitDefender
  • AWS
  • Hono
  • SolarWinds
  • Cloudflare
  • BitDefender
  • AWS
  • Hono
  • Opus
  • Office 365
  • Microsoft
  • Cisco Meraki
  • Microsoft Azure
  • SolarWinds
  • Opus
  • Office 365
  • Microsoft
  • Cisco Meraki
  • Microsoft Azure
  • SolarWinds

Enquiry Received!

Thank you for getting in touch. A member of our team will review your enquiry and get back to you within 24 hours.