Data quality problems are usually discovered through their symptoms: a customer receives two invoices, a dashboard disagrees with the ledger, a campaign goes to the wrong segment, or an AI assistant answers from an outdated product record. The natural response is to buy a tool. The more useful response is to define what "good" means for the data that drives those outcomes, then decide whether a tool, a service or a process change is the right purchase.
This buying guide is written for an SME that does not have a data team. It covers the requirements to settle first, a scorecard for comparing vendors, the pricing questions that expose hidden cost, how to run a pilot, and what to agree before you commit.
Start with the decisions the data supports
Data quality has no absolute standard. A customer's mobile number is critical for delivery notifications and less critical for a market-research segment. A product weight matters for shipping and not for a newsletter. Before comparing vendors, list the decisions and processes that depend on data:
- Billing and reconciliation.
- Order fulfilment and delivery.
- Inventory replenishment.
- Customer segmentation and campaigns.
- Regulatory or donor reporting.
- Management dashboards and forecasts.
For each one, name the data domain, the consequence of a defect and the person who currently fixes it by hand. That list becomes your requirements document, and it is far more useful than a generic feature checklist.
Define quality in six dimensions
- Completeness: required fields are present and usable.
- Accuracy: values match the real world or the authoritative source.
- Timeliness: data arrives early enough to act on it.
- Consistency: the same fact agrees across systems.
- Uniqueness: one customer, product or supplier is one record.
- Validity: values follow the agreed format, range and reference list.
Pick two or three dimensions per domain. Trying to measure all six everywhere produces a dashboard nobody uses.
Build a baseline before you shop
Profile one domain, usually customers or products, and record the defect rate, the manual correction effort and the business impact. A simple sample is enough to start: take a few hundred records, check the fields that matter, and classify the failures. The baseline does two things. It tells you whether a vendor's promised improvement is real, and it gives the pilot an acceptance test.
Do not clean the data before the baseline. You will lose the evidence of what the process actually produces.
Know which category you are buying
- Profiling and monitoring: measures quality, detects drift and alerts owners.
- Cleansing and standardisation: corrects addresses, phone numbers, names and codes.
- Matching and deduplication: finds the same entity across records and systems.
- Master data management: creates a governed golden record for a domain.
- Data observability: watches pipelines for freshness, volume and schema changes.
- Governance and catalogue: documents ownership, definitions and lineage.
Many SMEs need a combination of profiling, cleansing and a lightweight stewardship process. Master data management is justified when several systems disagree about the same customers or products and the disagreement has a direct financial cost.
A weighted vendor scorecard
Score each vendor from 1 to 5 on the criteria below, multiply by the weight, and compare the totals. Adjust the weights to your context, but decide them before the demonstrations.
- Connectors and data sources, 15 percent: can it read your ERP, ecommerce platform, spreadsheets and databases without custom scripts?
- Profiling and rule building, 15 percent: can a non-specialist define and test a rule?
- Stewardship workflow, 20 percent: who receives a defect, how is it fixed, and is the fix recorded?
- Lineage and audit, 10 percent: can you show where a value came from and who changed it?
- Security and deployment, 15 percent: access control, encryption, data residency and whether the tool can run in your environment.
- Support and local context, 10 percent: Bangla support, timezone, implementation partner and response terms.
- Pricing predictability, 10 percent: what happens to cost when records, users or sources grow?
- Exit and portability, 5 percent: can you export rules, mappings and history?
Pricing questions that expose the real cost
- Is pricing based on records, sources, users, domains or environments?
- What counts as a record, and are duplicates or historical versions counted?
- Are implementation, connector setup and training included?
- What does a re-scan or a full refresh cost?
- Is there a charge for storing cleaned data or audit history?
- How does the price change at renewal, and what notice is required?
- What is excluded from support, and what is billed hourly?
- What does exit cost, including export and deletion?
Ask for a three-year total cost under two volume scenarios. A low entry price with per-record growth can overtake a higher fixed subscription as the business grows.
Design the pilot
Run a four to six week pilot on one domain and one decision. Define the baseline, the target, the acceptance test and the owner before the pilot begins. Include the people who currently fix the data by hand, because they know where the exceptions live.
At the end, review three things: whether the quality measure improved, whether the stewardship workflow was actually used, and whether the team could operate the tool without the vendor in the room. A pilot that only proves the software can clean a sample has not proved much.
Agree the exit before you sign
- Export formats for records, rules, mappings and audit history.
- Ownership of cleaned data and derived attributes.
- Deletion timeline and a certificate of deletion.
- Transition assistance and its cost.
- Continuity if the vendor is acquired or discontinues the product.
Where this fits in our work
If you are preparing data for reporting or AI and need the foundation fixed first, start with Operations. The pricing page explains how we scope an assessment. For the wider programme, read the AI-ready data foundations guide.
Sources
- UK Government, The Government Data Quality Framework, accessed 17 November 2026.
- DAMA International, data management body of knowledge, accessed 17 November 2026.
- International Monetary Fund, Data Quality Assessment Framework, accessed 17 November 2026.
Responses (0)
No comments yet. Add the first one below.
Leave a response
Your email address will not be published. Required fields are marked *.