What we do · Fix my data

Bad data in, bad decisions out.

Your business runs on data about other businesses, where they operate, and the people who run them. We clean it, complete it and keep it current, so you can trust every decision you make on it.

Check my data free Talk to us
Your data Gmaven Your data + ours
Your dataMessy, duplicated, incomplete, going stale.
→
GmavenWe clean, enrich and maintain it.
→
Your data + oursClean, complete, current, trustworthy.
Why this matters now

Junk in, junk out.

Every decision your business makes, and increasingly every answer an AI assistant gives your people, is only as good as the data underneath it. Put junk in, and you get junk out: the wrong customers targeted, bad calls made, opportunities missed.

Fixing enterprise data properly is harder than it looks, and most once-off, in-house attempts fall short. They do the easy, surface-level part, run out of road on the hard part, and leave you with data that looks clean but still lies to you.

The benefit

Hand it to specialists.

We are enterprise data specialists, focused on data about businesses, their locations (commercial properties), and their decision-makers. We have already made the investment, and the expensive mistakes, over two decades. So you get data you can trust without hiring a data team, buying tooling, or climbing a learning curve. Your problem, solved, and off your plate.

The part others skip

There are two kinds of data fix. The difference is everything.

Vertical: makes it look right

Vertical fixes work down a column: putting every phone number in the same format, removing duplicates, checking a number is real. This scales, it is the easy part, and it makes your data look right.

Horizontal: makes it actually right

Horizontal fixes work across a single record: does this phone number actually belong to this person, or is it a dummy, or their spouse's? That takes judgement, the other clues in the record, and reference data. It is slow, it is where the real value sits, and it is the part most providers and in-house projects skip.

We do both. Reference data is the trusted master set you check each record against to know the right answer, and in South Africa we hold the deepest set there is for commercial property and the businesses and people attached to it. That is what lets us do the horizontal fixes others cannot. And knowing which record needs which only comes with experience.

Speed View
YOUR DATA, AS IT ARRIVES Eight records, as they might land in any customer file. Grey means your data, untouched. STEP 1 · VERTICAL: ANALYSE Down each column in turn, every value is checked against that column's rules. Nothing reads across a row yet. 19 faults flagged, 14 values pass. Five of eight company names verified on the register. STEP 2 · VERTICAL: FIX The same columns again. What a column alone can fix is fixed, in cyan: Gmaven's work. The rest is queued. 3 fixed, 1 cleared, 15 queued to investigate. This is where most fixes stop. Two errors have not even been seen. STEP 3 · HORIZONTAL: ANALYSE Now across one record at a time. Red says what is wrong; blue says what it should be, from our reference data. 16 faults diagnosed with the right value for each, and 2 errors found that no column check could see. STEP 4 · HORIZONTAL: FIX Record by record, the right values are written in, the duplicate pair becomes one business, and each record is scored. 8 corrected, 6 completed, 1 pair merged. 8 records in, 7 out, every one scored and none left needing review. BEFORE · YOUR FILE, UNTOUCHED This is the data exactly as it arrived. Every fix, flag and score of ours is hidden. Press After, top right, to bring our work back. Company Registration number Contact person The business's publicly listed representative. See the note below the figure. i Industry Acme Trading 123 2014-123456-07 T Example Business Sample Stationery 1998/012345/23 A. Retail Sample Stationers 1998 012345 23 Ayesha Sample Demo Logistics CC Lindiwe Demo Transportation Testco Projects 482913 Sipho Test Construction and Materials Placeholder Pl 2011/765432/07 Piet Voorbeeld Manufacturing Faded Fabrics 1111/111111/11 Nomsa Invented Munafact Testline Traders 876543 Zanele Mock Transporting correct fault the column pass can fix fault that needs the whole record VERTICAL CHECKS One standard format standardisation Valid pattern value validity No duplicates uniqueness On the register existence On the industry list conformance Complete, no blanks or initials completeness Not checked here: does each value belong to this record? (business-rule and derived validity) Runs on: Gmaven code + algorithms, with register lookups (reference data) no match format initial only no match duplicate initial only no match no match format · duplicate missing missing incomplete no match no match filler no match incomplete near match 2014/123456/07 1998/012345/23 1111/111111/11 Transportation correct, unchanged fixed or cleared here still wrong: needs the whole record VERTICAL FIX: OUTCOMES Fixed: 3 two numbers standardised, one industry name matched to the list Cleared: 1 filler removed, now treated as missing, not as a real value Investigate: 15 queued for the horizontal pass A column alone cannot resolve a no match, a gap or a duplicate. That needs the whole record, and reference data to check it against. Runs on: Gmaven code + algorithms, with register lookups (reference data) investigate fixed investigate investigate investigate investigate investigate investigate fixed · investigate investigate investigate investigate investigate investigate cleared, now missing investigate investigate fixed correct what is wrong what it should be, from our reference data an error no column check could see HORIZONTAL CHECKS Value belongs to the record business-rule validity Fields agree with each other derived validity Record makes sense as a whole record-level reasonableness Same business as another row entity resolution GMAVEN REFERENCE DATA the register entry behind row 4: Demo Logistics CC 2004/055123/23 Lindiwe Dube Runs on: Gmaven algorithms, the reference-data mesh, human specialists, and code and tooling. Slow, expensive, and the part others skip. wrong name · Acme Traders initials · Thabo Example not on list · Technology Hardware and Equipment duplicate · merge row 3 initial · Ayesha Sample not on list · Retailers not the registered name · Sample Stationery duplicate · merge row 2 missing · Retailers missing · 2004/055123/23 surname · Lindiwe Dube partial · 2016/482913/07 name cut short · Placeholder Plastics surname · Piet van Beeld not on list · Industrial Materials missing · 2007/334455/07 not on list · Construction and Materials partial · 2019/876543/07 Every column check passed "Lindiwe Demo". It is still wrong. The register's public representative for Demo Logistics CC is Lindiwe Dube. Someone typed the company name into the surname field. Only reading the whole record, against reference data, can catch that. value we corrected or completed merged away into another record Trust HORIZONTAL FIX: OUTCOMES Corrected: 8 wrong values replaced with right ones Completed: 6 gaps and part-values filled in Merged: 1 pair two rows, one business, best value kept Scored: 7 records THE TRUST MARK Verified: confirmed on the register High: one judgement involved Review: we flag it, never guess None left needing review in this example. Runs on: Gmaven algorithms, the reference-data mesh, human specialists, and code and tooling. Slow, expensive, and the part others skip. Acme Traders Thabo Example Technology Hardware and Equipment Retailers Ayesha Sample kept merged into the record above: one business, not two 2004/055123/23 Lindiwe Dube 2016/482913/07 Placeholder Plastics Piet van Beeld Industrial Materials 2007/334455/07 Construction and Materials 2019/876543/07 Vertical makes data look right. Horizontal makes it actually right. We do both. 21 faults in eight records. 3 could be fixed by the column pass alone. 18 needed the whole record, read against reference data. That 18 is the part most providers, and most in-house projects, skip. Contact person: the business's publicly listed representative, taken only from public, lawful records. Every value in this figure is invented.
Eight invented records carrying the faults real files carry, worked through four steps. One rule throughout: a tick means correct, red says what is wrong, blue says what it should be. The column pass finds nineteen faults and can fix three. Reading each record across, against Gmaven's reference data, finds two more that no column check could see, then corrects eight values, completes six, merges a duplicate pair into one business, and scores every surviving record. Use View to compare the original file with our work.
Why we can do it

Our secret sauce.

Five things, built over years of iteration.

Algorithms

Our own scale-friendly data flows and processes, founded on years of deep intellectual property.

Code and tooling

Mature, automated data processing, with technologies applied in ways others don't.

Reference-data mesh

Our own data woven together with public and private sources into one trusted layer.

People

Skilled specialists working to a clearly defined process, alongside the technology.

Horizontal interventions

We go beyond the surface-level work most others stop at.

What we actually do

Clean. Enrich. Maintain.

Clean

We find what is wrong and fix it. First we detect the errors and do the surface fixes: deduplicate, standardise and normalise, and correct obvious mistakes. Then we resolve what is left by replacing it with the correct data, which is where cleaning flows into enrichment.

Enrich

We make your records complete. We replace wrong data with correct data, fill the gaps in fields you already have, and add the new fields you always wanted.

Maintain

We keep it right going forward. We validate, clean and enrich each new record as it arrives, then keep the rest fresh on a schedule, so your data stays trustworthy instead of quietly going stale.

Boundaries

What we don't do.

We don't sell data software, and we don't replace your business-intelligence or systems teams. We clean, enrich and maintain your data, and that is all. If you would rather plug our data straight into your own systems, see Machine access.

Compliance, built in

Contacts, done compliantly.

When we work on the people who make the decisions, we use only public, lawful sources, and we tell you which outreach channels you may use without consent and which you may not. We are your data provider, not your marketer: the outreach stays yours, and we make sure the data under it is sound. This is built into how we work, not bolted on.

This describes how we build. It is not legal advice: confirm your own outreach design with your counsel.

Proof

We have done the biggest jobs in the market.

For about two decades, on some of the largest data jobs in the market:

We also hold South Africa's largest commercial-property database, more than 64,000 properties with the richest metadata (names, addresses, lettable area, category, extent, value, solar), and we run the country's largest business-to-business commercial-property marketplace. Our core data-processing methods were built on work recognised by an Industrial Development Corporation innovation grant. We are an RICS Technology Partner and sit on the SAPOA PropTech Committee.

Scoping and price

You pay for results.

Every data set is different, so we price each job to its size, its condition, and the fields you need. No off-the-shelf package, and no charge for work that doesn't land. We scope it with you first.

Start with a free check.

Send us up to 10 records. We will show you what is wrong and what we would fix, at no cost. It is a quick read to prove the point, not the full fix.

Check my data free → See what else we do