Startup deal flow: prevent duplicate records and entity errors

Know what each record means

Start by naming the thing in front of you. A lead may be a founder note. It may be a product name. It may be an old form. It may not name a legal firm. Give each type a clear tag. Then a reviewer can see what is being linked.

Use one main file for a firm. Keep all lead notes that point to it. Link to a source. Do not copy a fact with no source. This lets the team check where a fact came from.

A close name calls for a check. It is not proof. A brand name can differ from a legal name. Two firms can use close names. Do not turn a fast search into a join.

• Lead file: where and when the team found it.
• Firm file: the firm the team will check.
• People file: a founder, aide, or contact tied to a source.
• Source file: a page, filing, list, or firm note that backs a fact.

Build an identity card from source facts

Use facts from a good source. GLEIF says core LEI data has a legal name, registered address, and country of formation. [2] Do not fill in facts that you do not know.

Add an ID from the right public list if you have one. The SEC gives a CIK to each person or firm that files with it. [3] It helps with those filers. It does not name a firm that did not file.

UK firm data may be checked at Companies House. It holds the UK list of firms and makes firm data public. [5] Keep an LEI as it came from the source. GLEIF says its list holds old and new LEI files plus key data. [2]

• Legal name and known brand name.
• Country of formation and the list used.
• Registry ID, CIK, or LEI if the source gives one.
• Web address, source link, and check date.
• Known gaps, such as an old address or a brand link not yet checked.

Match in levels, then review

Start with hard proof. The same registry ID is better proof than a close name. A web address can help. It is not legal proof. Save why you made each link.

Then look at more than one fact. GLEIF checks likely repeat LEI files with many data points. It makes weak name terms more alike. It then checks fields such as place and firm type. This helps cut false links. [1]

Use four clear tags. Yes means strong proof backs a link. Maybe means a person needs more proof. No means the files are not one firm. Not known means proof is not there.

Do not let an AI score close the case. NIST says teams should check AI sources and quotes in tests and later checks. [6] A person must choose: join, link, or split.

• Exact ID match: check source and date, then link.
• More than one fact matches: send it to a named reviewer.
• One close name: keep files apart until there is better proof.
• Facts clash: stop the match and find a new source.

Keep a log and fix errors

Keep the match result. Keep the cause. Keep the source, date, and name of the reviewer. Keep the first lead note too. A later person may need to see why a link was made or undone.

Set a way to fix a file. Recheck it if a firm gives new legal facts. Recheck it if a public list changes. Recheck it if sources clash. Do not hide a fact that is in doubt.

Contact details can be personal data. Where GDPR rules apply, such data must be right and up to date when needed. Bad data must be fixed or erased with no delay. [4] Rules vary by case and place. Ask a skilled legal or data team about your case.

Track the work, not deal gains. Count repeat alerts. Count time in the check queue. Count links that change. Count files with no source. These show where the process needs help.

• Who made the choice and when.
• Which facts matched or clashed.
• Which source backs each key fact.
• What must happen before the file can move.