Leadership and readiness
Data startup investor communications accuracy: who signs each claim
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
Data startup investor communications accuracy comes from a written claims process, not a careful founder. Each public claim needs a class, an approver who owns the evidence, a record on file before it runs and a trigger that forces a correction. A Rule 506(c) raise can be advertised, and information given to investors must be free of false or misleading statements, so ask for the claims register before the traction slide.
Checklist
Data startup investor
Key takeaways
- Accuracy comes from a process: class, evidence owner, proof on file and a correction trigger for every claim.
- A Rule 506(c) raise can be advertised, and information given to investors must be free of false or misleading statements.
- One illustrative pipeline supports a headline 57.1 times its paid count; always ask which step a number counts.
- The author never approves; the owner of the evidence signs, plus counsel and the CEO for investor materials.
- A correction must reach every surface, which is why the register lists where each claim appears.
What makes a data startup's investor communications accurate?
A written process that runs before anything is published, not a founder who means well. Every public claim needs four things: a class that says what kind of statement it is, a named approver who owns the evidence, a record on file before the claim runs, and a trigger that forces a correction when the record changes. Data startup investor communications accuracy is what that process produces; a careful tone is not a substitute.
Data companies are unusually exposed. Their traction words, such as suppliers, datasets, buyers and pipeline, each have several honest definitions, and the gap between them can be large. Their technical claims, such as rights-cleared, de-identified or AI-driven, sound precise to an investor and mean little without a method behind them. And their buyers often insist on confidentiality, so outsiders cannot check a claim against the customer, which puts more weight on the company's own controls.
For an investment committee, the useful question is not whether a startup's claims sound reasonable. It is whether the company can trace any claim you pick to the record that supports it, name who approved it and show what would make it change.
Which rules reach a startup's public claims, in plain terms
Two sets of rules matter most in the United States: the securities anti-fraud rules for anything said to investors, and the Federal Trade Commission's substantiation doctrine for marketing claims. This section is general information, not legal advice; which rules apply to a given statement is a question for the company's counsel.
On the securities side, the SEC's guidance on Rule 506(c) says an issuer may broadly solicit and generally advertise an offering, provided all purchasers are accredited investors and the issuer takes reasonable steps to verify that status; the rule text at 17 CFR 230.506 sets out the verification duty. The SEC's summary of Rule 506 adds that companies decide what information to give accredited investors, so long as it does not violate the anti-fraud prohibitions, and that any information provided must be free from false or misleading statements. Rule 10b-5(b) prohibits untrue statements of material fact, and omissions of material facts needed to keep statements made from misleading, in connection with buying or selling a security. In Macquarie Infrastructure v. Moab Partners (2024), the Supreme Court read that prohibition as reaching half-truths but not pure silence.
On the marketing side, the FTC's 1984 Policy Statement Regarding Advertising Substantiation says advertisers need a reasonable basis for objective claims before the claims are disseminated; evidence developed afterward may shape how the agency proceeds, but the expectation is a basis in hand first. Announcing a sweep of cases against deceptive AI claims in September 2024, the FTC's chair said there is no AI exemption from the laws on the books. And in a July 2024 post, the FTC's technology office said hashing does not make data anonymous and that the agency evaluates a company's privacy representations against its actual data handling.
Two enforcement cases, one missing control
Both cases turned on claims that nobody had to prove before they went out. They are examples of what the control is for, not a measure of how often startups misstate traction.
In June 2024 the SEC charged the founder of Joonko, a recruiting software startup, alleging that she told investors the company had more than 100 customers, including Fortune 500 companies, more than $1 million in revenue and more than 100,000 active job candidates, and that she supplied fabricated testimonials, falsified bank statements and forged customer contracts. These are allegations in a complaint, not findings. Every figure named sits in a class this article's register would route to a different evidence owner: customer counts to contracts, revenue to accounting records, candidates to a counted database, testimonials to written permission.
In March 2024 the SEC announced settled orders against two investment advisers, Delphia and Global Predictions, over statements about their use of artificial intelligence. The SEC found that Delphia claimed to use machine learning on client data that was not part of its investment process, that Delphia had agreed in 2021, after an examination, to correct similar statements, and that the false statements continued. Global Predictions called itself the first regulated AI financial advisor, which the SEC found false. A promised correction that never reached every place the claim appeared is the failure a register with a surface list is built to prevent.
The claims register: eight classes, the proof each needs and who signs
Sort every claim into one of eight classes, because the class decides what evidence must exist and who may sign. The register is the original asset of this article, built as an editorial tool; the wordings are illustrative, and no row describes any company's actual claims.
A data company can run this register in a spreadsheet: one row per live claim, with the exact wording, the class, the evidence link, the approver, the approval date, every surface where the claim appears and the trigger below. The eight classes, in rough order of how much damage a wrong claim does to an investor's trust:
| Claim class | Illustrative wording | Evidence on file before it runs | Who signs | What forces a correction |
|---|---|---|---|---|
| Traction count | 40 companies under contract | A dated count with the pipeline step defined and the contract list behind it | Operations owner, then the CEO | A churned contract, a changed definition or a stale count date |
| Revenue or money | Revenue grew every quarter | Accounting records for the stated periods, reviewed by finance | Finance owner; counsel for investor materials | A restatement, a reversal or a closed period |
| Named customer or testimonial | Trusted by a leading lab | Written permission to name, a contract in force, the person's own words | CEO with counsel, and the named party in writing | A contract that ends or permission withdrawn |
| Data rights | Rights-cleared datasets | Signed licenses covering the stated use and the rights review notes | Rights or legal reviewer | A license that ends, a dispute or a new use |
| Privacy or de-identification | Anonymized records | The method, its test results and a statement of the risk that remains | Privacy reviewer | A new linkage risk, a method change or an incident |
| AI or product capability | AI-powered matching | Where automation runs, where people step in and the test results | Engineering owner | A process change or manual steps found behind the claim |
| Market or third-party figure | A market of stated size | The primary source, its date and the exact passage used | The author plus a second reader | A revised source or one older than a set age |
| Forward-looking statement | We expect to sign more suppliers | Written assumptions, labeled as an expectation, never a promise | CEO with counsel | A failed assumption or a passed date |
Which pipeline step does a traction number count?
Whichever step the company defines in writing, and a reader should insist on knowing it, because the same pipeline can support a headline many times larger than its paid count. A traction number that is literally true at an early step becomes misleading when readers assume a later one, which is the half-truth problem in miniature.
The downside sensitivity below runs one invented pipeline, not any company's, through seven claims. With 400 inquiries narrowing to 7 paid deals, a headline drawn from the first step is 57.1 times the paid count, and one drawn from the framework-contract step is still 5.7 times. Each wording is accurate for its own step; the risk sits in what the reader fills in.
The test for diligence is simple: for every count in the deck, ask which row it comes from and see the list behind it. The claims and their multiples, rounded once to one decimal place:
| Claim a reader might see | Step it honestly counts | Illustrative count | Multiple of the paid count |
|---|---|---|---|
| 400 companies in our pipeline | Inquiries received | 400 | 57.1 |
| 120 qualified suppliers | Passed screening | 120 | 17.1 |
| 40 companies under contract | Signed a framework agreement | 40 | 5.7 |
| 18 datasets signed | Signed a deal for one dataset | 18 | 2.6 |
| 12 datasets delivered | Data delivered | 12 | 1.7 |
| 9 datasets accepted | Accepted after review | 9 | 1.3 |
| 7 deals paid | Paid | 7 | 1.0 |
Who may approve a claim, and who may not
The person who owns the evidence approves, and the person who wrote the claim never does. That single separation catches most errors, because the owner of a count knows its definition and the author usually does not. Investor-facing materials add two more signatures, counsel and the CEO, because the anti-fraud rules apply to them directly.
Approval rules that work in a company of ten people as well as a hundred:
- The author drafts and never approves the same claim.
- The evidence owner signs: finance for money, operations for counts, the rights reviewer for rights, engineering for capability.
- The deck, offering pages, investor updates and data-room summaries also need counsel and the CEO.
- Each approval records the exact wording, the evidence link, the date, the surfaces allowed and an expiry date.
- Reusing a claim in a new medium, such as a podcast, a post or a press quote, counts as a new claim.
- Copy drafted with AI tools enters the same register; a generated sentence is a claim like any other.
Correcting a claim that stopped being true
Find every place the claim appears, fix each one, and tell the people who relied on the old version. The register makes the first step possible, because it lists every surface; without that list, a correction depends on someone remembering where a number was pasted, which is how a fix promised after an examination can leave the original statements running.
Corrections start from triggers, not from complaints. Each class in the register names its own: a churned contract for a traction count, a revised source for a market figure, a method change for a privacy claim. A monthly sweep compares each live claim's evidence date with today and flags anything past its expiry, so that a claim true when written does not stay up unchanged after the facts move. Whether any legal duty to update applies to a given statement is a question for counsel, but an investor reading stale numbers will draw conclusions either way.
Record every correction as its own entry: the old wording, the new wording, the date, the reason and who approved it. An investor who asks for the correction history learns more from a short list of fixed claims than from a register with none, because a register that has never corrected anything has usually never been checked.
What could make a claims process the wrong investment
Its cost, its drift into theater, and five risks the process cannot control on its own. The cost objection first: an early company with a handful of claims can spend more on process than the claims are worth. The answer is scale, not exemption; a one-page register with seven columns takes an afternoon. Theater is the second: approvals can become rubber stamps, and a register can be filled in after publication. The test is to pick five live claims at random in a diligence meeting and ask the company to open the evidence for each on the spot.
Steps
What could make a claims process the wrong investment
These risks could make a claims process fail to protect an investor even when it exists, and each is worth one question:
- Bad source data: a process approves what the records say, so a wrong count in the system becomes an approved wrong claim. Ask how the underlying counts are audited.
- Confidentiality: buyers and suppliers often bar naming them, so public claims stay vague and the real evidence sits behind an NDA. Ask what can be shown under confidentiality.
- Speed: AI-assisted writing multiplies claims faster than reviewers can read them. Ask how many claims went live last quarter and how many were reviewed.
- Shifting enforcement focus: regulators' priorities change, but the baseline that investor information be free of false or misleading statements does not. Do not read a quiet enforcement year as permission.
- Silence as strategy: a company can avoid inaccurate claims by making almost none, which leaves investors with nothing to test. Ask what is shared privately and how it is defined.
What SourceX's published pages commit to, and what to ask
SourceX's published pages show several claim habits an investor can check, and they say nothing about its internal approval process, so this section describes only what is public. Its investor page says it shares company and commercial metrics with investors directly, so no traction counts appear in public copy to test against the register. Its process page qualifies its own statements: deal terms are described as typical, with final terms in each agreement; the sample data summary is marked as a sample for illustration, not a real company; the fit panel says it is not a valuation; and stated durations are described as targets rather than promises.
Those habits are a start, not proof of a process. SourceX is raising from accredited investors under Rule 506(c), and an investor who requests materials through sourcex.si/investors can ask for the same evidence this article asks of any data company:
Checklist · 5 items
- Which pipeline step defines each count in the materials, and can we see the list behind it?
- Who approves investor-facing statements, and is the approver ever the author?
- What evidence supports each rights or de-identification statement, and when was it last checked?
- Which claims were corrected in the last year, and how were the people who saw the old version told?
- Which statements are expectations, and what assumptions sit behind each?
0 of 5 ticked. Tick each item as you go.
Methods note: sources, illustrative inputs and limits
Rules and cases come from primary sources checked on 2026-10-27: the SEC's Rule 506(c) guidance, its summary of Rule 506 and the rule text in the eCFR; Rule 10b-5 and the Macquarie decision on Cornell's Legal Information Institute; SEC press releases 2024-70 and 2024-36; and three FTC publications. Most were verified through search excerpts on their own domains and are marked for a reviewer to open. SourceX statements come from its investor and process pages fetched that day.
The claims register, the approval rules and the seven-row pipeline are editorial tools drawn up for this article. Every pipeline count is invented, and the multiples are each count divided by the paid count, rounded once to one decimal place. Not measured: how many startups keep a claims register, how often traction claims are overstated, and any company's internal approval process, SourceX's included. The two SEC cases are examples, not a sample, and the Joonko matter consists of allegations. Nothing here is legal advice; counsel decides which rules apply to a specific statement.
Frequently asked questions
Are a startup's website and posts part of its offering under Rule 506(c)?
Rule 506(c) lets an issuer generally advertise its offering, so public pages and posts can end up in front of prospective investors, and the SEC says information given to investors must be free from false or misleading statements. Whether a particular page is offering material is a question for the company's counsel.
What is a claims register?
A claims register is a list of every live public claim with its exact wording, its class, the evidence behind it, who approved it and when, every place it appears and what would force a correction. It lets a company trace any statement to its proof and fix every copy when the facts change.
Who should approve a data startup's traction claims?
The person who owns the underlying count, usually the operations or finance owner, should approve, never the author. Claims in the deck, investor updates or offering pages should also go to counsel and the CEO.
What evidence does an anonymized data claim need?
A description of the de-identification method, the test results and a plain statement of the risk that remains. The FTC has said hashing does not make data anonymous and that it checks privacy promises against actual data handling, so a method-based description is safer than the word anonymized.
How should a startup correct a public claim that turned out to be wrong?
Find every surface where the claim appears, fix each one, tell the people who received the old version and log the correction with the old and new wording, the date, the reason and the approver. Counsel should advise on any legal obligations.
Sources
- The SEC says Rule 506(c) permits issuers to broadly solicit and generally advertise an offering, provided all purchasers are accredited investors and the issuer takes reasonable steps to verify their status. Source
- 17 CFR 230.506(c) requires that all purchasers in a 506(c) offering be accredited investors and that the issuer take reasonable steps to verify that status. Source
- The SEC's summary of Rule 506 says companies decide what information to give accredited investors within the anti-fraud prohibitions, and that information provided must be free from false or misleading statements. Source
- Rule 10b-5(b) prohibits untrue statements of material fact and omissions that make statements misleading, in connection with the purchase or sale of a security. Source
- In Macquarie Infrastructure v. Moab Partners (2024), the Supreme Court held that Rule 10b-5(b) reaches half-truths but does not prohibit pure omissions. Source
- In June 2024 the SEC charged Joonko's founder, alleging false statements to investors about customers, revenue and candidates, fabricated testimonials, falsified bank statements and forged contracts. Source
- On March 18, 2024, the SEC announced settled charges against Delphia and Global Predictions for false and misleading statements about their use of artificial intelligence. Source
- The FTC's 1984 substantiation policy requires a reasonable basis for objective advertising claims before dissemination; evidence developed later bears mainly on how the Commission proceeds. Source
- Announcing Operation AI Comply in September 2024, the FTC's chair said there is no AI exemption from the laws on the books. Source
- The FTC's technology office said in July 2024 that hashing does not make data anonymous and that the FTC evaluates privacy representations against data handling. Source
- SourceX's investor page offers investor materials on request and says it shares company and commercial metrics with investors directly. Source
- SourceX's process page describes typical deal terms with final terms in each agreement, labels its data summary a sample for illustration and its fit panel not a valuation, and calls stated durations targets, not promises. Source
Request SourceX investor materials
SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.