Skip to content

AI data market

Source system dependency and the data moat: a five-shock stress test

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

Source system dependency turns a data moat into a lease when the platform holding the records controls the route out. Before crediting the moat, test five shocks: a terms change such as the ban on outside apps training LLMs on API data in Slack's current terms, an API retirement, a product sunset, a forced migration and buyer concentration. What survives belongs to whoever controls owner-run exports and signed licenses.

Value drivers

Source system dependency

This page covers source system dependency and the data moat: a five-shock stress test.Illustrative

Key takeaways

  • A moat built on one source system is only as durable as the route records take out of it; whoever can switch that route off holds the real advantage.
  • Platform terms now reach training use directly: Slack's API terms bar outside apps from training LLMs on API data, and X's developer agreement bars training foundation or frontier models.
  • Interfaces and products expire on published schedules, as Salesforce's retirement of API versions 21.0 to 30.0 and Google's deletion of Universal Analytics data showed.
  • Concentration can sit at either end of the pipe: a few record holders upstream, or a few buyers downstream, as Reddit's fiscal 2025 annual report discloses for its own licensing.
  • Score eight lines with a test that could break each, and treat a dependency that survives a replication pilot on a second system as resilience, not as a moat.

Is a dataset built on one source system a moat or a dependency?

It is a dependency until the company shows it controls the route the records take out of that system. A data moat that rests on one source system, such as a single CRM, help desk or messaging platform, inherits every decision the platform's owner makes about terms, interfaces and product life cycles. Investors should credit the moat only with the parts a platform change cannot reach.

Source system dependency, as used here, is the share of a data company's licensed value that originates in one software system and travels out through a route someone else controls. The definition has two parts on purpose. Concentration alone is common and often harmless: plenty of strong datasets come from Salesforce or Zendesk because that is where companies keep their records. The risk appears when concentration meets a route the data company does not control.

This article gives investors a way to test that risk: a map of who controls each route, five shocks with dated public examples, an eight-line scorecard with a disconfirmation test for each line, the strongest argument that the risk does not matter, and the questions to put to any company in the market, SourceX included.

Who controls the route out: the platform, the record owner or the data company?

Three parties can control how records leave a source system, and the moat belongs to whichever one can stop the flow. A platform controls routes that run through its developer program. The record owner, the business whose account holds the data, controls the exports built into its own account. The data company controls only what it generates itself.

Platform-controlled routes now carry explicit limits on AI use. Slack's API terms, effective October 10, 2025, say an application offered for use outside the developer's own organization may not use API data to train a large language model, and may not bulk export Slack message and file data unless an additional agreement expressly allows it. X's developer agreement lists using the X API or X content to fine-tune or train a foundation or frontier model among its prohibited uses, and X's restricted-use page makes a single exception for its own Grok.

Owner-controlled routes are different in kind. Salesforce lets an administrator generate a backup export from the org's own account once every 7 days on Enterprise, Performance and Unlimited editions, or every 29 days on the monthly schedule, and the zip file is deleted 48 hours after the notification email, weekends excluded. A route like that cannot be revoked by a developer policy change, but it runs only while the record owner keeps using it, which makes the owner relationship the real asset.

The practical test is simple. For each source system behind the dataset, write down the route, the document that governs it and the party that can end it. A company whose records arrive through owner-run exports under a signed license holds a different asset from one whose pipeline depends on an outside app's API access, even when the two datasets look identical on delivery.

Five shocks that break single-system data businesses

Five shocks can cut a single-system data business off from its supply or its buyers, and each has a dated public example. Run them in order against the company's main source system and record, for each one, what breaks and what survives. The figure shows where each shock lands on a platform-controlled route, and the owner-controlled route that survives most of them.

The first four shocks act on supply and the fifth acts on demand. A company can be well protected against platform changes and still depend on two buyers, which is why the demand shock belongs in a source system test even though it starts at the other end of the pipe.

Five shocks that break single-system data businesses
ShockDated public exampleWhat it breaksWhat survives it
Terms changeSlack API terms effective October 10, 2025: outside apps may not train LLMs on API data or bulk export messages without an additional agreementThird-party pipelines whose business is training dataExports the record owner runs from its own account and licenses directly
API retirementSalesforce retired Platform API versions 21.0 to 30.0 in Summer '25 after first scheduling it for Summer '23; calls to them fail as deactivatedExtraction code pinned to retired versionsPipelines with tested upgrade paths and versioned field maps
Product sunsetGoogle stopped processing standard Universal Analytics data on July 1, 2023 and deleted it from the week of July 1, 2024History that was never exported before the deadlineArchives exported in time, with record counts that prove completeness
Forced migrationAtlassian ended support for its Server products on February 15, 2024 and pointed customers to Cloud or Data CenterContinuity of IDs, fields and history across the moveDatasets with a documented crosswalk from old to new identifiers
Buyer concentrationReddit's Form 10-K for fiscal 2025 attributes substantially all licensing contract value to two partnersPricing power and leverage at renewalA buyer base spread across several tasks and contracts
A rented route from a source system through a third-party API to a dataset and buyers, hit by five shocks, above an owned route in which the record owner's own export and a signed license lead to an accepted dataset.
Five shocks on a platform-controlled route, and the owner-controlled route that survives most of them. Shocks come from the cited examples; the routes are illustrative.

Eight lines to score, each with a test that could break it

Score each line from documents and data rather than management's description, and attach the test that would prove the line weaker than claimed. The lines follow four comparisons, replacement cost, legal scope, technical utility and repeat buyer behavior, plus the route and exposure lines that matter only for businesses fed by one system.

Thresholds are editorial. Where a line names one, such as flagging a single system above half of accepted value, treat it as a starting assumption to adjust for the market, not a measured norm. Denominators matter more than thresholds: use accepted records or contract value, never raw rows exported, because raw volume flatters whichever system produces the most noise.

Eight lines to score, each with a test that could break it
LineWhat to measureDisconfirmation testWeak result
System shareShare of accepted records, by count and by contract value, from the largest single source systemRecompute on accepted records only; if the share jumps, raw volume was hiding the dependencyOne system above half of accepted value with no second system ever piloted
Route controlFor each system: the route, its governing document and the party that can end itRead the current developer terms and look for training or bulk-export limitsCore supply arrives through an outside app's API access
Legal scopeWhether rights come from the record owner's signed license, separate from the platform's permission to exportAsk counsel to find any platform clause that limits use after exportPlatform terms, not the owner's license, decide permitted use
Interface exposureAPI versions and fields the pipeline uses, against the vendor's published retirement datesList every endpoint version in use next to the vendor's retirement noticesProduction extraction on a version with a retirement date and no tested upgrade
Life-cycle exposureEnd-of-support and sunset calendars for each source product, and whether history is deleted at shutdownCheck that the archive was exported and reconciled before any announced shutdownHistory that exists only inside a product with a published end date
Replacement costCalendar time and engineering effort to rebuild the same scope from a second systemRun a replication pilot on a second system and score it with the buyer's acceptance testNo second system has ever been mapped, so the cost is unknown
Technical utilityBuyer task results on records normalized to a neutral schema versus native exportsRemove platform-generated fields, such as built-in AI summaries, and rerun the buyer's testValue depends on fields the platform generates and controls
Repeat buyer behaviorShare of contract value from the top two buyers, renewals and the number of distinct buyer tasksRemove the largest buyer and recompute the renewal and pricing evidenceMost value from one or two buyers, with renewals untested

Schema changes: the slow leak behind a stable-looking pipeline

Schema change is the shock investors most often miss, because it rarely stops a pipeline outright; it quietly changes what the records mean. An API version retires, a field is renamed, a picklist gains values, or a platform starts writing machine-generated summaries into fields that used to hold human notes. Each change can leave row counts intact while breaking the comparability a buyer pays for.

Retirements arrive on a schedule the platform sets. Salesforce first scheduled the retirement of Platform API versions 21.0 through 30.0 for Summer '23, postponed it to Summer '25, and said calls to those versions would then fail with an error indicating the endpoint is deactivated; the change covered the SOAP, REST and Bulk APIs in that range. A company that pinned extraction to an old version had years of warning, and a diligence team can check whether it used them.

Ask for three numbers, each with its period and denominator: pipeline breaks caused by source changes per year, median days to restore a comparable feed after a break, and the share of historical records that still map to current field definitions. A company that cannot produce them has not measured the leak, which is itself the finding.

Forced migrations are schema change at full scale. When Atlassian ended support for its Server products on February 15, 2024, customers moving to Cloud or Data Center carried years of Jira and Confluence history into a new deployment, where identifiers and field configurations may not carry over unchanged. A dataset that spans such a move needs a documented crosswalk, or its early and late years describe two different systems under one name.

Concentration at both ends: record holders upstream, buyers downstream

Supplier concentration and buyer concentration are the same risk in different positions: too much of the value passes through too few parties. Upstream, ask what share of accepted records comes from the three largest record holders and whether each holder's license has a stated term. Downstream, ask what share of contract value comes from the two largest buyers.

Even a well-known licensor discloses the downstream version. Reddit's Form 10-K for fiscal 2025 states that "substantially all of the contract value associated with our licensing revenue is derived from two of our partners", and the same risk factor flags renewal of those arrangements as uncertain. A company with a large, distinctive corpus can still have its pricing set by two counterparties.

Upstream concentration is harder to see because it hides inside volume. One large record holder can supply most of a dataset's rows while contributing few of the accepted records that carry value, or the reverse. Report both counts, by record holder and by accepted record, with the term remaining on each holder's license. A single holder whose license ends within the year is a supply shock with a date on it.

Could the records be rebuilt from another system or bought directly?

Often yes, and that is the strongest argument that source system dependency does not settle the moat question either way. If the same scope can be rebuilt from a competing system, the platform risk is manageable, but so is a rival's cost of entry. If a buyer can license directly from the record holders, the data company's position rests on service, not supply.

Test it with a replication pilot rather than an argument. Map a sample of the same record types from a second system, such as a different help desk or CRM used by a comparable business, and run the buyer's own acceptance test on both versions. If the second system passes, the dependency is a resilience problem only. If it fails because the first system captures fields the second lacks, that supports a narrow moat. If it fails because the first company keeps better records, the moat belongs to the record holder, not the system.

The direct-sourcing alternative needs its own test. Ask how many of the company's record holders have been approached by buyers directly, and what the data company adds that a direct deal would not: rights review, preparation, delivery, payment or aggregation across holders. If the honest answer is only access, the moat is a lease on someone else's relationship.

Regulation can lower the walls as well. The EU Data Act, which applies from 12 September 2025, gives users of connected products the right to access the data their use generates and to share it with third parties of their choice, and it limits what the data holder may pass on outside its contract with the user. Where it applies, it weakens a manufacturer's control of the route, which helps resilience and erodes exclusivity at the same time.

What could make the dependency thesis wrong

The thesis that source system dependency drains a moat can fail in at least five ways, and each deserves a test before an investor marks a company down.

  • Platforms may monetize access themselves. X already reserves model training on its content for its own Grok; a platform that licenses data directly would compete with data companies rather than cut them off, turning a route risk into a margin risk.
  • Owner-run exports may already carry the supply. If most records arrive through record owners' own exports under signed licenses, platform terms matter less than the five shocks suggest.
  • Terms can loosen. Developer terms change in both directions, and a restriction in today's terms can narrow later, lowering the value of today's caution and of today's moat alike.
  • Substitutes may close the gap. If synthetic or openly licensed records pass the buyer's acceptance test, dependency on any one system stops mattering because the scope itself stops being scarce.
  • Concentration may reflect an early market rather than fragility. A young market can look concentrated before more buyers arrive, so read the trend across several filings or contract years, not one snapshot.

Questions for any single-system data company, SourceX included

Put the same questions to every company that sells records drawn from business systems, and ask for documents rather than descriptions. SourceX belongs on that list. When a business licenses operational records to AI developers through SourceX, SourceX carries the deal from finding and qualifying that business through rights review, the contract and data preparation to delivery and payment. Buyers' datasets are sourced on request from U.S. businesses rather than kept in inventory, which makes upstream concentration and route control the scorecard lines that matter most here.

Two published practices bear directly on the scorecard. Nothing is delivered to SourceX until the business that owns the records has approved the scope and terms and signed the agreement, so the license runs from the record owner, not from a platform. When a delivery is large, the records either remain in the seller's own storage or travel on encrypted drives; SourceX does not host multi-terabyte datasets. Neither practice is a measured result, and the figures behind the eight lines are not public. SourceX's raise, made under Rule 506(c), is open to accredited investors, who can request materials at /investors and hold each of the eight lines to the definitions above. Useful questions for any company:

  • Which source systems supply the accepted records, and what share of accepted contract value comes from the largest one?
  • For each system, which route moves the records, which document governs it and who can end it?
  • Which API versions and fields does extraction use, and which of them carry published retirement dates?
  • Has any source product announced an end of support, and was its history exported and reconciled first?
  • What happened when a replication pilot rebuilt the scope from a second system and the buyer's test was run?
  • What share of contract value comes from the two largest buyers, and for how many distinct tasks do buyers use the data?

Methods note: how the stress test and scorecard were built

The five shocks and eight lines are an editorial framework built on 2026-10-26 from primary sources: Slack's API terms, X's developer agreement and restricted-use page, Salesforce's API retirement notice and export documentation, Google's Universal Analytics notice, Atlassian's Server end-of-support FAQ, Reddit's Form 10-K for fiscal 2025 and the EU Data Act. Each was checked on its publisher's own domain through search excerpts, so a reviewer should open the pages before release.

Nothing here is measured across companies. The scorecard has not been applied to any named data business, no shares or thresholds are observed norms, and no SourceX results are reported. The framework does not measure model quality, price, or the enforceability of any specific contract; those need counsel and the buyer's own tests. Platform terms and retirement schedules change, so recheck each source on the date of the review.

Frequently asked questions

What is source system dependency in a data business?

It is the share of a data company's licensed value that comes from one software system and leaves through a route someone else controls. Concentration in a popular system such as a CRM is normal; the risk is a route, such as an outside app's API access, that the platform can restrict or retire. Measure it on accepted records and contract value, not raw rows.

Can a platform stop a data company from licensing records exported from it?

A platform can restrict the routes it controls, such as API access for outside apps; Slack's API terms, for example, bar those apps from training large language models on API data. Whether a record owner may license records it exports itself depends on the owner's agreements and the platform's terms, so counsel should read both. This is general information, not legal advice.

Does relying on Salesforce or Zendesk records weaken a data moat?

Not by itself, because records often sit in one popular system simply because that is where companies work. The moat weakens when value depends on fields the platform generates, on API versions with retirement dates or on routes the platform can switch off. A replication pilot on a second system shows which of those applies.

What is the right denominator for a data company's buyer concentration?

Use contract value rather than deal count, and report the share from the two largest buyers alongside renewals and the number of distinct buyer tasks. Public companies disclose it in risk factors: Reddit's Form 10-K for fiscal 2025 attributes substantially all of its licensing contract value to two partners. Ask private companies for the same measure with its denominator.

What evidence shows a single-system dependency is not a threat to the moat?

Three findings together: records rebuilt from a second system pass the buyer's acceptance test, supply arrives through owner-run exports under signed licenses, and no interface in use carries a retirement date. Then the dependency is a resilience question to manage, not a reason to discount the moat.

Sources

  • Slack's API terms, effective October 10, 2025, bar applications offered outside the developer's organization from using API data to train a large language model and from bulk exporting message and file data without an additional agreement. Source
  • X's developer agreement lists using the X API or X content to fine-tune or train a foundation or frontier model among prohibited uses. Source
  • X's restricted-use page says X prohibits use of its APIs and content to fine-tune or train a foundation or frontier model, with the exception of Grok. Source
  • Salesforce lets administrators generate backup exports every 7 days (weekly, on Enterprise, Performance and Unlimited editions) or every 29 days, and deletes the zip files 48 hours after the email, not counting weekends. Source
  • Salesforce postponed the retirement of Platform API versions 21.0 through 30.0 from Summer '23 to Summer '25; calls to retired versions fail with an error that the endpoint is deactivated, across SOAP, REST and Bulk APIs. Source
  • Standard Universal Analytics properties stopped processing data on July 1, 2023; from the week of July 1, 2024 the properties and API became inaccessible and the data was deleted. Source
  • Atlassian Server products are no longer supported as of February 15, 2024, with customers directed to Cloud or Data Center. Source
  • Reddit's Form 10-K for fiscal 2025 states that substantially all of the contract value associated with its licensing revenue is derived from two partners, and flags renewal risk. Source
  • The EU Data Act applies from 12 September 2025, gives users of connected products the right to access and share the data their use generates with third parties, and limits data holders' use of non-personal product data to what their contract with the user allows. Source

Request SourceX investor materials

SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.

See if your company qualifies