Skip to content

Deal economics

Data licensing cycle time study: a protocol for timing each deal stage

By SourceX Editorial · Updated

10 sections

Short answer

A study can measure where data licensing deals spend time by recording a dated start and end event for every stage of each deal, plus who or what each stage was waiting on, then reporting elapsed, active and waiting time with open and abandoned deals handled as censored. This page publishes that protocol, a stage timestamp dictionary and a pre-registration checklist. It reports no results.

License terms

Data licensing cycle time

This page covers data licensing cycle time study: a protocol for timing each deal stage.Illustrative

Key takeaways

  • This page is a study protocol; it reports no durations or results.
  • Each stage is defined by an observable start event, end event and evidence record.
  • Recording what each stage waited on, and its predecessors, explains where time goes, not just how much.
  • Open deals are censored and abandoned deals are a competing outcome; neither is dropped.
  • Results would be medians with intervals by stage, published only for cells large enough to be stable and anonymous.

How could a study measure where data licensing deals spend time?

By treating each deal as a sequence of stages with observable start and end events, recording what every waiting period was waiting on, and analyzing the result as time-to-event data. Durations alone say a deal took a long time; dependencies say why, and which stage would have shortened the whole deal if it had moved faster.

Anecdotes about slow legal review or slow data exports are not evidence, and this page assumes no figures about how long any stage takes. This page is a protocol, not a finding. It sets out the question, population, sampling, variables, measures, uncertainty and limits, so that a study run later, by SourceX or anyone else, can be judged against a plan written before the data was seen. It reports no results, and every number on the page is either a field definition or labeled illustrative.

Which deals are in the population, and what is the unit of analysis?

The unit is one proposed license: one data-holding company, one licensee or intermediary, one defined dataset scope. A company that licenses two separate datasets contributes two deals; a renewal of an existing license is a separate, flagged deal.

At a glance

Which deals are in the population, and what is the unit of analysis

The population is every proposed license that reached a defined entry event, such as a first qualified conversation, inside a fixed calendar window, whatever happened afterward. Including abandoned and stalled deals is not optional. A study that counts only signed deals measures the survivors, and survivors tend to be the deals whose stages moved smoothly.

  • Sampling frame: the deal registers of participating intermediaries or companies, with the frame source and any known coverage gaps disclosed.
  • Recruitment: every deal in the frame within the window, or a random sample if the frame is large; never hand-picked examples.
  • Window: entry events between two fixed dates, with follow-up continuing to a stated cut-off date.
  • Strata recorded, not filtered: data type, record volume band, presence of personal data, buyer type, and whether an intermediary was involved.
  • Exclusions, stated in advance: deals with no recoverable dates for at least the entry and one later stage, reported as a count.

What goes in the stage timestamp dictionary, and how are its dates coded?

The core asset is a fixed dictionary of stages, each defined by an observable start event and end event and the record that evidences them. Two coders reading the same email thread or deal log should reach the same dates. Stages can overlap; the dictionary defines each one independently and the dependency model below handles the overlap.

Code every date from a record, never from memory, and keep a pointer to the record that supports it. Deal histories are scattered across CRM stage changes, shared document version histories, email threads, e-signature audit trails and transfer logs, and each source has its own idea of time. The coding rules below keep two coders consistent and make every timestamp auditable later.

A double-coded subset, drawn at random before coding starts, measures agreement between coders on stage boundaries and waiting_on attributions. Disagreements are resolved by a third reader using the evidence pointer, and the agreement statistic is reported with the results rather than tuned until it looks acceptable.

  • Precedence: a system-generated event (signature audit trail, CRM stage change, transfer receipt) outranks an email that mentions the same event.
  • Date only: stages are timed in calendar days in the counterparty's stated time zone; times of day are not coded.
  • Ambiguous starts: if a stage has no clear start record, its start is coded as the end of its last predecessor and flagged as inferred.
  • Missing ends: a stage with no end record by the cut-off date is open and censored, not closed on the last email.
  • Evidence pointer: every coded date carries the record type and an internal reference, held by the study team and never published.
  • Redaction at source: counterparty names, people and commercial terms are removed before coders see the material.
What goes in the stage timestamp dictionary, and how are its dates coded?
StageStart eventEnd eventEvidence record
QualificationEntry event (first qualified conversation)Go or no-go decision recordedCRM stage change, meeting notes
Data discovery and inventoryRequest for system and record inventoryInventory accepted by both sidesShared inventory document version history
Sample and buyer-fit reviewSample or schema sentFit decision communicatedTransfer log, decision email
Rights reviewRights questionnaire or document request sentRights position signed offReview tracker, counsel sign-off
Privacy and de-identification designPrivacy review openedDe-identification method approvedReview ticket, method document
Commercial termsFirst term sheet or offer sentTerms agreed in writingOffer and redline history
Contract executionExecution version circulatedLast signatureE-signature audit trail
Extraction and preparationSigned agreement or kickoffPrepared dataset passes internal checksJob logs, QA report
Delivery and acceptanceDelivery initiatedAcceptance or deemed acceptanceTransfer receipt, acceptance notice

How should dependencies between stages be recorded?

For every stage, record what it was waiting on, not only when it started and ended. The GAO Schedule Assessment Guide's scheduling practice is the right model: dependencies depict the sequence in which activities occur, most links are finish-to-start, every activity except the first and last has a predecessor and a successor, and the critical path is the continuous chain that sets the finish date.

Applied to a deal, that means each stage carries its predecessor stages and the type of link, and each waiting interval inside a stage carries a blocking party. GAO also warns that lags and date constraints can distort the critical path, which in a deal study means calendar holds such as a board meeting date must be recorded as constraints with a reason, not hidden inside a stage's duration.

How should dependencies between stages be recorded?
FieldValuesPurpose
predecessor_stagesList of stage idsRebuilds the deal's dependency network
link_typeFinish-to-start, start-to-start, finish-to-finishShows which stages could run in parallel
waiting_onData holder, licensee, intermediary, counsel, third party, calendar constraintAttributes waiting time to a party
wait_start and wait_endTimestamps for each blocked interval inside a stageSeparates waiting from active work
constraint_reasonFree text, coded afterwardExplains imposed dates such as board approvals
rework_flagYes or no, with the stage that triggered itCaptures stages that reopened after a later finding

Which measures does the study report?

Report five measures per stage and per deal, each with its denominator stated. Medians and interquartile ranges are the default summaries because durations of this kind are usually skewed; means are shown only alongside them.

Critical-path share needs the dependency fields from the previous section: for each deal, the study rebuilds the network of stages, finds the longest chain of linked stages from entry to acceptance, and records which stages sat on it. A stage that is often long but rarely on that chain is a poor target for anyone trying to shorten deals, and the report should say so explicitly rather than ranking stages by duration alone.

  • Elapsed time: end event minus start event in calendar days, per stage and for the whole deal from entry to acceptance.
  • Waiting share: the proportion of a stage's elapsed time spent in recorded wait intervals, broken down by waiting_on party.
  • Critical-path share: how often each stage lies on the critical path, and the share of total deal time it accounts for when it does.
  • Rework rate: the proportion of deals in which a stage reopened, and the added elapsed time attributable to reopening.
  • Conversion by stage: the proportion of deals entering a stage that complete it, the stage where abandoned deals stopped, and the recorded reason.

How are open and abandoned deals counted without biasing the result?

Treat them as censored observations rather than dropping them. A deal still open at the cut-off date has a known lower bound on its duration, which the NIST/SEMATECH e-Handbook describes as right-censored data: the event time is beyond the observation window, so only a minimum is known.

Time to signature and time to acceptance are therefore estimated with a distribution-free survival method such as Kaplan-Meier, which the e-Handbook describes as needing no assumed distribution, with censored units contributing up to the point they leave observation. Abandoned deals are handled as a competing outcome, not as long deals, and the protocol states in advance which analysis treats them which way. Censoring that depends on the deal itself, such as a stalled deal quietly dropped from a register, is listed as a threat to validity and checked by comparing register drop-outs with recorded abandonments.

How many deals, compared against what, and with how much uncertainty?

Set the minimum sample before collection and publish per-stage estimates only where enough deals completed that stage. A cell-size rule, written into the pre-registration, prevents a single unusual deal from appearing as a stage benchmark and also protects participants from being identified.

The baseline is internal, not external: each stage is compared with the same stage across strata, such as deals with and without personal data, rather than with figures from unrelated industries. Uncertainty is reported as intervals around medians and survival estimates, with resampling done by participating organization rather than by deal, because deals handled by the same team are not independent. Any comparison not named in the pre-registration is labeled exploratory in the results.

What would the study not show?

It would describe deals in the participating frames during one window, not the whole market. Deals run directly between large AI developers and large publishers, which may be negotiated differently, are outside the population unless a participant contributes them.

It would not show why a stage is slow in a causal sense; waiting_on attributes time to a party, it does not prove that party could have moved faster. It would say nothing about value, price or whether a license was a good decision. And it cannot show that a company has the right to license its records; rights, privacy permission, ownership and contractual licensing rights are separate questions decided case by case.

What must be registered before any deal is coded?

Register the plan in a time-stamped public or escrowed record before any deal data is coded. Preregistration's main purpose, as the Center for Open Science puts it, is to separate hypothesis-generating exploratory work from hypothesis-testing confirmatory work. The disclosure items below follow AAPOR's disclosure standards and the STROBE checklist for observational cohort studies, adapted to deals.

At a glance

What must be registered before any deal is coded

  • Research question and the exact primary measures, with their denominators.
  • Population, entry event, calendar window and follow-up cut-off date.
  • Sampling frame source, coverage gaps and recruitment method.
  • The stage dictionary version, with start and end event definitions and evidence rules.
  • Coding procedure, number of coders and the inter-coder agreement check on a double-coded subset.
  • Handling of open deals (censoring) and abandoned deals (competing outcome).
  • Strata and the comparisons planned between them.
  • Minimum cell size for publishing any estimate.
  • Uncertainty method, including resampling by organization.
  • Anonymity rules for participants and counterparties, and who reviews outputs before release.
  • Everything that will be reported regardless of outcome, including null and unfavorable findings.

What does the dictionary offer a company in the middle of one deal?

A company considering licensing its records does not need to wait for study results to benefit from the dictionary. Logging the start and end events for its own deal, and naming what each wait was blocked on, turns a vague sense that things are slow into a list of specific dependencies someone can act on.

On deals run through SourceX, the stages in the dictionary map onto work SourceX carries itself: qualification, rights review, the contract, preparation and delivery. The company approves the scope and terms and signs before any record moves, and each deal's signed Dataset Deal Confirmation fixes the delivery date. The qualification stage begins with a brief assessment of the business that needs no uploaded records.

Frequently asked questions

Does this page say how long data licensing deals take?

No. It publishes a measurement protocol and a stage dictionary only. Any results would come from a later study run under the pre-registered plan and would be reported with their population, period and uncertainty.

Why record what each stage was waiting on?

Elapsed time shows that a stage was long; the waiting_on field shows whose input it was blocked on. Without it, time spent waiting for a counterparty looks the same as time spent working, and the slow party is impossible to identify.

Why include deals that never signed?

Excluding abandoned and stalled deals measures only the survivors, which tend to be the deals that moved smoothly. Keeping them, with open deals treated as censored, avoids understating how long deals take and shows where they stop.

What is the critical path in a data licensing deal?

It is the chain of dependent stages that sets the earliest possible finish date. Shortening a stage on the critical path shortens the deal; shortening a stage that runs in parallel off the path does not.

Can one company use the stage dictionary on its own?

Yes. Logging start and end events and waiting parties for a single deal is useful for managing that deal. It is not a benchmark, because one deal says nothing about how other deals behave.

Sources

  • GAO's Schedule Assessment Guide says logic links depict the sequence of activities, most should be finish-to-start, and lags and constraints can distort the critical path. Source
  • The NIST/SEMATECH e-Handbook defines Type I censoring, also called right-censored data, where event times beyond the observation period are not observed. Source
  • The NIST/SEMATECH e-Handbook describes the Kaplan-Meier approach as distribution-free, with censored units contributing to the estimate up to when they leave observation. Source
  • The Center for Open Science says preregistration separates hypothesis-generating (exploratory) from hypothesis-testing (confirmatory) research. Source
  • AAPOR's disclosure standards ask for the sampling frame, coverage gaps, dates of data collection and sample sizes. Source
  • STROBE publishes reporting checklists for observational studies, including a cohort study version. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies