The production and distribution of cultural work has moved onto digital platforms, a shift that has reorganised how creators reach the people who value what they make (Poell et al., 2019). For most of the twentieth century an independent artist, writer, or musician reached an audience only after passing through an institutional gatekeeper. A publisher accepted a manuscript, a gallery hung the work, a label pressed the record, and that intermediary decided who was allowed to reach the public at all. Direct-support platforms invert this arrangement. A creator can now solicit recurring contributions from supporters without any curator standing at the entrance, and the relationship between creator and supporter is owned by the creator rather than mediated by a gatekeeper who selects on the creator's behalf. This inversion does not abolish mediation. Platforms still take fees, rank content through recommendation systems, and set the terms of participation. What changes is the locus of access: there is no gate that an external authority opens or closes, and the continuing relationship with supporters belongs to the creator. The phenomenon studied here is therefore the creator's ownership of a non-gatekept audience relationship, not a hypothetical condition of zero mediation. For independent creators this arrangement is demanding. The labour of building and keeping an audience is itself a form of work, often unpaid and frequently invisible, layered on top of the creative work for which a creator wants recognition (Duffy, 2016; Murad and Sfhea, 2023). 

A practitioner intuition, widely shared among creators who sustain themselves outside institutions, holds that durable independence depends less on being the most talented and more on treating audience cultivation as a core skill: communicating consistently, responding to supporters, and tending a relationship rather than assuming the work will find its audience on its own. This study takes that intuition seriously as a research question rather than as a conclusion.

Testing the intuition has been constrained by measurement. The patronage relationship has been described richly, but mostly in qualitative terms as an ongoing bond rather than a transaction (Swords, 2017; Wohn et al., 2019). Quantitative work on platform support has concentrated on the funding event, that is, whether a campaign reaches its goal (Lukkarinen et al., 2016; Frydrych et al., 2014; Bidouei and Bidouee, 2024). A campaign that succeeds once is not the same as an audience that stays, yet the field has lacked a measure that separates the two. Without such a measure, the claim that cultivation sustains a relationship cannot be examined, because the outcome the claim is about, the sustained relationship, has not been captured in numbers.

This article makes two contributions. First, it introduces a plain, reproducible measure of the sustained creator–audience relationship, a score that summarises how well a creator keeps its recurring supporters over time and that is designed to reflect a lasting relationship rather than a one-time windfall. Second, it uses the measure to examine the cultivation intuition on public data, first across a sample of 381 independent creator and community projects and then through a mixed-methods layer that codes what 101 communicating creators actually write to their supporters. The aim is not to crown cultivation as the cause of success, but to ask whether the practices a creator can be observed performing on a public platform account for a sustained relationship once the quality of the work is held constant.

Background and related work

From the funding event to the sustained relationship

Research on platform-based support has produced a robust account of what drives a single funding outcome. Success has been modelled as a function of a creator's social capital and early contributions (Colombo et al., 2014) of project and pitch characteristics (Lukkarinen et al., 2016), and of the perceived legitimacy a project signals to prospective backers (Frydrych et al., 2014). Campaign-design choices such as reward structure have been treated as further levers (Kraus et al., 2016). Correlates of funding success have also been characterised on large public collections of an entire platform category, with descriptive analysis preceding regression and explicit caution about observational limits, a template this study adopts for the separation of exploratory from confirmatory analysis (Sauermann et al., 2019). This literature is largely organised around a binary or threshold outcome, namely whether a project is funded. 

A funded campaign and a sustained audience are different objects. The first is an event; the second is a relationship that persists over time. The distinction matters most precisely in the independent, direct-support setting this study address, where a creator's livelihood depends not on one successful appeal but on supporters who keep contributing. The vocabulary for the second object exists. Patronage has been described as an ongoing, relational arrangement between creator and supporter rather than a one-off purchase, in both cultural-sociological (Swords, 2017) and human–computer interaction accounts of digital patronage (Wohn et al., 2019). What has been missing is a measure of persistence that can be computed from the data platforms actually expose.

Observable cultivation behaviours

If cultivation sustains a relationship, some of it should be visible in what creators do on the platform. The most directly attested observable behaviour is communication cadence: the posting of updates has been linked to participation during crowdfunding campaigns (Block et al., 2017). Relationship-centred framings in communications research treat funding as growing out of the personal relationships and social-media activity that a creator maintains, rather than as an anonymous market transaction (Borst et al., 2017). Early contributions and internal social capital have been read as social-proof signals that a creator can cultivate (Colombo et al., 2014). These literatures supply a menu of behaviours, including the regularity of communication, responsiveness to supporters, community-building, and the mobilization of supporters across channels, that a study can attempt to observe and relate to an outcome. A persistent difficulty is separating cultivation from quality. A creator whose work is simply better may both attract a lasting audience and, incidentally, post more. Published work handles perceived quality without subjective rating by building it from observable production and legitimacy signals, such as the richness and distinctiveness of a project's self-presentation (Taeuscher et al., 2020; Frydrych et al., 2014). The present study adopts that strategy, building a quality proxy from observable signals and holding it constant when it relates cultivation to the sustained relationship.

The Research gap

Three observations define the gap this study addresses. The sustained relationship is described but not measured; cultivation behaviours are catalogued but rarely separated from quality; and the two have not been brought together on data from the direct-support setting where the creator, not a gatekeeper, owns the relationship. A measure of the sustained relationship, together with a design that holds quality constant, is the prerequisite for examining the practitioner intuition at all.

Measuring the sustained audience relationship

We measure the sustained audience relationship by how well a creator keeps its recurring supporters over time. For each creator we follow the groups of supporters who signed up for recurring support and track what fraction are still contributing month by month. That month-by-month retention is summarised over the first two years of the relationship into a single sustained audience relationship (SAR) score between 0 and 1, where a score near 1 means the creator kept almost all of its recurring supporters and a score near 0 means supporters left quickly. Early months count for more than distant ones, because the near-term continuation of a relationship is the clearest sign of its health. One-time contributions are deliberately left out of the score, because a single payment says nothing about whether a relationship lasts. Two features follow directly from how the score is built, and both are what one would want from a measure of a lasting relationship. A windfall of one-time contributions, however large, does not change the score, so a good month of one-off giving cannot be mistaken for a durable audience. And a creator who keeps supporters at least as well as another creator at every point in time never receives a lower score, so the measure respects the plain ordering of who retains an audience better. These features make the score a measure of a sustained relationship rather than of momentary funding, which is the distinction the field has lacked.

Data and sample

The study uses public data from Open Collective, a platform built on financial transparency whose public interface reports, for each project, whether supporter contributions are recurring or one-time, the status of those contributions over time, the text of the updates a project posts to its supporters, and descriptive information about the project. Open Collective fits the measure because it records genuinely recurring support and distinguishes it from one-time giving, which is what a retention score requires. Open Collective is treated here as direct-support infrastructure in the sense set out in Section 1: a project is not selected by a curator who controls access, and it owns its relationship with supporters. The platform still mediates through fees and discovery, so the object of study is the creator-owned, non-gatekept persistence of support, not support under an absence of mediation.

The population is independent creators and community projects sustained by direct supporter contributions. This framing is broader than cultural creators alone, and the breadth is a deliberate, disclosed consequence of the data: the usable population of purely artistic projects on the platform is small, too small to carry a stand-alone analysis. Cultural production remains the organising lens and is examined directly as a subset (Section 5.3). A project entered the sample when it had at least 15 recurring-capable supporters and was created on or before 31 December 2024, so that a retention horizon could be observed. The resulting sample comprises 381 projects: 235 technology or open-source projects, 67 community projects, and 79 creative projects (artists, musicians, writers, journalists, and podcasters); Table 1 reports the composition.

Table 1: Analysis sample composition (N = 381).

Note. Inclusion: at least 15 recurring-capable supporters and created on or before 31 December 2024. Source: Open Collective public data.

Measures

The outcome is the SAR score described in Section 3. Cultivation is represented first by structural behavioural signals that can be computed for every project: the rate of updates per year, the regularity of supporter acquisition, and early acquisition momentum. The quality proxy is built from observable production and legitimacy signals, following prior work that captures perceived quality without subjective rating (Taeuscher et al., 2020; Frydrych et al., 2014), and the analysis also adjusts for project age and the number of supporters. The structural cultivation signals are entered alongside quality before any text-based measure is added, so that a text signal is judged net of how often a creator simply posts.

The mixed-methods coding layer

Structural signals capture how often and how regularly a creator acts, but not the content or style of what a creator communicates, which is closer to what the cultivation intuition is about. To observe content, a mixed-methods layer codes the text of the updates that creators post to supporters. The coding scheme comprises eight practice categories drawn from the relationship and cultivation literature reviewed in Section 2: communication regularity, responsiveness and dialogue, tone and warmth, community and collective identity, transparency and accountability, off-platform mobilisation, storytelling and mission, and reciprocity and recognition. Each update is coded for the presence of each practice, and a project's score on a practice is the share of its updates that show it. Coding followed a transparent, reproducible protocol applied uniformly across languages, with the definition of each practice held identical across languages and only the surface wording differing. 

A language was coded when it reached an inclusion threshold of at least five percent of updates or at least eight projects; updates in languages below the threshold were marked uncoded rather than silently scored zero. Of 929 updates observed, English and Portuguese met the threshold and were coded, and per-language coverage is reported with the results. Reliability was assessed as the agreement between two independent coding passes, expressed as a per-category agreement coefficient in the family of standard reliability measures for coding data (Hayes and Krippendorff, 2007). Categories reaching an agreement of 0.67 or above were treated as reliable for interpretation; categories below that level were kept only as exploratory. This agreement is a check between two coding passes and is not a substitute for a second human coder, a limitation stated in Section 7. Because it needs projects that communicate, the coded layer runs on the 101 projects that posted at least three updates, and within that group it asks a narrower question: among creators who do post publicly, does the content and style of their communication explain a sustained relationship once quality is held constant?

Study Analysis

The analysis relates the measures to the SAR score using regression with robust standard errors and standardised predictors, and it compares a model containing quality and controls with a model that adds cultivation, testing whether the added block explains more of the score. Rank correlations are used to check what the SAR score tracks. The exploratory and confirmatory parts are kept explicitly separate. A single confirmatory test is stated in advance: the association of responsiveness, the most directly relationship-cultivating practice, with the SAR score once quality is held constant. Every other association, including those for the structural signals and the remaining practices, is exploratory.

What the score captures

The SAR score behaves as a measure of a sustained relationship should. Across the full sample of 381 projects, it is strongly correlated with a simple month-by-month retention benchmark (Spearman rho = 0.75, p < .001), confirming that it tracks ongoing retention. At the same time it is only weakly related to the size of one-time funding (rho = 0.13, p = .011) and is unrelated to the share of contributions that are recurring (rho = −0.08, ns). 

The weak but non-zero link to one-time funding is worth stating plainly: the score is not perfectly independent of funding size, but it is far more a measure of who stays than of whom received a large one-off sum. It therefore separates a sustained relationship from a one-time funding event, which is the distinction the field has lacked and the basis for what follows. Fig. 1 shows the retention curves of a sustained and a decayed project with their scores annotated, and Fig. 2 plots the score against one-time funding magnitude.

Fig. 1: Retention curves for a sustained and a decayed project, with SAR scores annotated.

Cultivation behaviours and the sustained relationship

The central question is whether observable cultivation explains the SAR score once quality is held constant. Across both the full sample and the coded layer, it does not, within the limits of what is observable. In the full sample of 381 projects, adding the structural cultivation signals to a model containing the quality proxy and controls raised explained variance only from 0.037 to 0.039, an increment of 0.0016 that is not significant (F = 0.21, p = .89).

The strongest structural signal, early acquisition momentum, had an adjusted association with the score of 0.003 with a 95% confidence interval of [−0.012, 0.019], a null result; this and the other structural signals are reported as exploratory.

Fig. 2: SAR score versus one-time funding magnitude.

In the coded sub-sample of 101 communicating projects, adding the eight coded practices to a model with the quality proxy, controls, and the structural signals raised explained variance from 0.083 to 0.234, an increment of 0.151 that is marginal and not significant at the conventional level (F = 2.02, p = .054). 

The single pre-specified confirmatory test, responsiveness, was null (adjusted association −0.008; 95% confidence interval [−0.035, 0.012]; coding agreement 1.00). Of the eight coded practices, four reached the reliability threshold, namely communication regularity, responsiveness, off-platform mobilisation, and storytelling, with agreement between 0.85 and 1.00, and each of the four was null.

Fig. 3: Cultivation associations, unadjusted versus net of quality, with confidence intervals.

Fig. 4: How common each coded practice was, overall and within the creative subset.

The four categories below the reliability threshold, including tone and warmth, community, transparency, and recognition, are reported only as exploratory; among them, community had the largest association (0.027, p = .066), but its low coding reliability (agreement 0.47) means it must be read as a weak, exploratory signal rather than a finding

Fig. 5: Coded practices versus SAR score: adjusted associations with confidence intervals.

.Table 2 reports the per-practice associations with their reliability and prevalence, and Fig. 3, 4, and 5 displays the coefficient pattern and how common each practice was.

Note. Regression with robust standard errors, standardised predictors. Adding the practice block over quality, controls, and structural signals: increment in explained variance 0.151, F = 2.02, p = .054. Agreement is between two coding passes; categories below 0.67 are exploratory only.

Two features of the coded layer deserve emphasis. The practices that could be coded reliably were also rare: responsiveness, for instance, appeared in only about two percent of updates. And the practice with the richest narrative content, tone and warmth, was among the least reliably coded, because warmth in free text resists rule-based detection. The observable surface of cultivation on this platform is thin, and the part of it that can be coded reliably is thinner still.

Table 2: Coded cultivation practices and their association with the SAR score (communicating sub-sample, n = 101).

Creative projects

The creative subset is examined directly to honour the study's cultural-production focus, while respecting its small size. Across the full sample, the 79 creative projects had a mean SAR score of 0.69 compared with 0.68 for the 302 non-creative projects, a difference that is not significant (Mann–Whitney U = 12,572, p = .46). Creative and non-creative projects sustain audience relationships about equally well on this platform. Within the creative subset, early acquisition momentum was weakly and non-significantly related to the score (rho = 0.16, p = .17, n = 79). In the coded layer the creative subset makes up roughly a third of the 101 communicating projects, and a practice-by-creative interaction on the lead practice was null (−0.005, p = .83). The creative cases are reported as a described subset and as a variable in the analysis rather than as the basis for a separate strong claim, because their number is too small to carry one. Fig. 6 contrasts the score for creative and non-creative projects.

Fig. 6: SAR scores for creative versus non-creative projects.

The results combine into a single, careful story. The SAR score gives the field a measure it lacked: a number that tracks the persistence of a creator's audience and is clearly distinct from the size of a one-time funding event. On the substantive question, the observable behaviours a creator performs on a public direct-support platform did not explain a sustained relationship once the quality of the work was held constant, and this held in the full sample and again in the coded layer, including among the practices that could be coded reliably.

The interpretation matters as much as the result. This is absence of evidence, not evidence of absence. The cultivation intuition is about relational practices, the personal communication, responsiveness, and warmth that pass between a creator and individual supporters, and the work of community that often happens off the platform entirely. Public artefacts show very little of this. The reliably codable practices were rare, the richest practice was the hardest to code, and the projects available for the coded analysis lean toward community and open-source work whose cultivation looks different from that of a solo artist. The study therefore did not so much disconfirm the intuition as fail to find its mechanism in the only traces a public platform makes available. That outcome fits the intuition's own logic: if the decisive practices are relational, they should be largely unreadable from public posts, which is what was found.

For independent creators the practical reading is twofold. The relationship a creator builds is measurable, and a creator can in principle track whether supporters are staying rather than only whether a campaign succeeded. But the part of cultivation that public metrics reward, the visible cadence and content of posting, is not what separates creators who keep an audience from those who do not, at least not on this platform and not once quality is accounted for. If cultivation matters in the way creators believe, it does so through channels that neither a platform dashboard nor an outside analyst can see. This shifts a creator's attention from performing visible activity toward the harder, less visible work of relationship.

The finding also speaks to the difference between gatekeeping and mediation. Direct-support platforms remove the gatekeeper but not the mediation, and the part of the creator–audience relationship that this study could measure is the creator-owned persistence of support, not the relationship in full. The absence of an observable cultivation effect is itself a reminder that the most consequential parts of a non-gatekept relationship may live precisely in the spaces platforms do not record.

Several limitations bound these conclusions. First, the cultivation measures are observable behaviour, not the invisible relational practice the intuition names; the study measures the shadow, not the object. Second, because those same invisible practices could shape both what a creator posts and whether supporters stay, the associations reported here are descriptive and are not claims about cause. 

Third, the population was broadened from cultural creators to independent creators and community projects because the usable creative population was too small to stand alone; cultural production is the lens and a described subset, not the full analytical population. Fourth, updates posted on this platform is sparse, so the best-attested cultivation signal from prior work had little variation here. Fifth, coding reliability is the agreement between two coding passes rather than agreement with a second human coder, and four of eight categories fell below the reliability threshold and were confined to exploratory use. Sixth, supporter tenure is estimated from contribution records rather than read from a per-payment ledger, a robust approximation rather than an exact measurement. Seventh, the sample requires established projects, which tilts it toward collectives that already persisted. Finally, the platform still mediates through fees and discovery, so the measure captures creator-owned, non-gatekept persistence rather than relationship under an absence of mediation.

This study set out to take seriously a practitioner intuition, that independent creators sustain themselves by cultivating an audience rather than by talent alone, and found that the intuition cannot be confirmed from the public surface of a direct-support platform, while contributing a measure that makes the underlying relationship visible in numbers for the first time. The sustained audience relationship score distinguishes a lasting relationship from a one-time funding event and is simple to compute from open data. Using it, observable cultivation behaviours did not explain sustained relationships beyond the quality of the work, a result best read as evidence that the decisive practices are relational and largely unreadable from public artefacts. The most direct route forward is the close-up, qualitative study of cultivation that public data cannot substitute for: observing, in the exchanges between creators and supporters, the relational work whose consequences a measure of outcomes can detect but whose content it cannot. The measure offered here gives that future work an outcome worth explaining.

S.L.L.: conceived the study, assembled and analyses the data, produced the figures, and drafted the manuscript. V.L: provided methodological guidance and scientific supervision throughout and critically revised the manuscript for important intellectual content. Both authors read and approved the final version and agree to be accountable for the work.

The authors gratefully acknowledge the Open Collective community and its member projects for maintaining the public financial transparency data on which this study depends. The authors also thank the anonymous reviewers and the editorial team of the Canadian Journal of Business and Information Studies for their careful reading and constructive suggestions, which improved the clarity and the presentation of this article. No external assistance, financial or technical, was received beyond that acknowledged here.

The authors declare that they have no conflicts of interest, financial or non-financial, with respect to the research, the authorship, or the publication of this article.