AI Revenue Attribution: Solving the Scaling Problem

What AI revenue attribution actually means

AI revenue attribution is the challenge of showing whether an AI system contributed to business outcomes such as pipeline, closed-won revenue, retention, expansion, or sales velocity. In practice, many organizations can see AI usage but cannot connect that usage to measurable commercial lift, leaving AI stuck as a pilot rather than an operating capability [1][20].

A useful way to think about the problem is simple: adoption tells you whether people touched the tool, while attribution tells you whether the tool changed a business result. Those are related, but they are not the same. A team can use AI every day and still fail to improve conversion or revenue if the output does not affect the buying journey, the offer, or the handoff logic [11][20].

That is why AI revenue attribution is less about proving a perfect causal chain and more about building a decision-ready picture of contribution.

Why AI usage does not automatically translate into revenue

AI usage is only a behavior signal, not a revenue signal. A content assistant, a lead-scoring model, or a sales copilot can be used widely and still fail to improve outcomes if it is not tied to a meaningful business step.

This distinction is often collapsed into a broader confusion between activity and impact. Activity metrics tell you whether employees used the tool. Impact metrics tell you whether the tool influenced the outcome. In many programs, adoption dashboards look healthy while commercial results remain flat because the system is instrumented to track usage rather than decision quality or downstream lift.

The same logic applies to AI programs judged solely by volume. More drafts, more prompts, or more logins do not necessarily mean more revenue. They only mean the workflow has more AI inside it.

Efficiency gains are not the same as demand creation

A second source of confusion is the difference between efficiency and growth. Efficiency gains reduce the time, cost, or labor required to complete work. Demand creation increases the volume or quality of buyer intent. Those are not the same thing.

An AI system may make a team faster without making the market more willing to buy. Faster output can be helpful, but speed by itself is not proof of commercial lift. The business question is always whether the faster work actually changes buyer behavior or improves the economics of the funnel.

This matters because CFOs and revenue leaders do not fund every AI use case in the same way. If AI only reduces effort, it belongs in an efficiency or operations budget. If AI improves pipeline or conversion, it can justify revenue investment. At a strategic level, attribution is not optional.

Automation is not growth

Automation removes manual steps. Growth increases commercial output. Many AI initiatives deliver the first without achieving the second.

A workflow can be more automated and still produce the same sales results. That is why “we automated it” is not a sufficient business case. The organization still needs proof that automation changes a buyer outcome, not just an employee task.

This is the core operating mistake behind many AI rollouts: the tool is added to an unchanged process. When AI is bolted onto the existing workflow rather than redesigning decision-making, the organization often gains incremental efficiency but little measurable change in revenue [5].

Strong AI programs need proof, not assumptions

Unmeasured AI programs remain experiments because no one can justify their budgets with evidence. The business may like the tool, but liking a tool is not the same as proving value.

That is especially true when a program cannot show a baseline, a comparison, or a revenue proxy. In those cases, leaders are left with intuition instead of evidence. The result is familiar: AI gets discussed in enthusiastic terms, but it does not survive planning cycles, finance review, or cross-functional debate [6].

This is also why many AI efforts stall after the pilot stage. They are not rejected outright; they simply fail to be included in the operating budget because the evidence is too soft to support scaling.

Why companies fail to see AI increase revenue

Companies usually fail to see AI increase revenue because they frame it as a productivity initiative, launch disconnected pilots, and measure outputs instead of commercial outcomes. When ownership is unclear and data is fragmented, even useful AI work can be hard to translate into credible revenue proof [7].

They optimize for cost-cutting instead of revenue lift

Many teams adopt AI to reduce headcount pressure, shorten task duration, or avoid hiring. Those are valid goals, but they do not automatically create revenue.

If the success metric is hours saved, the organization will naturally optimize for efficiency. If the metric is pipeline created, it will optimize for growth. The problem is not that efficiency is bad; it is that efficiency and growth need different measurement systems [8].

This is why AI often looks successful to operations teams and disappointing to the board. Both views can be true at once, but they are not interchangeable. The operating model must define which outcome matters before deployment begins.

They launch disconnected pilots with no path to scale

Disconnected pilots are common because they are easy to approve and hard to govern. A marketing team tests content generation, sales tests meeting summaries, and customer success tests ticket drafting. Each team gets a small win, but no one is responsible for connecting those wins into a repeatable commercial system [9].

This is the classic pilot trap. The organization accumulates local gains but never creates a repeatable mechanism for enterprise-wide revenue lift. If a shared owner does not exist across marketing, sales, RevOps, customer success, and finance, the program will usually remain a set of isolated experiments rather than a scalable capability.

They measure activity, not commercial outcomes

Activity metrics include logins, prompt volume, draft counts, and content output. Commercial outcomes include pipeline created, conversion rate, average deal size, renewal rate, and expansion revenue. If the dashboard stops at activity, leadership has no basis for deciding whether AI is working financially [10][20].

This is a common reporting failure. Teams present adoption graphs because they are easy to create and flattering to share. But adoption is not a value. A tool used by most of the team can still produce zero lift if it is not embedded at the right decision point.

They lack clean data, aligned ownership, and shared definitions

Attribution fails when the CRM, marketing automation platform, product analytics, and finance systems disagree about what a lead, opportunity, customer, or renewal means. If teams use different definitions, they will also produce different versions of success [11][20].

Shared definitions matter because attribution is as much a governance problem as a measurement problem. Revenue operations, finance, and marketing must agree on what counts as sourced revenue, influenced revenue, assisted conversion, and expansion lift. Without that alignment, the organization ends up debating spreadsheets instead of scaling systems.

Why revenue attribution is hard for AI

Revenue attribution is hard for AI because it spans multiple touchpoints, revenue arrives with a delay, and the buying journey is shared across teams and channels. Unlike a single campaign click, AI usually influences decisions indirectly, which makes causal proof more difficult but not impossible [12][20].

AI influences multiple touchpoints, not one clean source

AI often touches discovery, qualification, messaging, prioritization, follow-up, and retention. That means its effect is distributed across the funnel rather than concentrated in a single event. A lead scored by AI may be better routed, better nurtured, and better closed, but none of those outcomes alone captures the full influence [13].

This is why simplistic dashboards fail. They look for a single channel, source, or event to receive credit. AI does not work that way. Its value is frequently embedded upstream, in better choices made earlier in the journey.

Revenue is delayed, multi-causal, and often shared across channels

Revenue is rarely the product of one cause. Buyers may interact with several touchpoints before a deal closes, and AI may improve one or more of those steps while still sharing credit with the rest of the system [14].

Delay makes the problem harder. If an AI intervention in Q1 drives closed-won revenue in Q3, short reporting windows will miss the effect. That is why revenue attribution needs both immediate proxies and longer-horizon measurement, especially in B2B SaaS and agency environments where buying cycles are longer.

Attribution gets messy across marketing, sales, product, and service

AI does not stay inside one department. Marketing may use it for content and scoring; sales for sequencing and qualification; product for in-app guidance; and customer success for expansion and support. Each function sees only part of the story, making total attribution difficult unless the data model is shared [15].

The practical implication is simple: the AI system must be tied to a single commercial objective, even if multiple teams participate. That reduces the common “everyone influenced it, therefore nobody owns it” problem that often kills scale.

Why perfect attribution is the wrong standard

Perfect attribution is impossible in most commercial systems because buyers behave nonlinearly and revenue is shared across multiple variables. The useful standard is not perfect certainty; it is enough evidence to make a better decision than guesswork [16][18].

This matters because teams often use the demand for precision as an excuse for inaction. If AI cannot be attributed perfectly, they conclude it cannot be attributed at all. That is a false binary. The right question is whether the measurement is good enough to tell you what to repeat, refine, or stop.

Common attribution traps

Last-click attribution overcredits the final interaction and hides upstream influence. Vanity metrics make AI look busy without proving revenue. Overclaiming happens when a team assigns all improvement to AI simply because it was present in the workflow [17].

Those traps damage credibility. Once a team overstates impact, finance and executive leadership become skeptical of future AI claims. Credible attribution is conservative, repeatable, and specific about what the AI did and did not influence.

The operating model problem behind low AI ROI

The low ROI problem is usually an operating model problem, not a model quality problem. AI fails to scale when it is treated as a side project instead of a workflow redesign, because teams, incentives, and ownership structures remain unchanged [18][19].

AI as a side project versus AI as a workflow change

A side project adds a tool to an existing process. A workflow change reorganizes work, so the tool changes how decisions are made. The second model is harder, but it is the only one that scales reliably because it modifies the system rather than decorating it [20].

This is where most AI investments stall. Teams buy software, write policies, and run a few tests, but they never redesign the surrounding workflow. As a result, AI becomes another tab in the browser rather than a revenue infrastructure layer.

Siloed teams and unclear accountability

If marketing owns the AI tool but sales owns the pipeline target, neither team fully owns the outcome. That split creates reporting ambiguity and slows decision-making. Clear accountability means one named owner is responsible for the business result, not just the implementation [20].

This is especially important in enterprise B2B, where multiple teams influence the buyer journey. The AI adoption process becomes easier to manage when ownership spans revenue functions rather than being confined to a single, isolated team.

Broken handoffs across marketing, sales, and customer success

AI can improve handoffs, but only if the handoffs themselves are measurable. For example, AI-assisted lead qualification is limited if sales ignore the score, and AI-assisted expansion recommendations are wasted if customer success does not act on them. The systems must connect end-to-end [[21]].

Broken handoffs are one of the highest hidden costs in revenue organizations. They create friction, delays, and inconsistent follow-up, which weaken attribution and performance. Better AI does not fix bad handoffs on its own; it exposes them faster.

Incentives that reward output instead of revenue impact

When teams are rewarded for content volume, call volume, or ticket closure volume, they will optimize for output. AI then amplifies the wrong behavior. To scale revenue, incentives must shift toward revenue impact, conversion quality, or retention outcomes [22].

This is why many AI programs fail even when the tooling works. The team complies with the workflow, but the incentive system points elsewhere. AI can only scale when it is embedded in the same success criteria that leadership uses to allocate capital.

One practical framework for AI revenue attribution

A useful framework has five steps: define the use case, choose a revenue proxy, establish a baseline, compare against a control or prior period, and combine quantitative lift with qualitative proof. The goal is not perfect certainty; it is a measurement system that is good enough to support investment decisions.

Step 1: Define the use case and the business outcome

Start by naming the exact job the AI is meant to do. For example, AI might improve lead scoring, sales prioritization, quote generation, knowledge retrieval, retention outreach, or content creation. Then define the outcome it is expected to influence, such as pipeline creation or cycle time reduction [[24]].

This step prevents vague success criteria. “We use AI in marketing” is not measurable. “We use AI to improve lead-to-opportunity conversion in mid-market SaaS” is measurable. The more discrete the use case, the easier it is to attribute impact.

Step 2: Choose the right revenue proxy

A revenue proxy is a metric that connects the AI action to a commercial outcome before revenue fully materializes. The proxy should be close enough to revenue to be meaningful, but early enough to help the team make decisions quickly [[25]].

Pipeline creation

Use pipeline creation when AI helps generate more qualified opportunities. This is a strong fit for demand intelligence, account targeting, and lead scoring systems. Track the volume and quality of pipeline created in the AI-assisted segment versus the baseline [[26]].

Conversion rate

Use conversion rate when AI improves progression from visit to lead, lead to opportunity, or opportunity to closed-won. This is useful for chat, personalization, recommendations, and sales enablement interventions that affect decision points [[27]].

Average deal size

Use average deal size when AI improves upsell strategy, package selection, or pricing guidance. If a model helps identify better-fit accounts or shape stronger offers, deal size can reveal impact even when volume remains stable [[28]].

Retention and expansion

Use retention and expansion when AI supports customer success, support, or account management. Churn reduction and expansion revenue are especially relevant in subscription businesses where the commercial value of existing accounts is often larger than that of new acquisitions [[29]].

Sales velocity

Use sales velocity when AI reduces time to close, shortens stage duration, or improves follow-up timing. Faster cycles matter because they reduce cost and increase throughput, which can often be measured before total revenue is fully visible [[30]].

Step 3: Establish a baseline before deployment

A baseline is the pre-AI performance level against which lift is measured. Without one, teams cannot know whether the new system improved anything. Baselines should be captured for the same segment, team, campaign, or workflow before AI goes live [[31]].

This seems obvious, but many teams skip it in the rush to launch. That mistake makes later claims fragile. If you do not know what normal looked like, you cannot prove the shift. Baselines should include volume, conversion, cycle time, and revenue proxy data.

Step 4: Compare AI-assisted performance to a control group or prior period

The strongest evidence comes from a control group, holdout, or A/B design. When that is not feasible, compare AI-assisted performance to a matched historical period or a similar non-AI segment. The goal is not laboratory purity; it is credible comparison [[32]].

This is where experimentation and revenue attribution intersect. The point is not to assign flawless credit to every experiment. The point is to use the best available comparison so leaders can make a better investment decision than they could with guesswork alone.

Step 5: Track lift by segment, workflow, and team

Lift should be segmented, not averaged away. AI may improve performance in enterprise accounts but not in SMB accounts, or in one campaign but not another. Tracking by segment helps identify where the model actually works and where it should be retired or retrained [[33]].

This is one of the most practical ways to avoid broad overclaiming. If one team has a strong result, do not generalize it blindly across the company. Use segmented reporting to determine where the signal is sufficiently real to scale.

Step 6: Combine quantitative lift with qualitative proof

Numbers show lift; operators explain why it happened. Qualitative proof from sales reps, marketers, and customer success managers helps verify whether AI changed judgment, prioritization, or workflow quality. That combination is more credible than either source alone [[34]].

A revenue model is easier to trust when users can explain how it changed their work in practical terms. In that sense, attribution is not only a measurement exercise but also a management exercise.

What to measure if you want AI to scale

To scale AI, measure executive revenue outcomes and operational signals that show whether the initiative can be repeated reliably. The right metric set includes both lagging indicators, such as revenue, and leading indicators, such as adoption, quality, and workflow speed [[35]].

Revenue metrics that matter to executives

Executives care most about metrics that connect directly to the P&L. If AI cannot affect one of these, it will struggle to survive the budget cycle. Revenue metrics should be selected by use case, not by habit [[36]].

Net new pipeline

Net new pipeline is the clearest early commercial signal for many go-to-market AI use cases. It shows whether AI is helping the organization create more qualified opportunities, not just more raw activity [[37]].

Closed-won revenue

Closed-won revenue is the strongest proof of direct value, but it is also the slowest to show. It is ideal for mature programs where the AI intervention has sufficient scale and where the buying cycle is stable enough to measure [[38]].

Expansion revenue

Expansion revenue is especially useful for customer success and account management AI. It captures upsell, cross-sell, and growth within the installed base, which often provides a more efficient return than new acquisition [[39]].

Retention or churn reduction

Retention and churn reduction are critical in subscription businesses. If AI improves support resolution, proactive outreach, or risk prediction, the most meaningful result may be preserving revenue rather than creating it [[40]].

Sales cycle reduction

Shorter sales cycles increase throughput and often improve cash conversion. If AI helps prioritize accounts, improve follow-up, or automate admin, cycle time can be a credible leading indicator of future revenue lift [[41]].

Operational metrics that show whether scaling is possible

Operational metrics matter because a program cannot scale if it is too fragile, cumbersome, or dependent on heroics. These metrics show whether the AI workflow is repeatable across teams and time [[42]].

Time saved per workflow

Time saved is useful when it is tied to a revenue-critical task, not just a convenience task. Saving 20 minutes on a low-value activity is less important than saving 20 minutes on lead routing or deal follow-up [[43]].

Adoption by role or team

Adoption by role reveals whether the right people are actually using the system. A high adoption rate among junior staff but low adoption by decision-makers can indicate shallow utility rather than true workflow change [[44]].

Task completion rate

Task completion rate tells you whether AI is helping users finish the work, not just start it. This matters in sales and marketing workflows where partial completion can create hidden leakage [[45]].

Output quality and consistency

Quality and consistency are essential in customer-facing work. AI that produces more output but lowers message quality, accuracy, or brand fit can reduce revenue even as it increases volume [[46]].

Cost per revenue outcome

Cost per revenue outcome is one of the most useful scale metrics because it connects investment to actual business return. It helps answer the question of whether the AI system is becoming more efficient as it grows [[47]].

Leading indicators versus lagging indicators

Leading indicators predict future performance, while lagging indicators confirm results after the fact. A scalable AI program needs both because leadership must manage today’s funnel while waiting for end-revenue confirmation [[48]].

In practice, leading indicators include adoption, conversion proxies, and cycle time. Lagging indicators include pipeline, closed-won revenue, retention, and expansion. If a program relies solely on lagging indicators, it will be too slow to manage. If it only has leading indicators, it will be too weak to trust.

The scaling problem: why proof matters more than hype

The scaling problem is not that AI cannot work in one pilot. It is that organizations cannot prove enough value to justify repeating it across the business. Without proof, the budget stalls, the sponsorship weakens, and the program remains isolated [[49]].

Why teams stop after the pilot phase

Teams stop after the pilot phase because pilot results are often anecdotal, localized, and difficult to generalize. One success story does not prove maintainability, repeatability, or cross-team fit. When ownership and measurement are weak, pilot momentum eventually fades [[50]].

This is the point where many AI initiatives quietly die. They are not rejected; they are simply not renewed. The organization treats them as interesting but not essential.

How a lack of proof kills budget and sponsorship

Budget follows evidence. Sponsorship follows confidence. If leaders cannot see a credible revenue path, they will not allocate meaningful resources to scaling. In multi-quarter planning, “it seems useful” is not as persuasive as “it improved conversion by X% in this segment” [[51]].

That is why revenue attribution is central to scale. It translates experimentation into budget language. Without it, AI adoption competes with every other initiative that can show a clearer business impact.

Why leadership needs evidence before standardizing AI

Standardization means changing training, governance, process, and sometimes headcount design. Leaders will not standardize a tool that has only produced isolated wins unless the gains are measurable and repeatable. Evidence lowers the perceived risk of organizational change [[52]].

This is especially important in mature businesses, where the cost of widespread change is high. AI has to earn its place in the operating model. Proof is the price of entry.

Why “works in one team” does not equal “ready for enterprise scale”

A tool can work in one team because that team has better data, stronger managers, or simpler workflows. Enterprise scale introduces variation in process maturity, data quality, customer segment, and compliance requirements. What works locally can fail globally [[53]].

That is why scaling requires segmentation, governance, and workflow instrumentation. A local win is promising; scalable proof is different.

How to build a revenue-attributed AI program that can scale

Build the program by starting with one revenue-critical use case, naming an owner, instrumenting the workflow, sharing a common dashboard, and regularly reviewing lift. Scale should follow proof, not ambition [[54]].

Start with one revenue-critical use case

Pick a use case that already affects money, not one that merely looks impressive. Good candidates include lead scoring, deal prioritization, renewal risk prediction, or campaign content tied to measurable demand. The goal is to secure a clean commercial story early [[55]].

Tie each use case to a named owner and KPI

Every use case should have one accountable owner and one primary KPI. That owner may coordinate across teams, but accountability must be clear. This reduces ambiguity and speeds up decision-making when results need to be actioned [[56]].

Instrument workflows so AI actions are trackable

Track the AI action, the human response, and the commercial consequence. If the system recommends a next step, log whether the recommendation was accepted and what happened next. This creates an evidence trail that is much more useful than tool usage logs alone [[57]].

Create a shared dashboard for marketing, sales, ops, and finance

A shared dashboard keeps every function aligned on the same definitions and outcomes. Marketing can see lead quality, sales can see conversion and velocity, operations can see process adherence, and finance can see revenue impact. Shared visibility prevents attribution arguments from fragmenting the program [[58]].

Review lift regularly and retire weak use cases quickly

Not every AI use case deserves to survive. Review lift on a regular cadence and remove programs that do not show repeatable value. Quick retirement is a feature, not a failure, because it protects resources for stronger opportunities [[59]].

Reinvest in the use cases with the clearest revenue signal

Once a use case shows repeatable lift, reinvest to expand scope, coverage, or workflow depth. This is how AI becomes infrastructure. The program becomes compounding rather than episodic [[60]].

Common mistakes that block AI revenue scale

The most common mistakes are buying tools before defining the problem, measuring usage instead of value, involving finance too late, treating attribution as a reporting task, and scaling outputs before proving lift. Each mistake weakens credibility and delays growth [[61]].

Buying tools before defining the business problem

Tool-first buying leads to searching for use cases after the fact. That usually leads to a weak fit and poor adoption. The business problem should define the tool, not the other way around [[62]].

Measuring usage instead of commercial value

Usage is easy to track, which is why it is overused. But a tool can be heavily used and commercially irrelevant. Always connect usage to a revenue proxy or a protected cost-outcome target [[63]].

Letting finance and operations get involved too late

If finance and operations are brought in after launch, they will often dispute the assumptions, definitions, or governance model. Early involvement helps set the baseline, choose the proxy, and preserve credibility when results are reviewed.

Scaling outputs before proving lift

More output is not the same as more impact. Scaling a content engine, for example, can create a lot of material without improving conversion. Before expanding, make sure the workflow has proven it can move a business metric that leadership cares about.

FAQ

What is AI revenue attribution?

AI revenue attribution is the process of showing whether an AI system contributed to outcomes such as pipeline, closed-won revenue, retention, expansion, or sales velocity. It does not require perfect certainty, but it does require enough evidence to support better business decisions.

Why is revenue attribution harder for AI than for a single campaign?

AI usually influences several steps in the buying journey rather than one isolated event. That makes its contribution more distributed and harder to isolate, especially when multiple teams, channels, and time delays are involved.

What is the difference between activity and impact?

Activity measures usage, such as logins, prompts, or drafts. Impact measures business change, such as conversion rate, pipeline created, average deal size, retention, or expansion revenue.

What should leaders use as a baseline?

Leaders should use pre-AI performance for the same workflow, segment, team, or campaign. Useful baselines often include volume, conversion, cycle time, and the selected revenue proxy.

What is the best way to prove AI value?

The strongest approach is a control group, holdout, or A/B design when possible. If that is not feasible, compare AI-assisted performance to a matched historical period or similar non-AI segment, then combine the numbers with operator feedback.

Which metrics matter most?

The metrics depend on the use case, but common executive measures include net-new pipeline, closed-won revenue, expansion revenue, retention, churn reduction, and sales-cycle reduction. Operational measures such as adoption, task completion, and output quality help indicate whether the program can scale.

Does AI have to be attributed perfectly to be useful?

No. Perfect attribution is rarely possible in commercial systems. The goal is consistent, defensible evidence that is strong enough to guide investment, governance, and scaling decisions.

Conclusion

AI revenue attribution is not about forcing perfect precision onto complex business systems. It is about proving, with enough discipline, whether AI changes commercial outcomes in ways that matter to the organization. The programs that scale are usually the ones that connect a clear use case to a meaningful revenue proxy, establish a baseline, and measure lift consistently across teams.

For leaders who want the operational side of that work translated into a practical program, Multiplier AI is one example of a team-oriented approach to workflow instrumentation and revenue-focused measurement. The broader principle, however, is simple: if AI cannot be tied to a business outcome, it will usually remain a useful experiment rather than a scalable capability.

References

1. https://davidleemannheim.medium.com/no-you-cant-accurately-attribute-nor-forecast-revenue-to-experimentation-here-s-why-a22b1ac2a5c7
2. https://www.ssonetwork.com/finance-accounting/columns/the-biggest-obstacle-to-scaling-any-business
3. https://www.youtube.com/watch?v=tUA59ZRqf9o
4. https://www.facebook.com/SkylarLewisOC/posts/most-business-owners-dont-have-a-scaling-problemthey-have-a-complexity-problemyo/27056705880606821/
5. https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/superagency-in-the-workplace-empowering-people-to-unlock-ais-full-potential-at-work
6. https://www.demandgenreport.com/demanding-views/how-to-get-real-about-revenue-attribution/6902/
7. https://pmc.ncbi.nlm.nih.gov/articles/PMC10065758/
8. https://www.linkedin.com/posts/tonyaturrell_scaling-isnt-always-a-strategy-problem-activity-7477922574027554816-nLeC
9. https://blog.blueprintprep.com/lsat/flawctober-the-absence-of-evidence-fallacy/
10. https://www.linkedin.com/posts/cory1_95-percent-of-organizations-arent-seeing-activity-7394893344562262016-5r8w
11. https://www.linkedin.com/pulse/marketers-need-stop-obsessing-over-revenue-misha-abasov
12. https://www.pbs.org/wgbh/frontline/article/artificial-intelligence-work-jobs-robots-v-humans/
13. https://www.craftliterary.com/2025/03/26/show-dont-tell-what-ai-cant-do/
14. https://www.reddit.com/r/cscareerquestions/comments/1muu5uv/mit_study_finds_that_95_of_ai_initiatives_at/
15. https://www.instagram.com/reel/DT2wFJ2E7_u/?hl=en
16. https://www.reddit.com/r/DebateReligion/comments/dx4ux4/absence_of_evidence_is_not_evidence_of_absence_is/
17. https://www.reddit.com/r/ArtificialInteligence/comments/1q671wf/why_cant_ai_tell_us_how_it_works/
18. https://en.wikipedia.org/wiki/Evidence_of_absence
19. https://www.paulgraham.com/ds.html
20. https://dealhub.io/glossary/revenue-attribution/