Marketing budgets in the world of LLM: how to plan for AI revenues?

In this article, you will find a comprehensive analysis of the role of AI channels in the P&L structure, including an overview of the main types of solutions from chatbots to pricing systems and their direct impact on revenue, margin and CAC costs. You'll also learn why classic forecasting fails with learning models, and how to implement a 5-step budget planning framework based on specific use cases and measuring revenue incrementality with tools like Semly.

What are AI channels in marketing and sales from a P&L perspective

In practice, "AI channels" are not a single, abstract revenue stream, but several classes of solutions that "plug into" the existing marketing and sales funnel.

Main types of AI channels

• Chatbots and conversational assistants (conversational commerce, customer service bots)

• Sites of operation: site chat, WhatsApp, Messenger, IG DM, in-app chat, voice bots.

• Features:

1. answering pre-purchase questions,
2. guiding through the product selection and ordering process,
3. handling order statuses, returns, complaints.

Meta's research shows that companies using conversational assistants on WhatsApp and Messenger report higher conversions and shorter purchase paths in mobile commerce.

Features:

• scoring and prioritization of leads,
• automatic call summaries and preparation of follow-ups,
• cross-sell/upsell recommendations.

Gartner estimates that by 2027, 95% of sales research processes will be started using AI, and GenAI implementations could increase sales productivity by 25% and revenue by about 12-13%.

Generative content marketing (LLM + generative AI)

• Applications: SEO, performance advertising, email, social media, landing pages.

• Features:

1. quickly create multiple variants of creations,
2. copy personalization for segments,
3. a/B and multivariate test automation.

• Business effect: more experimentation, better message alignment and higher CTR and CR, which translates into revenue. In e-commerce, this advantage is further enhanced by content created under LLM - product descriptions and FAQs.

Product recommendation systems and 1:1 personalization

• Applications: e-commerce (website, app), email, push, onsite banners.

• Features:

1. "next best product/offer"
2. personalized listings and order of products
3. dynamic product bundling.

• McKinsey indicates that advanced personalization can increase revenue by 5-15% and increase the effectiveness of marketing spending by 10-30%.

Autonomous performance campaigns (AI campaign management)

• Applications: Google Ads, Meta Ads, programmatic, retail media, marketing automation.

• Features:

1. automatic budget allocation between channels and campaigns,
2. optimization of rates and creation under KPIs (ROAS, CAC, revenue),
3. continuous testing of hundreds of combinations.

• Case studies show CAC reductions of up to 30% and significant increases in ROI from campaigns.

AI in pricing and revenue management

• Applications: dynamic pricing, promotions, margin management.

• Features:

1. optimizing prices for maximizing revenue or margins.
2. automatic adjustment of discounts to price sensitivity.

• Revenue management class tools (e.g., PROS) communicate documented incremental revenue with AI.

AI for analytics, prediction and attribution (marketing/revenue analytics)

• Applications: all channels - the role of the "meta-layer" over marketing and sales activities.

• Features:

1. predictive CLV, churn and propensity to buy models.
2. marketing mix modeling (MMM).
3. multichannel attribution, incrementality measurement.

• Tools in this class - such as Semly - make it possible to estimate incremental revenue from AI channels and measure sales generated by AI search and other traffic sources, and optimize the allocation of budgets under revenue, not just clicks.

Where AI channels "touch" P&L

• Revenue (Revenue): higher conversion, higher average basket (AOV), more frequent purchases, higher CLV.

• Acquisition Costs (CAC): better targeting and campaign optimization lower the cost of customer acquisition.

• Operating Expenses (OPEX): automation of customer service and sales reduces the cost of service, but requires investment in data and integrations.

• Margin: AI in pricing and managing promotions allows raising margins while maintaining or increasing volume.

Why classic revenue forecasting doesn't work directly for AI

Transferring the logic of "let's add a new channel in Excel" often ends up either overestimating or underestimating the effect of AI. The reasons are threefold.

High variability and learning effect of models

AI channels are not static:

• models learn from data and user behavior,
• the results after 2-3 weeks can be dramatically different than in the first days,
• performance increases by leaps and bounds when new prompts, data or integrations are introduced.

Forecasting "rigidly" (one CR, one AOV) is a dangerous illusion - what is realistically needed is forecasting in scenarios and with "learning" periods.

Attribution problem: AI as a "coprocessor," not a separate channel

AI often runs in the background:

• improves the copy of the campaign, so the results of existing channels (Google, Meta).
• recommends products, but sales are attributed to "email" or "organic search."
• qualifies leads, but revenue appears as "Direct" or "Sales."

In classic last-click attribution, most of the AI value "dissipates" to other channels. Therefore, without advanced attribution tools and incrementality models, management will only see part of the effect.

Risk of hype and underestimation of organizational costs

Reports from BCG and Deloitte indicate that:

• only a minority of AI projects go beyond the pilot and reach scale with a positive ROI

• TCO (Total Cost of Ownership) of AI projects is often 40-60% higher than assumptions, and 60-80% of the effort is data work

If management only plans for licensing and "rapid deployment" costs, and ignores data, integrations, process change and competencies, the ROI forecast will be systematically inflated.

AI channel revenue forecasting framework for management

Management needs a simple, repeatable approach that can be applied to various AI use cases. Below is a proposal for such a framework.

Step 1 Define the CONCRETE use case and funnel stage

Instead of "we are investing in AI," be more specific:

• Use case: e.g., product recommendations in e-commerce, AI chatbot on a product website, sales assistant in B2B.

• Funnel stage: acquisition, conversion, retention, upsell.
• Key KPIs: CR, AOV, CLV, churn, SQL/MQL count, etc.

Step 2 Determine the scenarios: conservative / baseline / aggressive

For each use case, plan three scenarios for 12-24 months:

• Conservative: low adoption, moderate uplift (e.g. +3 p.p. CR).
• Baseline: in line with benchmarks and market experience (e.g., +5-7 bps CR).
• Aggressive: optimistic, but still realistic (e.g., +10 p.p. CR, more traffic coverage).

Step 3 Counting the "model on a napkin" - numerical example

Let's assume you are running an e-commerce business with the following parameters (monthly):

• Traffic: 500,000 sessions,
• Current conversion (CR): 2,0%,
• Average basket value (AOV): 58,13 EUR
• Gross margin: 40%.

Implementing AI product recommendations on the website and email has:

• cover 60% of the traffic (300,000 sessions),
• in the baseline scenario, raise the CR by +0.5 pp. (from 2.0% to 2.5%).

Current status (without AI - only covered traffic):

• Orders = 300,000 × 2.0% = 6,000,
• Revenue = 6,000 × 58.13 EUR = 348,750 EUR
• Margin = 348,750 × 40% = 139,500 EUR.

Status after AI implementation (baseline scenario):

• Orders = 300,000 × 2.5% = 7,500,
• Revenue = 7,500 × 58.13 EUR = 435,938 EUR
• Margin = 435,938 × 40% = 174,375 EUR.

Incremental monthly effect (only on covered traffic):

• Additional orders: 1 500,
• Additional revenue: 87,188 EUR
• Additional margin: 34,875 EUR.

If the cost of an AI solution (tool + integrations + maintenance) is 13,950 EUR per month, the incremental gross profit (before other costs) is ~20,925 EUR/month. Scenarios can be scaled (coverage of 80-100% of traffic, different levels of uplifts) and build a revenue "corridor" in them.

Step 4 Incorporate incrementality and control groups

The "on a napkin" model is based on an assumed uplift. To make it more realistic:

• conduct A/B testing with a control group (e.g., 80% of users see AI recommendations, 20% do not).
• after 4-8 weeks, compare CR, AOV and margin between groups.
• count incremental revenue as the difference.

It is this level of incremental revenue that budget analytics should adopt - and this is where the role of tools such as Semly comes in, helping to integrate test data, count incremental revenue and translate it into budget scenarios.

Step 5 From use cases to AI portfolio

At the board level, individual experiments are not budgeted for, only a portfolio:

• 3-5 priority use cases AI
• each with three revenue and cost scenarios

• all bundled in a budget model (12-24 months) with visible contribution to revenue, impact on CAC and CLV, and payback period for each use case.

Here, Semly - as an analytics and distribution platform - can act as a "single source of truth": pinning data from AI and traditional channels into a single revenue model, which is especially crucial in e-commerce building competitive advantage through AI.

How to plan marketing budgets in the world of LLM

1. Split budget for R&D vs "core performance"

In practice, the division works:

• R&D / AI experiments (5-15% of marketing budget) - pilots, POCs, testing of new use cases.
• Core performance (the rest of the budget) - activities with an established ROI, including scaled AI channels that have passed the pilot phase.

"High performers" companies, according to McKinsey, often allocate more than 20% of their technology budget to AI. In marketing, some of that budget "comes back" in the form of media spending and AI-optimized tools.

2. Think in the logic of a "portfolio of tests," not one big bet

Instead of a large, one-time investment in "one AI project."

• build a roadmap of 3-5 use cases with different risk and potential profiles,
• each use case has a pilot (3-6 months), success criteria (uplift, incremental revenue, ROI) and a scaling plan (if it works).

3. Agree with CFO on 12-24 month horizon

• Deloitte and McKinsey stress that a realistic return horizon for larger AI investments is 12-24 months.
• Pilots should show "evidence of value" in 1-2 quarters, but full ROI requires time to scale and improve models.

4. Example of AI budgeting scheme for 12 months

Let's assume that the annual marketing budget is 4,650,000 EUR.

• AI experimentation fund - 10% (465,000 EUR): 3-4 pilots (e.g. recommendations, chatbot, sales assistant, AI content). Each with its own P&L (cost + projected incremental revenue).

• Scaling successful use cases - 10-15% of the budget (465,000–697,500 EUR): Moving some spending from less effective performance channels to scaled AI initiatives. Data-driven decisions from tools such as Semly (real revenue impact).

• The rest of the budget - 75-80% (3,487,500–3,720,000 EUR): Classic channels (search, social, affiliates, offline), but increasingly managed and optimized by AI (campaign automation, bid management).

The role of data, attribution and tools like Semly

What management needs to see to trust AI forecasts

• Consistent view of revenue and costs per channel: revenue data (orders, subscriptions), media and operating costs, touchpoint data (including: exposure to chatbots, recommendations, AI content).
• Multichannel attribution and incrementality: classic last-click does not show the true value of AI; data-driven attribution models, marketing mix modeling (MMM) and experiments with control groups are needed.
• Revenue and CLV-oriented management dashboards: not just ROAS and CPC, but: incremental revenue per channel, impact on CAC and CLV, payback period per use case.

• consolidates data from traditional channels and new AI channels into a single model.
• helps measure the incremental impact of AI on revenue and key KPIs.
• allows you to build and monitor budget scenarios (what happens to revenue if you shift X% of the budget to AI channels).
• provides management with clear dashboards for discussions with the CFO and the board.

As a result, decisions to scale AI channels are not based on vendor promises or the "feel" of the team, but on hard data.

Cases and numerical examples: how AI channels translate into revenue

Example 1: e-commerce - AI product recommendations

Assumptions (baseline scenario):

• Additional monthly margin from AI recommendations (after deducting the cost of the tool): 20,925 EUR
• Scaling for more traffic per year (average 1.5× effect) → 31,388 EUR/month.
• Horizon: 12 months.

Annual incremental effect: about 377,100 EUR in additional margin.

If the initial investment in integration and implementation is 93,000 EUR, and annual operating costs are 167,400 EUR (licenses, maintenance), then:

• Total cost in year 1: 260,600 EUR
• Additional margin: 377,100 EUR
• ROI in year 1 ≈ 45%, payback period < 12 months.

Example 2: B2B SaaS - AI assistant for SDR (lead qualification)

Assumptions:

• The SDR team generates 1,000 MQLs per month,
• Current MQL → SQL = 20% (200 SQL/mc),
• SQL → won = 25%,
• Average annual revenue per customer (ARR) = 9,300 EUR.

Implementing an AI co-pilot improves the effectiveness of qualifications and follow-ups:

• MQL → SQL rises to 26% (+6 p.p.),
• SQL/mc = 260,
• Won/mc = 260 × 25% = 65 (vs. 50 previously),
• Additional 15 customers × 9,300 EUR = 139,500 EUR ARR per month.

Annual horizon (assuming parameters are maintained): An additional 1,674,000 EUR ARR. If the annual cost of the solution (tool + integrations + maintenance) is 348,750 EUR, the gross ROI is very high. In practice, some of the additional revenue will be consumed by sales costs and churn, but even after adjustment the ROI remains attractive.

Most common management mistakes when budgeting for AI in marketing

• Lack of clearly defined use cases and KPIs: "AI budget" is not enough. Specific targets are needed: e.g. +X p.p. CR, +Y zl incremental revenue, -Z% CAC.
• Too much or too little pilotage: Too large: difficult to control, no clear conclusions. Too small: statistically insignificant, difficult to generalize.
• Focus on technology implementation instead of process: AI without changes in processes (sales, service, content production) often misses potential.
• Ignoring data and integration costs: Underestimating TCO by 40-60% is the norm in AI projects without a mature approach to data.
• No central tool for monitoring effects (analytics and attribution): Without a tool like Semly, management sees scattered data and cannot reliably count incremental revenue, compare channel efficiencies or model budget scenarios.

Checklist for CEO/CMO/CRO: how to approach AI budgeting

• Define 3-5 key use cases of AI in marketing and sales (with funnel stage and KPIs).
• Make sure you have the data to measure their effect (revenue, costs, touchpoints, CLV).
• Establish revenue scenarios (conservative / baseline / aggressive) for 12-24 months.
• Design a pilot with a control group and clear criteria for success.
• Provide an attribution and analytics tool (e.g., Semly) that shows the incremental impact of AI on revenue.
• Agree with the CFO on the ROI horizon (typically 12-24 months) and evaluation rules.
• Budget AI as a portfolio of tests, not a single project; verify ROI of use cases quarterly.
• Incorporate AI channels into continuous management reporting (revenue dashboards, CAC, CLV, payback per use case).

FAQ: the most common management questions

1. How much of our marketing budget should we allocate to AI?

There is no universal number, but the practice of "high performers" suggests:

• 5-15% of the marketing budget for AI experiments and pilots,
• an additional 10-15% for scaling proven use cases,
• over time, AI will permeate the entire budget (campaign automation, personalization, attribution), so it's more about how than "how much."

2. When can we expect a return on investment in AI?

• The first signals of value from the pilots should appear within 3-6 months.
• A realistic full payback period for larger AI programs is 12-24 months.

3. How to avoid "AI-hype" and budget burn-through?

• Use case-first approach and clearly defined KPIs.
• Require experiments with a control group and incremental revenue measurement.
• Use tools like Semly to verify predictions vs. actual data, including in the context of new traffic sources from LLM-based search engines.

4. Do we need to have our own data science team to use AI in marketing?

Not always - many solutions provide ready-made models. On the other hand:

• you need the competence to understand, interpret and use the results.
• investment in data quality and an analytics platform that ties the entire ecosystem together is key.

Summary

AI channels - from chatbots to product recommendations and generative content to advanced analytics - can realistically increase revenue, lower CAC and raise CLV. Reports from McKinsey, Gartner, BCG and numerous case studies show double-digit increases in revenue and efficiency at companies that implement AI head-on.

The key for management, however, is not "how much we spend on AI," but how we link AI to budgets and revenue forecasts:

• defining specific use cases
• building revenue scenarios
• conducting pilots with measured incrementality
• integrating AI channels into budgeting and reporting processes

Without a strong foundation of data and attribution - which tools like Semly can provide - AI will remain just an expensive experiment. With such a foundation, on the other hand, it becomes one of the most important levers for increasing revenue and business value.

Sources

• Meta - Win with Conversations (2024)
• Gartner - The Role of Artificial Intelligence (AI) in Sales in 2025
• Gartner - GMs: How to Use Generative AI in GTM for Sales and Productivity Gains (2024)
• Northbeam - The Marketer's Guide to AI
• McKinsey - Agents for growth: Turning AI promise into impact
• Adnan Masood - AI Use-Case Compass - Retail & E-Commerce: Personalization at Planet Scale
• SuperAGI - AI-Powered Marketing Automation: Case Studies on How AI Agents Boost Efficiency and ROI in 2025
• PROS - Airline Revenue Management Software
• Averi.ai - The 2026 Marketing Budget Reality Check
• Madanchian M. - The Impact of Artificial Intelligence Marketing on E-Commerce Sales (MDPI, 2024)
• Envive.ai - 32 AI-Driven Customer Lifetime Value Statistics for Ecommerce
• Cascadia Capital - Sales & Marketing Tech Industry Report 1H 2024
• Wair.ai - AI for strategic promotion planning and execution
• Haptik - The State of WhatsApp Marketing 2024
• McKinsey - The state of AI in 2025: Agents, innovation, and transformation
• McKinsey - How are organizations capturing the value of AI at scale?
• Gartner - 2024 Tech Marketing Benchmarks Survey: AI and Generative AI Insights
• BCG - AI Adoption in 2024: 74% of Companies Struggle to Achieve and Scale Valu
• LinkedIn - How AI can save your sales team from wasting 66% of their day
• Sparkco.ai - Sales Funnel Conversion Rates Analysis 2025
• McKinsey - The state of AI in early 2024: Gen AI adoption
• Deloitte - AI and tech investment ROI
• Mark Orsborn - Measuring What Matters - AI ROI Beyond the Hype
• Wair.ai - Uplift modeling, predicting promotion ROI
• Sellforte - Dynamic scenarios:
• PrescientIQ - Forecasting AI ROI with Aether
• Spinta Digital - Predictive ROI: The Future Beyond ROAS & AI Revenue Models
• LinkedIn - Your 2026 B2B GTM Budget Guide
• Tomi.ai - 2x marketing-driven revenue & LTV:CAC with Predictive AI & first-party data

Share:

Copy link