AI Marketing ROI Calculator, The Honest Maths (2026)

In this blog post I'm going to give you the actual ROI calculation framework I use to decide whether an AI marketing project is worth doing. The version with the hidden costs included. The version that gives you a real answer, not the vendor's marketing answer.

This is the framework I use. It's less flattering than the vendor versions. It's also closer to the truth.

I've been a marketing consultant for twenty-one years. I went all in on AI in 2024. I've now run this framework on roughly 50 AI projects (mine, my clients', and ones I've audited after they failed). The pattern is consistent: real-world ROI lands at about 35-50% of vendor projections, and payback timelines are typically 3-4x what vendors claim.

By the end of this blog you'll have the formula, the cost categories most calculations skip, the realistic payback timeline by project type, and a decision rule you can apply in 15 minutes.

TL;DR

The honest AI ROI formula:

Annual ROI = (Annual Value Created − Annual True Cost) / Annual True Cost

Where:

• Annual Value Created = Time saved × Realistic hourly value × Realistic adoption rate × (1 − Quality discount)
• Annual True Cost = Tool cost + Implementation + Training + Maintenance + Opportunity cost + Failure provision

Realistic adoption rate is rarely above 60% in year 1. Quality discount is usually 10-20% for AI-assisted work. Implementation is typically 5-10x the tool cost in the first year.

Most projects show real ROI in months 9-18, not month 2.

Step 1: Calculate the gross value (the vendor's number)

Start with what the vendor shows you. We'll subtract from this later.

This is the headline number. It's wrong, but it's the starting point.

Step 2: Apply the adoption discount

The first reality check. Nobody uses any tool 100% of the time in year 1.

Realistic adoption multipliers by tool type:

• AI writing assistant: 40-60% adoption in year 1
• AI meeting summarisation: 60-80% (lower friction)
• AI sales outreach: 30-50% (high friction)
• AI CRM enrichment: 50-70%
• AI content production at scale: 25-45%
• AI customer support: 60-80%
• AI design tools: 30-50%
• AI analytics/reporting: 50-70%

Default to the lower end of the range unless you have evidence otherwise. Most teams I work with overestimate adoption by 50%.

Adjusted value = Gross value × Adoption rate

Continuing the example: 8h/week assumed × 50% adoption = effective 4h/week

Step 3: Apply the quality discount

AI outputs are not equivalent to human outputs in most cases. They're either lower quality (requiring rework) or require additional review time.

Realistic quality discounts by work type:

• Content drafts (long form): 15-25% discount (extra editing)
• Email copy: 10-15%
• Meeting summaries: 5-10%
• Sales personalisation: 20-30% (lower response rates)
• Code generation: 15-25%
• Analytics summaries: 10-15%
• Design assets: 25-40%
• Customer support responses: 10-20%

This is where vendors are most aggressive in lying. AI-generated outputs without human review consistently underperform human work by 15-30% on quality metrics that really matter (engagement, conversion, response rate).

Quality-adjusted value = Adjusted value × (1 − Quality discount)

Step 4: Calculate the true cost (the part vendors hide)

4a: Tool subscription

The number on the pricing page. Include any seat-based scaling.

4b: Implementation cost

Time to set up, configure, integrate, test. This is where most ROI calculations break.

Realistic implementation hours by project type:

• Standalone SaaS tool (no integration): 20-40 hours
• Tool with light integration (Zapier/Make): 40-80 hours
• Tool integrated with CRM/website: 80-200 hours
• Custom AI workflow build: 200-500 hours
• AI agent system: 400-1,000 hours

Most B2B AI tools fall into the "light to moderate integration" bucket. Budget 80 hours minimum.

4c: Training cost

Per-person hours to reach proficiency. Multiply by the team using the tool.

Realistic training hours per person:

• Simple tools (writing assistants, summarisers): 4-8 hours
• Medium complexity (CRM AI, design tools): 12-20 hours
• Complex (custom agents, multi-tool workflows): 20-40 hours

If you have 5 people using a tool that needs 12 hours of training each:

4d: Maintenance cost

Ongoing care. Updates, prompt tuning, integration repairs, vendor escalations.

Budget 15-25% of annual subscription as maintenance overhead minimum. For custom workflows, 30-50%.

4e: Opportunity cost

Most calculations skip this entirely. It's the largest hidden cost.

4f: Failure provision

Some percentage of AI projects don't work. You need to provision for the cost of finding out, plus migrating away.

Industry default: 25% failure rate in year 1. Add 25% of your other costs as a provision.

Total true cost

Step 5: Calculate honest ROI

Honest ROI = (Annual Value − Annual True Cost) / Annual True Cost

Year 2 typically looks better because implementation is sunk and adoption improves:

Year 2 honest ROI (same example, with adoption climbing to 75%, no implementation, lower training, smaller failure provision):

This is the realistic shape. Years 1 negative, year 2 strongly positive, year 3 onward at run-rate.

Step 6: Calculate honest payback timeline

Most projects with positive long-term ROI pay back in months 9-18, not month 2.

Realistic payback by project type:

• Off-the-shelf SaaS, light integration: 9-15 months
• Custom workflow build: 12-18 months
• AI agent system: 15-24 months
• Enterprise integration: 18-30 months

Anything claiming sub-6-month payback is either trivially small or maths is wrong.

Decision rule: when to proceed

Apply this rule:

Proceed if all three are true:

1. Honest year 1 ROI is no worse than −50%
2. Honest year 2 ROI is at least +50%
3. You have the cash to fund the year 1 loss without compromising operations

Pause if:

• Honest year 1 ROI is worse than −60%
• Honest year 2 ROI projects below +30%
• You're relying on the project to fund itself within 12 months

Walk away if:

• The vendor's calculator and your honest calculator differ by more than 3x
• The honest calculator never shows positive cumulative ROI inside 24 months
• The project requires team adoption above 70% in year 1

The categories of AI work that truly pay back fast

Some categories pay back faster than the general framework suggests. These have lower implementation and higher genuine value creation.

Meeting summarisation and CRM enrichment

Low implementation. Adoption is high because friction is minimal. Quality discount is small because outputs are reference material, not customer-facing.

Typical honest payback: 4-8 months.

Customer support deflection (FAQ bots done well)

High value if your support volume is significant. Quality discount is real (15-20%), but volume often justifies it.

Typical honest payback: 6-12 months.

Internal knowledge retrieval

Tools that let your team query internal documentation in natural language. Value depends on how often the team needs to find things.

Typical honest payback: 6-10 months.

Transcription and meeting intelligence

Almost universally positive ROI because the work being replaced is truly low-value.

Typical honest payback: 3-6 months.

The categories that look great and don't pay back

Mass AI content production

The maths looks brilliant. The reality is search engines penalise it, audiences disengage, and you spend the saved time on damage control.

Typical honest outcome: Negative cumulative ROI over 24 months in most cases.

AI sales outreach at scale

Volume goes up. Response rate drops. Reply quality drops further. Net new revenue often flat or negative.

Typical honest outcome: Break-even at best in year 1, positive only if combined with strong human follow-up.

AI-generated design at agency quality

Looks 90% right at 5% of the cost. The 10% that's wrong is the part clients notice.

Typical honest outcome: Negative ROI for client-facing work. Positive for internal use.

AI strategy or planning tools

Strategy is judgement, not pattern matching. Tools that promise to generate strategic plans almost never produce work senior leaders use.

Typical honest outcome: Negative ROI, tool sits unused inside 6 months.

What to do with the honest number

Most of my clients run this framework and end up:

• Cancelling 1-2 AI subscriptions where the honest maths shows persistent loss
• Doubling down on 1-2 areas where the honest maths is strongly positive
• Delaying 2-3 projects until the prerequisite conditions are in place

The framework is not designed to talk you out of AI. It's designed to talk you out of the wrong AI projects so you can fund the right ones.

Free resource: The AI Agent ROI Calculator Cheat Sheet.

Frequently asked questions

Why is my vendor's ROI calculator so much more optimistic? Because they're selling you the tool. Their calculator assumes 100% adoption, zero implementation, no quality discount, and full hourly value capture. Each of those is wrong, and the errors compound.

What's the most common ROI mistake? Counting saved hours as if they convert to billable revenue. They rarely do. Saved hours go to other work, slack, or sometimes nothing. Only count saved hours that are demonstrably reinvested in revenue-generating activity.

Should I include the strategic value of "being seen as an AI-forward company"? Quantify it or don't include it. "Strategic value" is where bad business cases hide.

Can I model multiple AI tools together? Yes, but model interactions. Two tools doing similar work cannibalise each other. Total adoption doesn't go to 2x; it goes to maybe 1.3x.

What discount rate should I use? For SMBs, simple cumulative ROI is fine. For larger businesses, apply your standard cost of capital (typically 8-12%) to multi-year flows.

Want help running this on a specific project?

If the maths says don't do it, that's the recommendation, and you walk away with a saved cheque and no further pitch.

[Book an AI marketing audit →](/contact)

I'm Lilach Bullock. I've been a marketing consultant for twenty-one years. I went all in on AI in 2024. I work with founders and marketing leaders who want AI to really move their numbers, not just their tool stack.

Related: running a guest posting campaign