Calculating Time Savings
Time savings is the most common AI value driver, and also the most commonly overstated.
The correct approach:
1. Identify the specific task being automated or assisted 2. Measure how long it currently takes per occurrence (be specific, "writing a client report" could mean 45 minutes or 4 hours) 3. Measure how long it takes with AI assistance (AI draft + human review) 4. Calculate the difference and multiply by volume
Example:
Current state: Sales team writes 40 personalized outreach emails per rep per week, averaging 12 minutes each = 8 hours per rep per week. AI-assisted state: AI drafts, rep reviews and edits = 3 minutes per email = 2 hours per rep per week. Time savings: 6 hours per rep per week × 8 reps = 48 hours per week.
At a fully loaded cost (salary + benefits + overhead) of $50 per hour for sales rep time, that is $2,400 per week in time savings, or roughly $125,000 per year, before you have counted whether better outreach also improves pipeline. The same exercise works for marketing teams producing weekly newsletters, operations teams processing vendor invoices, legal teams summarizing contracts, and support teams drafting response macros.
Common overstatement traps:
Do not count time savings against peak capacity as if every saved hour is immediately billable. If your team saves 5 hours per person per week but is not capacity-constrained, the direct dollar value is lower. Those hours return to the business, but not as immediate revenue. The value is in what the team does with the recaptured time: more accounts covered, faster follow-up, deeper research on higher-value prospects. Make that conversion explicit in the business case, or discount the time savings by 40 to 60 percent to reflect that not all recaptured time converts to output.
Count fully loaded employee costs, not just base salary. A $70,000 per year employee typically costs $100,000 to $115,000 fully loaded once you include payroll taxes, benefits, equipment, software seats, and proportional management overhead. Using base salary understates the value of time savings by roughly a third, which can make a marginal ROI look bad when it is actually fine.
Also watch for the adoption gap. A tool that could save 6 hours per rep per week saves zero hours for the reps who never open it. Real-world adoption rates for new internal AI tools are typically 40 to 70 percent in month three. Plan your ROI on expected adoption, not on headcount.
Calculating Cost Avoidance
Cost avoidance quantifies AI's contribution to not spending money you otherwise would have spent.
Common cost avoidance scenarios:
- AI handles inbound support volume that would otherwise require adding a support headcount
- AI processes documents that would otherwise require manual review by a higher-cost employee
- AI catches quality issues that would otherwise result in expensive rework
- AI-generated content reduces the need for freelance copywriters or contract designers
- AI-assisted SEO services work replaces a portion of agency retainer hours
Example:
Current trajectory: Business is growing, adding a new customer service rep at $55,000 per year every time support volume increases by 20 percent. An AI chatbot handles 35 percent of support inquiries without human involvement, with a fallback to a human agent for anything the model is not confident about. The business can grow 35 percent more before needing the next headcount addition.
Avoided cost in year one: $55,000 in hiring avoided. This is conservative. It does not count recruiting fees (often $8,000 to $15,000 for a filled role), onboarding and training costs (typically 20 to 40 percent of first-year salary in lost productivity), or the management overhead of a larger team. Adding those in, the true avoided cost is closer to $75,000 to $85,000 per deferred hire.
Cost avoidance is the cleanest category to measure because the counterfactual is usually documented somewhere: a hiring plan, a budget forecast, a vendor invoice. Tie your AI savings to a specific line item you would have otherwise spent, and the business case stops being hypothetical.
Calculating Revenue Impact
Revenue impact is harder to isolate but often the largest value driver.
Direct revenue impacts:
- AI-personalized outreach generates more pipeline and more revenue closes
- AI-assisted sales prep shortens deal cycles, so revenue is recognized faster
- AI product recommendations increase average order value
- AI retention systems reduce churn and raise customer lifetime value
- AI-assisted website design iterations improve conversion rate on high-traffic pages
Example:
Current outbound: 100 emails per week, 2 percent response rate, 1 meeting per 5 responses = 0.4 meetings per week per rep. AI-assisted outbound: 200 emails per week (increased throughput), 3.5 percent response rate (better personalization), same meeting conversion = 1.4 meetings per week per rep.
That is a 3.5x increase in meetings booked per rep. If a meeting converts to a qualified opportunity at 30 percent and an opportunity closes at 25 percent with an average contract value of $24,000, one additional meeting per rep per week is worth roughly $93,000 per rep per year in incremental bookings. Across a ten-person team, that is almost a million dollars of net-new pipeline influence, even before you credit improvements to brand identity consistency that better-written outreach tends to deliver.
Revenue impact calculations require assumptions, so build your case with conservative numbers and your actual close rates. Run the model three ways: pessimistic (half the lift), expected (the lift you have reason to believe), and optimistic (the best observed case). If the pessimistic case still clears hurdle rate, the investment is defensible. If only the optimistic case works, you are gambling.
Calculating Quality Improvement Value
Quality improvements do not always have direct financial impact, but many do.
Quantifiable quality improvements:
- AI-assisted document review catches errors before they become client problems, reducing rework cost or relationship damage
- AI-generated content passes review at higher rates than manually written first drafts, reducing revision cycles
- AI monitoring catches production issues earlier, reducing mean-time-to-detect and incident cost
- AI-assisted UI/UX design review flags accessibility issues before launch, avoiding remediation cost
- AI uptime monitoring paired with web hosting and maintenance reduces downtime minutes, which for e-commerce can run $3,000 to $10,000 per hour
For each quality improvement, trace the value chain: error rate × cost per error = error cost, then multiply by improvement percentage. If a ten-person contract review team catches 94 percent of issues manually and AI-assisted review raises that to 98.5 percent, the 4.5-point improvement on a base of roughly 2,000 contracts per year at an average downstream cost of $6,000 per missed issue is worth about $540,000 per year in avoided loss. That is not a soft benefit. That is the reason general counsels are approving six-figure AI budgets.
The Full ROI Calculation
Total costs to include:
- Implementation cost (one-time): agency fees, engineering time, integration development
- Annual tool licensing: AI platform subscriptions, API usage fees, vector database hosting
- Ongoing management: staff time to maintain, monitor, and improve the system
- Training: one-time rollout training plus ongoing refreshers as the team changes
- Compliance and security review: legal time, security audits, data processing agreements
Example full calculation:
AI-assisted outreach personalization system for a 10-person sales team:
Costs: - Implementation: $35,000 (one-time) - Annual licensing: $12,000 per year - Ongoing management: 5 hours per month at $80 per hour = $4,800 per year - Total Year 1 cost: $51,800; Total Year 2+ cost: $16,800 per year
Value created (conservative): - Time savings: 6 hrs per rep per week × 10 reps × 48 weeks × $50 per hour = $144,000 per year - Revenue impact: 3x increase in meetings resulting in attributable closed revenue (conservatively) = $200,000 per year additional
Year 1 ROI: ($344,000 – $51,800) / $51,800 = 564 percent Year 2+ ROI: ($344,000 – $16,800) / $16,800 = 1,948 percent
Even with much more conservative assumptions (half the time savings, a quarter of the revenue lift), this project still clears any reasonable hurdle rate. That is what a defensible AI business case looks like: it survives being pressure-tested.
How to Evaluate Your Options
Before you commit budget to any specific AI implementation, run these five checks. Skipping them is how pilots become write-offs.
First, pressure-test the baseline. Ask the team doing the work today how long the task actually takes, not how long it should take. Shadow one person for a day if you can. Baselines that come from gut feel are off by 40 to 60 percent in either direction, and a bad baseline makes every downstream number wrong.
Second, confirm the volume. A workflow has to run often enough to earn back the fixed implementation cost. A task that runs once a week, even if AI speeds it up by 80 percent, rarely justifies a custom build. A task that runs 500 times a day almost always does. Plot the volume honestly before building anything.
Third, quantify the downside. What happens when the AI is wrong? If the failure mode is "the draft is slightly awkward and the human rewrites it," the cost of being wrong is near zero. If the failure mode is "the model sent a commitment to a customer you cannot honor," the cost is extremely high. High-consequence failure modes demand review gates, and those review gates cost time that has to come out of your savings estimate.
Fourth, verify the buy-versus-build decision. For most standard workflows (meeting summaries, email drafting, support triage, content generation), an off-the-shelf product plus light configuration beats a custom build on total cost of ownership. Custom builds make sense when the workflow is differentiated, the data is sensitive enough to require self-hosting, or the integration touches proprietary systems. If you cannot articulate why a custom build is necessary, buy instead.
Fifth, agree in advance on what will be measured and when. The ROI number in a business case is a prediction. The ROI number six months after launch is a result. Decide up front who owns the measurement, which metrics are tracked, and what the decision rule is if the numbers miss. Projects without a measurement plan tend to become projects without accountability.
When the ROI Calculation Doesn't Work
Not every AI implementation has clear ROI. Red flags:
- The workflow is low-volume or highly variable, so it is not worth the fixed implementation cost
- The value is primarily "it would be nice" rather than a measurable outcome
- The required integration complexity makes the cost disproportionate to the savings
- The human time saved does not translate to anything, the person simply has less to do
- The regulatory overhead of deploying AI in this workflow (healthcare, financial services, employment decisions) exceeds the productivity gain
A rigorous ROI calculation also tells you when not to invest. If the numbers do not work, that is useful information. It is better to kill a bad project at the business case stage than to absorb six months of implementation cost and then quietly shelf the result.
Running Start Digital builds business cases for AI implementations and builds the systems that deliver the projected results. Our engagements start with a measurement plan and a kill criterion, so every project either clears the bar or ends on purpose.
Frequently Asked Questions
How do we set realistic savings estimates if we've never run AI in production?
Start with conservative benchmarks from comparable implementations, then validate against your specific numbers. For outreach personalization, industry benchmarks show 2-4x response rate improvements with good implementations. For document processing, AI assistance typically reduces review time by 50-70 percent for routine documents. Apply these ranges to your actual volumes and costs. If the ROI does not hold up even with conservative benchmarks, it may not be the right use case. If it works at conservative estimates, you have confidence in the investment.
Should we include productivity gains from AI tools employees choose to use themselves (like ChatGPT for drafting)?
These are real but harder to measure because they are distributed and unsystematic. Individual AI tool use can create value, but it is inconsistent. Some employees use it heavily, some do not, quality varies. It is worth tracking through periodic surveys, but it is better to build your core ROI case on systematic AI implementations where you can measure inputs and outputs directly. Shadow AI savings are a nice secondary story, not the foundation of the business case.
How long does it take AI implementations to start showing positive ROI?
For focused workflow automation, most businesses see positive ROI within 3 to 6 months of full deployment, sometimes faster if the implementation is straightforward and adoption is high. Complex multi-system implementations with longer deployment timelines may take 9 to 12 months to reach payback. Year 2 and beyond is typically where the returns become substantial because the fixed implementation cost is amortized and the annual operating cost is much lower than year one.
What if we can't quantify the value in dollars?
Some AI value is genuinely hard to quantify: better employee experience, faster response to competitive situations, reduced cognitive load. Do not force false precision on these. Instead, build your ROI case on the quantifiable value, and note the qualitative benefits as additional upside. If the quantifiable ROI does not justify the investment on its own, be honest about that. Qualitative benefits rarely close a gap that the numbers do not support.
How do we account for AI costs that scale with usage, like API calls?
Model usage-based costs as a per-unit charge in your operating expense line, then multiply by expected volume. For a team of 20 heavy users of a Claude-powered internal tool, typical costs land between $40 and $120 per user per month depending on context length and query volume. Build in a 25 to 40 percent buffer for usage growth, and reprice the model every six months as both demand and per-token costs shift. If usage is hard to predict, negotiate a committed-use discount or a volume cap with the vendor.
Who should own the ROI measurement after launch?
The business owner of the workflow, not the IT or AI team. If sales is getting the benefit, sales owns the number. If support is getting the benefit, support owns the number. When measurement lives with the implementation team, the incentive tilts toward reporting success. When it lives with the business owner, the incentive tilts toward honest assessment, which is what you actually want.