--- title: "They tried to sell me 'analytics automation' for $12k. I opened Excel and did the maths" url: "https://www.mikefluff.com/en/blog/roi-automation-formula" locale: "en" author: "Mike Fluff" updated: "2026-03-05" description: "Ten minutes, three numbers, one sheet. A formula after which half of the proposed rollouts fall off on their own because they'll never pay back." tags: - "automation" - "metrics" - "strategy" alternates: - locale: "en" url: "https://www.mikefluff.com/en/blog/roi-automation-formula" markdown: "https://www.mikefluff.com/en/blog/roi-automation-formula.md" - locale: "ru" url: "https://www.mikefluff.com/ru/blog/roi-automation-formula" markdown: "https://www.mikefluff.com/ru/blog/roi-automation-formula.md" - locale: "pt" url: "https://www.mikefluff.com/pt/blog/roi-automation-formula" markdown: "https://www.mikefluff.com/pt/blog/roi-automation-formula.md" - locale: "es" url: "https://www.mikefluff.com/es/blog/roi-automation-formula" markdown: "https://www.mikefluff.com/es/blog/roi-automation-formula.md" --- # They tried to sell me 'analytics automation' for $12k. I opened Excel and did the maths > **In short:** Multiply human minutes per operation by cost per minute and operations per month, then set that against setup cost spread over 12 months plus running cost. A quarterly report worth 32 hours a year costs about $800, so a $12,000 automation with $500 a month upkeep never pays back. Under 20 human hours a month, skip automation; over 80, it almost always pays. An acquaintance called and asked me to look at a proposal. "We're thinking of buying it," he said. "Price isn't low, but they promise 10× efficiency. Could you look?" I looked. It was about automating a quarterly analytics task. A financial report that an analyst compiles every three months, eight hours of work, pulling data from four sources. The automation would do it in ten minutes. Price — $12,000 setup, plus $500/month maintenance. I opened Excel. In ten minutes it was clear this was a **failing rollout**. Not because of bad company or tool. Because **the analyst does this task once a quarter**. Do the maths: Eight hours × four times a year = **32 hours per year**. Analyst costs about $4,000 a month = $25 an hour. So the task's annual cost is **$800**. Setup — $12,000. Maintenance — $6,000 a year. For the automation to break even in two years, you'd have to save **$12,000 a year**. You save a maximum of **$800**. No, this won't pay back. Not in two years, not in five, never. I called my acquaintance and said: "Let the analyst keep doing it. Honestly — it's cheap." --- I have conversations like this two or three times a month. Someone sells automation, someone buys, someone complains "we rolled it out, it didn't pay off." In fact, **in the overwhelming majority of cases** whether a rollout will pay back or not is a question you can answer **in ten minutes** in Excel. With precision sufficient for a decision. I've been doing this for a few years, and I've arrived at a simple formula with three variables. It doesn't give pennywise precision. It gives precision that separates **yes** from **no**. --- ``` ROI = (T × C × N) - (S + O) ───────────────────── S + O ``` Where: **T** — minutes of human time per one operation *before* automation. **C** — cost of one employee-minute (monthly salary ÷ 168 ÷ 60). **N** — number of operations **per month**. **S** — setup cost (dev or implementation), amortised over 12 months. **O** — operating cost per month (LLM tokens, SaaS subscription, maintenance). If `ROI > 1` — pays back in under a month. If `ROI > 0` — under two. If `ROI < 0` — **don't do it**. --- Two examples from real projects. **Example one. Email triage.** T = 3 minutes per email (I sat next to the operator and timed). Operator: $1,000/month ≈ $0.10/minute. N = 2,000 emails/month. Setup: two weeks of development × $2,500 = $5,000. Over 12 months → **$420/month**. LLM tokens: 2,000 × $0.02 ≈ $40/month. Benefit = 3 × $0.10 × 2,000 = **$600/month**. Cost = $420 + $40 = **$460/month**. ROI = ($600 − $460) / $460 ≈ **0.30**. Pays back in ~3 months. **Do it.** **Example two. Quarterly financial report** (the one I started with). T = 480 minutes per report. Analyst: $4,000/month ≈ $0.41/minute. N = **0.33** operations per month (4 times a year). Setup: $12,000 = **$1,000/month**. Operating: $500/month. Benefit = 480 × $0.41 × 0.33 ≈ **$65/month**. Cost = $1,000 + $500 = **$1,500/month**. ROI = ($65 − $1,500) / $1,500 ≈ **−0.96**. Loss. **Don't do it**. Let the analyst keep doing it — 32 hours a year is **cheaper**. --- Three mistakes I most often see when this formula *is* applied. **First — not counting verification time.** Automation rarely gives 100% accuracy. If the system errs in 5% of cases and a human reviews those errors, **actual T_after = 5% × T_before + some time for review**. Keep those 5% in mind. > **Hidden tax** > > If reviewing an error takes **longer** than the original operation — figuring > out "why the AI decided that" — the automation can fail to pay back **even at > 95% accuracy**. Counterintuitive, but I've had three projects like this in > the last two years. So I now always ask the client: "If the system makes a > mistake — who catches it, and in how long?" **Second — forgot about maintenance.** In six months the input format will shift, and the prompt will stop working. Or the LLM API will change pricing. Or a new model drops and is worth trying. That's **5–20% of operating cost every year**. Rule: if there's **no team** that can maintain the automation, it will break sooner or later and ROI flips negative. I now check this at the start. If the client says "the developer leaves after launch" — we need **another** developer, someone internal, who'll maintain it. **Third — counting only direct savings.** Automation often brings **indirect benefits** that don't land in the formula: — Customer response time from 24 hours to 5 minutes = conversion × 1.3. — Load taken off a key expert = they don't burn out = they don't quit. — Ability to scale without hiring = growth without headcount. These aren't in the formula, but sometimes they're **bigger than direct savings**. Especially the third one — with one client we once calculated that automating their qualification process let them take three times more leads without growing the sales team. That's not "savings per lead." That's **new revenue** that otherwise didn't exist. So I count direct savings **first**, indirect benefits **last** — but I count them. And if the direct formula gives ROI near zero, indirect benefits can tip it. --- > **Three minutes, not three hours** > > 1. **T × N** — how many human hours per month? Under 20 — don't automate, > less than 0.25 FTE, not worth it. > 2. 20–80 — **run the formula**, look at it. > 3. Over 80 — automation **almost always pays off**, the only question is which one. --- That conversation about the analytics report I ended with the line: "Automation ROI isn't a matter of faith. It's three numbers multiplied and divided. If someone sells you an implementation and **can't fill in this formula** — they're not solving your business case. They're selling **themselves** work." That's, I think, the most honest rule I can give on this topic. The formula is free. Excel is free. Ten minutes you have. **[Talk about specific ROI](https://www.mikefluff.com/en#contact)** — If you're calculating payback on a specific rollout and you're not sure — come in, we'll look at the numbers. Usually an hour is enough for me to say 'yes, it'll pay back,' 'no, it won't,' or 'it will, but not the way they're selling.'