Measuring automation ROI: time, capacity, revenue, cost

"It saves loads of time" is not a number, and numbers are the only way to know whether an automation should be kept, expanded or switched off. Four lenses, one baseline and a monthly ten-minute scorecard will tell you the truth about every system you run.

Baseline first, always

ROI is a comparison, and you cannot compare against a week you never measured. Before switching anything on, spend one normal week counting the boring things: hours spent on quotes, invoices and chasing; enquiries received and how fast each got a response; quotes sent and quotes chased; invoices outstanding and their age. One week of honest tallying on the back of a diary is enough. It does not need to be precise; it needs to exist.

Firms that skip the baseline end up in the worst position: paying for tools, feeling vaguely busier, unable to say whether anything improved. Firms that have one can say "quotes used to take two evenings, they now take forty minutes" and know the second number is real because they wrote the first one down.

The four lenses

Automation returns value through four distinct mechanisms, and mixing them up produces mush. Score each system through each lens separately:

Time is the easiest to feel and the easiest to overstate. Capacity is the one owners forget: if quoting drops from two evenings to one hour, the real prize is not the free evening, it is that you can now price twelve jobs a month instead of five without hiring an estimator. Revenue is the hardest to attribute but the most persuasive; count it conservatively and only where the link is direct, such as a job that began as a recovered missed call. Cost must include your own fiddling hours, because an automation that needs constant babysitting is quietly expensive.

A worked example, honestly framed

Take quote production, the heaviest admin task in most small firms. Suppose your baseline week showed six hours of evening quoting for two quotes sent. After adopting an AI estimating tool, the same two quotes take ninety minutes of review and editing. The illustrative sums, and these are rules of thumb applied to made-up round numbers, not research:

• Time: 4.5 hours per week returned, roughly 18 hours a month.
• Capacity: at 45 minutes per quote you could send eight quotes in the old six hours, so your pricing throughput roughly quadruples without hiring.
• Revenue: if faster turnaround and consistent follow-up win you one extra job a quarter at your average margin, attribute that and nothing more.
• Cost: the subscription, plus the two hours it took to set up your rates, spread over the year.

Run your own numbers with your own baseline. The structure of the sum matters more than anyone's example figures. And keep the experiment cheap: QS Quoter's first quote is free, so the trial costs you one evening and gives you a real before-and-after data point on your own job, your own rates.

Attribution without kidding yourself

The revenue lens needs one extra discipline: a rule for what counts, written down before the numbers arrive. The clean method is to tag the source at the moment of capture. A lead that came in through the missed-call text-back keeps that tag all the way to invoice, so at month end you can list won jobs by source without archaeology. Where a system merely helped, faster quoting on a lead that arrived normally, do not claim the whole job; claim nothing, or claim it only in a separate "assisted" column you never add to the headline. This sounds excessively strict, and that is the point. The moment your ROI numbers contain generosity, you stop trusting them, and then you stop looking at them, and then you are back to running on vibes with subscriptions attached. A conservative number you believe beats a flattering one you quietly discount. It also makes the kill decisions easier, because nobody argues to keep a system whose honestly-counted contribution is zero.

The ten-minute monthly scorecard

Once a month, fill one row per automation: hours saved (measured or sampled), capacity change, revenue attributed, full cost, and a keep/fix/kill verdict. The verdicts do the work. Keep means leave it alone. Fix means the system helps but leaks, a follow-up sequence with the wrong timing, a text-back message that reads robotic. Kill means three months of scores that never justified the cost; switch it off without sentimentality, because a dead automation still bills monthly. This scorecard is also the natural sequel to the priority order in where to start automating: build in that order, score in this format.

When someone else is accountable for the number

There is a structural version of all this: make the installer accountable for the value. That is how AGMM's commercial model works by design. It starts with a discovery call and a quantified business case, so the expected value is written down before anything is built. Onboarding is priced by scope, from 2,000 to 10,000 GBP, and the ongoing fee is 10 percent of the value tracked in your own CRM, which means the measurement lives in a system you control and can audit. Every install runs week by week with a pilot before go-live, so the scorecard starts producing evidence in month one rather than after the contract is signed. Whether you build it yourself or have it installed, the principle is identical: no baseline, no business case, no automation.