Most companies use AI today, but few measure what they actually get out of it. The sense that "it's faster" isn't enough to make good decisions about where to invest next. It can be measured – but it takes a little structure.
ROI here doesn't have to mean a precise financial model. It means an honest answer to a simple question: for this workflow, what changed, by how much, and was it worth what it cost? Everything below is in service of answering that for one workflow at a time, rather than for "AI" as a whole.
Start with the baseline
You can't measure an improvement without a before. Before introducing AI into a workflow, capture the current state: how long does it take, what volume is handled, what quality does the output hold, what's the outcome?
Without that baseline, any claim about value becomes a guess.
The baseline doesn't have to be elaborate. A week or two of honest notes on a single workflow – minutes per task, number of items handled, how often the result was good enough to use as is – is usually enough to compare against later.
A worked example
Say a small team does research and outreach: someone looks up a prospective account, finds the right contact, works out what they care about, and writes a tailored first message. For example, suppose this takes 40 minutes per account today, the team gets through 10 accounts a week, and roughly 1 in 10 messages gets a reply.
Now suppose AI drafts the research summary and a first version of the message, and the person edits and sends. For example, time per account might drop to 15 minutes, so the same hours now cover 25 accounts a week instead of 10. If the messages are also better targeted, the reply rate might rise from 1 in 10 to 1 in 7.
These numbers are illustrative, not measured. The point is the shape of the calculation: 25 minutes saved per account is real time you can cost, 15 more accounts a week is added capacity, and a reply rate moving from 10% to 14% is a quality gain that sits on top of the extra volume. Three replies a week could become five. Put a value on a closed deal and you can see when the cost of the tooling is small next to the return – and when it isn't.
Three kinds of value
The return usually shows up in three ways, and each one is measured differently:
- Time saved – hours freed up, translated into cost. Measure it by timing the same task before and after on a handful of real cases, then multiply the minutes saved by a loaded hourly cost and the volume you actually run.
- Quality and outcomes – higher response rates, fewer errors, better decisions. Measure it with an outcome you already track: reply rate, error or rework rate, win rate, or a simple count of how often the result was good enough to use as is on a sample.
- Capacity and revenue – more work getting done, or more business, without more people. Measure it as throughput, meaning items handled per week, or pipeline and revenue per person, and watch whether freed time turns into more output or just disappears.
Most workflows show value in more than one of these at once. The example above moves all three. Pick the one or two that matter most for the workflow in question, and resist the urge to measure everything.
Common measurement mistakes
The most common mistakes are measuring activity instead of outcomes, forgetting operating and maintenance costs in the calculation, and crediting AI with the entire improvement when other factors also played a part.
Keep it simple: pick a couple of metrics tied to a real workflow, measure before and after, and track the trend over time. That's how you see what's worth scaling – and where the next investment should go.
From measurement to decisions
Numbers are only useful if they change what you do next. Once a workflow has a baseline and a few weeks of after-data, three decisions usually become clear.
Scale what works: if a workflow shows a clear, repeatable gain, widen it – more of the team, more of the volume, the same pattern applied to a neighbouring task. Drop or rework what doesn't: if the gain is thin once you count the editing, oversight and subscription cost, stop and try a different approach rather than forcing it. And let the results point to the next investment: the workflow that improved most, or the one with the most volume behind it, is usually where the following bit of effort pays back fastest.
That's the whole loop: baseline, measure a real workflow, read the result honestly, decide. It keeps AI spending tied to outcomes instead of enthusiasm.
If you'd like help setting that measurement up – choosing the workflow, capturing a baseline and reading the numbers – we're happy to help. Get in touch with us at Nodal and we'll work through it with you.
