Understanding ROI

Understanding ROI

The ROI multiplier is the headline number on your Dashboard. It tells you, in plain English, how many dollars of value you got back for every dollar your AI tools cost in the period.

A 5× ROI means the tools cost you $1 and returned $5 of value. 1× is break-even. Below 1×, a tool is costing more than it returns.

The method changed in August 2026, and for most organisations the figure it produces is lower than it was before. What changed is set out below, with the same data run both ways.

The formula

ROI multiplier  =  Estimated value of hours saved  ÷  Monthly cost

Where:

• Monthly cost = for ordinary software, assigned users × cost per seat per month. For a tool recorded as an agent it is the flat monthly cost instead, because an agent isn't bought by the seat: an assignment there means you supervise its output, not that you hold a licence. Summed across every tool that has a cost set.
• Estimated value of hours saved = estimated hours saved × your blended hourly rate
• Estimated hours saved = what your team reported, scaled up per person to cover the days they weren't asked about

The multiplier runs on the estimated hours rather than the reported ones. Both figures sit side by side on Reports, so you can always see how much of the number is evidence and how much is estimate.

Every tool gets its own ROI first. The org-wide headline is then the sum of every tool's estimated value divided by the sum of every tool's cost, which is why the Dashboard headline and the by-tool table in Reports always reconcile.

How the estimate is built

Each person is scaled against their own answers, and the results are added together:

1. Add up what one person reported in the period, and how many working days those answers cover. An answer about today covers one working day. An answer about this week covers a working week.
2. Their factor is the period's working days divided by their own days covered. Never below 1, and never above 2.5.
3. Their scaled figure is then capped at one full working day of saving for each working day in the period. The length of your working day comes from Standard hours per week divided by your Working days setting, so a four-day week is measured against its own reality rather than someone else's.
4. Add everyone's figures together.

The longer walkthrough, including what each answer is worth in minutes and how repeat answers collapse, is in How we work out hours saved.

A worked example

Imagine:

• 30 ChatGPT Team seats at AUD $30 each → $900/month cost
• A reporting month with 22 working days
• Blended hourly rate = $95/hr
• 15 of the 30 answered The AI Pulse, which asks about today, so each answer covers one working day:
  • 5 regulars answered on 20 days each, reporting 300 minutes each
  • 10 occasional responders answered on 2 days each, reporting 80 minutes each

The maths:

Reported            = (5 × 300) + (10 × 80)   = 2,300 min = 38.3h

Regulars:   factor  = 22 ÷ 20                 = 1.1
            scaled  = 300 × 1.1 × 5           = 1,650 min
Occasional: factor  = 22 ÷ 2 = 11, capped at    2.5
            scaled  = 80 × 2.5 × 10           = 2,000 min

Estimated           = 1,650 + 2,000           = 3,650 min = 60.8h
Estimated value     = 60.8h × $95             = $5,779/mo
ROI multiplier      = $5,779 ÷ $900           = 6.4×

The working-day ceiling never bites here: 22 days at 8 hours is 176 hours a person, and nobody in this example is close to it.

A 6.4× gives a health score of about 81, which reads as Strong on the dial.

What changed in August 2026

The old method worked out one extrapolation factor from the average response rate across everybody, then applied it to the whole total. On exactly the data above, the 15 responders averaged 8 days answered each, giving a factor of 22 ÷ 8 = 2.75, capped at 2.5:

Old method:  2,300 min × 2.5 = 5,750 min = 95.8h → $9,104 → 10.1×
New method:                    3,650 min = 60.8h → $5,779 →  6.4×

Same answers, same costs, and the figure moves from about 10× to about 6.4×.

The old number ran high, and it is worth being precise about why. It took the regulars, who had already told us about 20 of their 22 days, and scaled them by a factor borrowed from colleagues who had answered twice. There was almost nothing left to estimate for the regulars, and the old maths estimated it anyway.

So the direction of the change is downward for most organisations. We would rather give you a smaller figure you can defend line by line than a larger one that falls over the first time someone in finance asks how it was derived.

How ROI becomes a health score

On your Dashboard, the ROI gauge shows two numbers: the raw multiplier (6.4× in the example above) and a health score from 0 to 100. The health score is what drives the dial colour.

The mapping isn't linear, because a 2× ROI is fine but not compelling, and a 15× ROI is nearly indistinguishable from a 20× ROI in practice. The curve is opinionated:

We cap at 20× so a single runaway tool doesn't peg every gauge to "excellent" and wash out the signal from the rest. More on normalised health scores →

Three limits worth knowing

1 · It's an estimate

Self-reported time saved runs optimistic. Four things hold it down:

• A 2.5× ceiling on any one person's factor, so a single enthusiast on a sparse pulse can't blow up the org-wide estimate
• A working-day ceiling on top of that: nobody is credited with saving more time in a day than there is in their working day
• A negative option in the time question ("I lost time because of it", counted as −30 minutes) that gets subtracted rather than ignored
• Reported hours shown alongside estimated hours, always, so you can choose how strong a claim to make

2 · It excludes tools with no cost set

A tool with no monthly cost set is left out of the cost denominator entirely, and an agent with no flat monthly cost behaves the same way. To see what the number is leaving out, open AI tools: anything without a cost carries a No cost set badge. On Reports, the wasted-seats card says Set tool costs to quantify while it has nothing to price them against.

3 · It depends on your hourly rate

The dollar value of an hour saved varies widely by industry and seniority. If your rate is too low, ROI looks artificially weak; too high and it looks artificially strong. Pick a number you'd be comfortable defending in a board pack, and see Setting your blended hourly rate.

What if it shows "—"?

One of the inputs is missing. See Why is ROI showing "—"?.

Related: How we work out hours saved · Setting your blended hourly rate · Setting tool costs and why it matters · Normalised health scores (blog)