Analytics spend is easy to cut because it is hard to defend. Count dashboards built and it looks optional. Count decisions changed and it becomes an investment.

Stop counting dashboards. Count decisions.

Analytics spend is unusually easy to cut, because it is unusually hard to defend. When budgets tighten, the data team is often first to feel it, not because it delivers little, but because it has never measured what it delivers in terms anyone upstairs recognises. It counts dashboards built and users licensed, which is activity, and stays silent on the thing that matters, which is decisions made better.

We cover the practical side of this in our data strategy and leadership work, and Why The AAP Is Not A Traditional Retainer looks at a closely related question.

The return is real. It is just framed wrong almost everywhere.

Output is the wrong unit

Dashboards delivered, reports published, licences issued: these are measures of effort, not value. A hundred dashboards nobody opens is not a return, it is a cost with a number on it. Reporting your output to a finance director is reporting your busyness, and it invites the obvious reply, which is that you could presumably be just as busy with less.

The unit of analytics value is a decision: one made faster, one made better, or one made at all that could not be made before. Start from the decisions your analytics is meant to serve, the ones you named when you scoped the work, and measure what happened to them. That reframes the whole conversation. You are no longer defending a cost centre that produces charts. You are showing how the business decides, and what deciding better is worth.

The four returns, in the order to use them

There are four kinds of return, and they are not equal in an argument. Better decisions is the largest, and the one to lead with: pricing set on current rather than stale data, stock held at the right level, churn caught early, the unprofitable line spotted. These move real money, and they are what analytics is for. Reduced risk comes next: errors avoided, decisions no longer made on wrong numbers, audits made defensible, and a single bad decision avoided often dwarfs a year of efficiency. New questions, the capability to ask what nobody could ask before, is genuine but hard to quantify, so use it to colour the case, not to carry it. And time saved is real but the weakest argument to lead with, because a finance director discounts saved hours that do not show up as fewer people. Mention it. Never build the case on it.

Set a baseline, or you can prove nothing

This is the discipline almost nobody applies, and the reason most ROI claims are unprovable. Before you build, capture the before. How long does month-end take today. How often is a decision made on a number that later turns out to be wrong. How long to answer a question the board asks. Write it down. Without a baseline, every later claim of improvement is an assertion, and assertions lose to spreadsheets.

Attribute honestly, because credibility is the asset

Do not claim the whole revenue uplift for the dashboard. Claim the contribution, and claim it conservatively. Over-claiming is the fastest way to lose a finance director, because the one number they can check and find inflated discredits every number you present. A small, defensible figure you can stand behind under questioning is worth far more than a large one that evaporates on contact.

So before your next budget conversation, pick one thing analytics changed and capture the before and after honestly, contribution only. One defensible number does more to protect your spend than a deck of dashboards ever will. Tying the work to named decisions from the start is how we scope every Analytics Acceleration Programme and our analytics strategy engagements, and the full method is in our Measuring the Return on Analytics guide.

If any of this sounds familiar, talk to us about your data.

Related reading

• A Fixed Price For Unscoped Work Is A Red Flag
• What The Pub Trade Teaches About Your Numbers
• A Dashboard Nobody Opens Is Not A Tool Problem