You Probably Don’t Need the Expensive Model

Most analytics spending never pays off

There is an uncomfortable statistic every analytics vendor knows and few will tell you. Back in 2019, Gartner projected that through 2022 only about 20 percent of analytic insights would actually deliver a business outcome. Four out of five never change a decision. And it shows up in usage too. By BARC’s long running survey of business intelligence users, only about a quarter of employees ever use the BI tools their company buys, a share that has barely moved in seven years, and inside large companies it drops to roughly one in six. The expensive dashboard does not fail because the math is wrong. It fails because it was scoped to impress rather than to move a number, so it becomes the screen nobody opens.

A predictive model, by the way, is just software that uses your past data to forecast something: which lead will close, what to charge, who is about to cancel. Useful, sometimes. Necessary, rarely. A home services owner does not need to be in that four out of five.

The two boring numbers that beat any algorithm

Before you buy a predictive anything, get these two things right, because they are where the money actually is:

1. How fast your team reaches a fresh lead, and what it costs you when they do not. (We wrote a whole piece on the five minute window.)
2. Which marketing channels book sold jobs, not just leads. (That is where the money leaks, and why your reports never agree.)

Get those honest and visible and you will find more money than any fancy model would have surfaced, usually faster and for a fraction of the cost. The big build can wait until you have exhausted the boring wins, and most owners never have to reach for it.

Small, targeted builds win

We have watched this play out across very different businesses. Drive My Way, a recruiting marketplace, did not rebuild its whole pricing system. A small, targeted model left the price on about 70 percent of jobs exactly where it was and raised it only on the unprofitable tail. That one surgical change lifted annual revenue about 24 percent, roughly 23 times what it cost. The whole story is here.

A DTC home and garden brand got a $906K swing in its revenue forecasting from a refresh that cost about $2,000. And a building products manufacturer came to us wanting a more sophisticated forecasting model. We built one, found it was technically more accurate, and then told them to keep using their simpler existing formula anyway, because the small accuracy gain was not worth the added complexity and the maintenance it would cost them every quarter. The simpler tool they would actually trust and keep running beat the fancier one that would slowly rot. We left money on the table on purpose. That is the whole point.

When you actually do need the bigger build

The honest answer is: sometimes you do, and it is worth knowing when. Reach for a real predictive model when three things are all true. You have already nailed the two boring numbers, so this is not a substitute for fixing the basics. Your data is clean enough to trust, because a model built on messy inputs just launders the mess. And you are operating at enough volume that a small percentage improvement is real money, since a two point gain on a thousand jobs pays for a build that the same gain on fifty jobs never will. That is the exact calculus we ran for the building products manufacturer, where the gain did not clear the bar. At higher volume, with clean data and the boring wins already banked, the same model would have been an easy yes. The test is never how clever the model is. It is whether the gain clears the cost.

The right size test

Before you sign for any analytics build, ask three questions:

1. Do I know, in dollars, what slow lead response is costing me? Most owners do not.
2. Can I name which channel booked my last ten jobs? Most cannot.
3. Would my team actually open whatever gets built? Be honest.

If the first two are fuzzy, you do not need a bigger model. You need those two answers. If the third is no, you do not need a dashboard at all. You need one wired to a decision your team makes every day.

Right size first. The build, if you ever need it, comes after the diagnosis, not before. If a vendor will not tell you what you do not need, that tells you something.