Find the Leak Before You Fix It

A good plumber walks into the basement and does nothing for a while. He listens. He puts a hand on the pipe, watches the gauge, finds the spot where the pressure drops. Only then does he reach for the wrench. The bad plumber starts cutting the second he hears a drip, replaces three feet of perfectly good copper, and hands over a bill while the real leak keeps going behind the drywall.

Most companies buy AI like the bad plumber.

The pattern is everywhere now. A team decides this is the year AI fixes the funnel, so they bolt a model onto every stage at once. Chatbot on the homepage. Scoring on the leads. Drafting on the emails. Summaries on the calls. It feels like progress because something is happening at every joint. Six months later the numbers haven't moved, and nobody can say which of the eight new tools earned its keep. They watered the whole garden to save one wilting plant.

The reason it fails is that a funnel doesn't leak evenly. It leaks in one place, hard.

Look at the benchmarks and the shape jumps out. Stage-to-stage progression in a typical SaaS funnel runs 20 to 40 percent, and the heaviest drop-off sits right at the top, at awareness, where 60 to 80 percent of prospects fall out before they ever become a real opportunity (HubSpot 2023 and Reforge 2022, via SparkCo). E-commerce has its own famous bleed: cart abandonment runs 70 to 90 percent (SparkCo). These aren't rounding errors. They're the burst pipe. Everything downstream is a slow weep by comparison.

So the question isn't whether AI can help. It's where the water is actually escaping, and whether a tool at that exact joint moves enough to pay for itself.

That second half is where teams get lazy. They'll quote the upside without ever checking the gauge. And the upside is real when it lands in the right spot. Automation at the leaking stage has been tied to a 20 percent lift in conversion rate (HubSpot State of Marketing 2023), an 80 percent cut in manual spreadsheet work (McKinsey Global Institute 2023), and a 25 percent improvement in CAC payback (Salesforce State of Sales 2024). Those are the headline numbers people put in the deck. The trouble is they put them in the deck before they've found the leak, so the numbers describe a job that may not be the one in front of them.

Here's the part that makes a finance person sit up. You can model the lift before you spend a dollar, because the benchmarks give you the baseline. Say a build costs $100,000. A 10 percent lift in conversion works out to roughly $30,000 in additional annual revenue (SparkCo). Run that out and a ten-person team saving 20 percent of its time pencils to a first-year ROI of about 33 percent, with payback landing somewhere between three and twelve months (SparkCo). Notice what that math demands. A starting conversion rate. A current cost. A real number for where the funnel sits today. Without the baseline, the ROI line is a wish with a percent sign.

This is the discipline the spreadsheet-everywhere crowd skips. They cut before they check the pressure.

A plumber who actually knows the trade runs the order in reverse. Find the leak first. Confirm it with the gauge, not a hunch. Size the repair against what the fix is worth. Then, and only then, reach for the wrench. Translate that and it's almost boring in its sensibility: measure the funnel stage by stage, find the one with the 60-to-80-percent fall (HubSpot 2023 and Reforge 2022, via SparkCo), model what closing even part of that gap is worth, and aim the AI there. One joint. The loud one.

Two things make this hard in practice, and both are about instrumentation rather than ambition.

The first is that a lot of teams genuinely don't know their stage rates. They have a CRM full of activity and no clean read on what percent of awareness becomes interest, or interest becomes a deal. You can't find the leak if you've never put a gauge on the pipe. Benchmarks help you guess, but your own numbers are the only ones that tell you the truth about your basement.

The second is that without a baseline you can't prove the fix worked even after it does. If nobody wrote down the conversion rate before the model went in, the after number is just a number. The McKinsey figure on spreadsheet work and the Salesforce figure on CAC payback are averages across companies that measured. A team that skips the baseline forfeits its own version of those stats, and ends up arguing about a tool's value on vibes.

The good news is that the fix for both problems is the same one, and it's cheap. Instrument the funnel before you touch it. Capture the stage rates, the cost, the time spent. That single habit turns AI from a faith purchase into an underwriting decision, the kind where the model is a line item with an expected return instead of a hope with a login.

So before the next big spend, before the chatbot and the scoring and the summaries, somebody should go stand in the basement and listen.

Find the leak. Then pick up the wrench.

Sources: HubSpot 2023 and Reforge 2022, via SparkCo (SaaS funnel stage-progression of 20-40% with 60-80% awareness drop-off); SparkCo (e-commerce cart abandonment 70-90%; the $30,000-on-$100,000 worked example; 33% modeled first-year ROI and 3-12 month payback); HubSpot State of Marketing 2023 (20% conversion-rate uplift); McKinsey Global Institute 2023 (80% reduction in manual spreadsheet work); Salesforce State of Sales 2024 (25% CAC-payback improvement).