Every time I review a new client's marketing deck, I make a quiet bet with myself: somewhere in those slides there will be a funnel diagram. I almost always win. The sales funnel might be the most-drawn diagram in digital marketing, and at the same time one of the least used in practice: it lives in the deck, but not in the decisions. This article is about moving it from the slide to the operation: what it is exactly, what its stages are, how to build one that reflects your actual business, and which numbers tell you whether it's working.
What is a sales funnel?
A sales funnel is a representation of the path a person travels from having no idea you exist to becoming your customer. It's drawn as a funnel for a simple arithmetic reason: at every step, fewer people move forward than arrived from the previous one. A thousand people see your content, a hundred visit your pricing page, ten book a call, three buy.
The value of the model lies in what it forces you to do: break a fuzzy process ("getting customers") into concrete stages you can measure and improve separately. It's the difference between saying "we need to sell more" and saying "we're losing most people between the quote and the close, let's start there". The first sentence fuels a long meeting; the second one fuels a plan.
What are the stages of a sales funnel?
The names change depending on which handbook you read (TOFU, MOFU, BOFU; awareness, consideration, decision; top, middle, bottom), but the underlying logic is the same everywhere. To spare you the acronyms, let me walk through them with a made-up example: a company selling inventory management software for small businesses.
Discovery
The person has a problem and doesn't quite know what it's called yet. The owner of a distribution business keeps losing sales because he never knows how much stock he has, so he googles something like "how to keep track of inventory in my business". If your content answers that question, you just entered his radar. This is where educational articles, videos and talks do their job: anything that answers a question for someone who is just starting to understand their problem.
Consideration
Now the problem has a name and the person is comparing solutions. Spreadsheet or software? And which software? At this stage people look for comparisons, case studies, demos and reviews. And increasingly, they ask the question directly to ChatGPT or Gemini, something like "which inventory software makes sense for a small business". I'll come back to this below, because it's redrawing the map for this entire stage.
Decision
Two or three candidates remain, and the conversation moves to price, implementation, support and trust. What pushes people forward here are things like a free trial, a well-run demo, fast answers to concrete objections, and signals that you're a real company that will still exist next year.
And after the purchase
The classic funnel ends at the sale, and that cutoff leaves out the most profitable part of the whole thing: a happy customer renews, expands and refers. If your business runs on recurrence (subscriptions, services, repeat purchases), treat retention as one more funnel stage, with its own actions and its own metrics.
Why does the real funnel look so little like the drawing?
The funnel is drawn as a straight line because models simplify, but real people don't move in single file. They come in, drop out, disappear for three months, return through another channel, move backwards a stage. Someone can see you for the first time in a comparison article (entering straight into the middle of the funnel) or arrive through a referral and skip almost the entire path.
That doesn't invalidate the model, it just changes how to use it: the funnel tells you what a person needs depending on how far along they are, even if the actual order and timing are a mess. The practical consequence is that you need content and actions for every stage at the same time, because you always have people at every stage at once.
The typical mistake here is the unbalanced funnel. I've seen companies with twenty educational articles and not a single page comparing their product to the alternatives (the middle is missing), and also the opposite: sites that only talk about themselves and never answer a single discovery-stage question, which amounts to a funnel with no entrance.
How do you build your sales funnel step by step?
Here's the full recipe. It doesn't require a budget, just a few hours and some honesty with your own data.
Step 1: reconstruct how your last customers arrived
Take your last 10 customers and find out, for each one, where they first saw you, what made them consider you, and what finally convinced them. If you don't know, ask them: it's a 10-minute conversation and most people are happy to answer. This step is what keeps your funnel from being a generic template copied from the internet, because the funnel that emerges from 10 real stories almost never looks like the one on the slide.
Step 2: define three or four stages with clear transition criteria
With those cases in front of you, define stages that reflect your real process and, for each one, the concrete event that marks the move to the next. In the inventory software example it could be: visited the site, signed up for the free trial, activated the trial, paid. The acid test is being able to count how many people are in each stage with data; if you can't, the criteria are poorly defined.
Step 3: assign one main action to each stage
For each stage, a single question: what does this person need in order to move forward? The answer defines your marketing to-do list. Educational content so people discover you, comparisons and case studies for consideration, a free trial and a demo for the decision, decent onboarding for retention. And the filter works in both directions: if something you produce doesn't serve any stage, it's fair to ask why it exists.
Step 4: put a number on every transition
Measure how many people move from one stage to the next. With made-up numbers to illustrate: 2,000 visits a month, 100 sign up for the trial (5%), 40 activate it (40%), 10 pay (25%). It doesn't matter whether your rates are better or worse than these; what matters is having them, because without them the question in step 5 can't be answered.
Step 5: fix the biggest leak first
With the funnel measured, deciding where to invest becomes almost arithmetic. If only 40 out of 100 signups activate the trial, doubling your traffic is expensive and drags the same hole downstream; raising activation from 40% to 60% is free by comparison and moves the same final result. Even so, most teams invest at the top (more traffic, more leads) when the leak is in the middle, basically because traffic is the metric they already look at every day.
Which metrics matter (and which ones to ignore)?
The ones that matter are the conversion rates between stages and the time people spend in each one. Those two tell you where the problem is and whether it's moving. Top-of-funnel volume (visits, followers, impressions) is the easiest number to inflate and the one that says the least on its own: you can double your visits with content that attracts the wrong audience, and your entire funnel gets worse while your dashboard gets better.
A warning about the word "lead", which in many teams means any captured email. An email that downloaded a guide and an email that requested a quote are at different stages of the funnel, and treating them the same (sending them the same things, measuring them together, handing them to the same salesperson) is one of the most common ways to ruin the analysis.
What is AI changing about the sales funnel?
The structure holds: people still discover, compare and decide. What's moving is where the middle stage happens. A growing share of the comparisons that used to go through Google now goes through ChatGPT, Gemini or Perplexity: the person describes their situation and asks for a direct recommendation. The assistant answers with two or three names, and that shortlist is built from the information the models have about you: your site, reviews, press mentions, forums, third-party comparisons.
For your funnel, this means consideration no longer happens only on your site or in the search engine, and working on your presence across all those sources has gone from experiment to a normal part of middle-of-funnel work. If you want to go deeper into how it's done, I wrote a full guide on Generative Engine Optimization.
Wrapping up
If you take a single task away from this article, make it step 1: this week, take your last 10 customers and reconstruct how they got to you. With those 10 stories, draw the stages, put a number on the transitions and find the biggest leak. The funnel that comes out of that will probably look very little like the one in your last deck, and that one, the one built from data, is the one you can actually work with.