Fix SaaS Onboarding Drop-Offs, Step by Step

To fix SaaS onboarding drop-offs, count how many people reach each setup step, find the step where the most of them disappear, and then cut that step, move it after the first result, or fill it in for the user. The step that loses the most people is rarely the one your team argues about. It is usually a step nobody has looked at since launch, and one that never needed to come first.

For an early-stage founder this is the cheapest growth you have. Every person who quits during setup was already paid for, in ads, in content, or in your own hours on LinkedIn. You know your product well enough to skip every step in your head. A stranger has to do each one in order, and the order is where they leave.

An AI sales assistant that ends setup with 71 drafts

Picture an AI sales assistant. It writes the first cold email to each prospect in your own voice, using what it reads on their LinkedIn profile and company site. The founder spent a year on the research agent behind it, and the drafts are good. A new user meets five screens before seeing one, under a progress bar that reads Step 1 of 5.

• Tell us about your company. A Company website field, then a large box labelled Describe your ideal customer, with grey placeholder text about Heads of RevOps at Series B fintechs. The Continue button stays disabled until the box holds 50 characters.

• Connect your CRM. Two tiles, HubSpot and Salesforce. There is no skip link. HubSpot opens a consent screen asking to read and write contacts, companies, deals and email. On Salesforce, many people see a notice that an admin has to approve the app.

• Import contacts. Choose a CRM list with at least 25 contacts in it.

• Choose your voice. Three cards, Friendly, Direct and Formal, plus an Upload three sent emails option.

• Write my first drafts. A button, then a spinner with the words Researching your prospects under it, for about a minute and a half, with nothing else on the screen.

Now the counts for one month of signups. 1,000 people land on step one. 720 reach Connect your CRM. 240 reach Import contacts. 170 reach Choose your voice. 124 press Write my first drafts. 71 open a drafted email. So about seven in every hundred signups ever see the thing the product exists to make.

Counting the steps without an analytics team

You do not need a data hire or a new tool to get those numbers. Most of them already sit in your database, because each setup step leaves a row behind when someone finishes it. A workspace row with a website saved. A CRM connection row. A contacts row. A voice setting. A draft with an opened time on it.

The method fits in an afternoon:

• Write every screen a new user sees before the first result on one line, in order. Include email checks, pop-ups and anything with a Continue button.

• Next to each screen, name the row or event that only exists if someone got past it.

• Take everyone who signed up last month and count how many have each row. One query per step, or one export into a spreadsheet.

• Write the loss under each step: how many people reached it, minus how many reached the next one.

• Circle the step with the largest loss in people. Then, on a video call, watch three people who fit your buyer do that step cold, on their own laptop, with their own accounts. Do not help them.

Compare each step to the one just before it, not to the top of the funnel. And count people, not visits. One person refreshing a broken screen five times is still one lost person.

Why the worst step is the one called a formality

In the sales assistant, the voice step the team argued about for three weeks loses 46 people. The draft quality they run evals on every release is seen by 71. The CRM step loses 480, which is two out of every three people who reach it. In standups, the founder calls that step a formality, because every customer who pays already lives in HubSpot.

He never sees it the way a new user does. His own HubSpot was connected months ago, so in every demo the step takes one click and no consent screen appears. He opens the paid conversion chart every morning. There is no chart for the steps, so the step that costs the most has no number anyone reads out loud.

This is the pattern, and most onboarding advice misses it. Teams spend their time on the steps they built with care, because those are the steps they can see. The step that bleeds is plumbing someone added in week two so the product would have data to work with. It was never designed. It was needed, so it went first.

The cost is concrete. On product-led SaaS benchmarks, activation for most products sits between 20 and 40 percent, and a ten point gain typically lifts free-to-paid conversion by 15 to 25 percent. Seven in a hundred is far below that range. It is also worse for an AI product than for classic software. The 53 people who pressed Write my first drafts and left during the spinner still cost money, because the research agent ran for each of them. AI products average around 52 percent gross margin, against 70 to 80 for traditional software, so there is less room to absorb that waste.

Permission, the blank box and the spinner

Almost every onboarding cliff is one of three kinds, and each has a different fix.

The permission cliff. The step asks the user to trust you, or to go and ask someone else. OAuth consent screens, admin approval, a card, an invite to a teammate. The sales assistant asks a stranger for write access to every deal in their CRM, four minutes after meeting them. On Salesforce it asks them to message their RevOps lead, which means tomorrow, which usually means never.

The blank box cliff. The step asks the user to write something before the product has done anything. Describe your ideal customer. Name your workspace. Upload three sent emails. Each one asks the user to do part of the product's job, with no example of what a good answer looks like. Our post on empty state design covers the same problem on screens after setup.

The wait cliff. The product is working and the screen says nothing. A spinner and one line of text for a minute and a half reads as broken. People switch tabs, and many never come back to see the result they were waiting for.

Cut it, defer it, or fill it in

For each step on your list, ask two questions. Does the first result need this? And does the user already have the answer in their head, ready to type? The answers sort every step into one of four moves.

• Defer it if the first result does not need it. Connecting the CRM, importing a list, inviting the team and picking a plan can all come after the user has seen what they get.

• Fill it in if the result needs it but the user would have to think hard to answer. You have their website, so read it and draft the ideal customer description for them to edit.

• Cut it if only your team needs it. Role, company size and how did you hear about us can wait for a later email, or go.

• Show the work if the step is a wait. Put each research step on screen as it happens, and let the draft appear line by line.

Most advice stops at shortening the flow. Fewer steps help, but the count is not the main thing. Order is. A step that loses two in three people when it comes first can lose far fewer when it comes after the user has a reason to want it. The screen list is a queue, and the fix is usually to change who waits in it. Our write-up on sequencing over screens shows how this played out on a real onboarding.

The same setup with the CRM after the first draft

Here is the sales assistant rebuilt with those four moves. Nothing new was built.

• Screen one asks for the company website and nothing else. The product reads the site and fills the ideal customer box with a first attempt the user can edit or accept.

• Screen two asks for one prospect: paste a LinkedIn URL, or pick one of three sample prospects already on the screen.

• Screen three writes the draft in front of them. Lines appear as the agent works, such as Reading their company page, then Found their latest job post, then the email itself, word by word.

• Above the finished draft, a bar reads Send this from HubSpot, with a Connect button. Under the draft, a Change tone link replaces the voice cards.

The HubSpot consent screen is word for word the same as before. What changed is the moment it arrives. Before, it asked for trust in exchange for a promise. Now it asks for access so the user can send an email they have already read and liked. The ask did not get smaller. The reason to say yes got bigger.

What to count after you ship the change

Run the same count on the next month of signups, with the same rows and the same end point: people who open their first draft. Keep that end point fixed for the whole test, whatever you later decide activation means.

• Count the moved step on its own. Of the people who saw a draft, how many connected their CRM? It will still lose people. The change worked if more people saw a draft and more people connected than the 240 who did before.

• Skip the A/B test at this volume. With a thousand signups a month, a split test on a step can take months to say anything clear. Ship the change to everyone, compare the month before with the month after, and write down anything else that changed in the same window, such as a launch or a price change.

• Find the next cliff. Fixing one step moves the biggest loss somewhere else. Circle the new largest loss and start again next month.

If you want a structured pass over the rest of your setup, the onboarding checklist scores it in a few minutes, and the onboarding UX guide covers why people leave before value at all.

Studio Maydit designs the first run of AI products for founders in the US, UK and Europe, which means the screens between signup and the first result. A fixed-scope project runs three to four weeks and ends with a diagnosis of where your product is losing people, step by step, before anything gets redesigned. If you have a setup flow and no idea which step is the expensive one, book a 30-minute call with Studio Maydit and we will count it with you.