How to automate sales with AI: a step-by-step walkthrough

A small B2B sales pipeline can be automated in five steps: capture → enrich → qualify → book → follow up. AI and automation take the collection, the research and the chasing; humans keep the conversations and the decisions. With us, a setup like this costs from 49,000 NOK one-off plus 2,300 NOK/month in tuning — or from 5,500 NOK/month as an ongoing service. Here's the full walkthrough, including the places where it breaks.

The short version

• Five steps: capture, enrich, qualify, book, follow up — in that order.
• Automate the admin and the chasing first. That's where the hours disappear.
• Humans take the conversations, the negotiation and the close. AI prepares.
• The most common failures: messy CRM data, over-aggressive automation, and nobody owning the setup.
• Our pricing: from 49,000 NOK one-off + 2,300 NOK/month, or ongoing from 5,500 NOK/month.

Before you automate anything

Two preconditions — otherwise you're automating chaos:

1. A defined process. If "the sales process" lives in one person's head, write it down first. Automation needs rules, and rules require agreeing on what happens when.
2. A clean CRM. Duplicates, dead deals and empty fields produce automation that makes confident, wrong decisions. We've covered why in AI in sales and CRM — the short version is that messy data in means messy decisions out, just faster.

With both in place, you can build. Let's take the steps in order.

Step 1: Capture — every enquiry into one place

Enquiries arrive through contact forms, email, phone, and the occasional social media message. The first job is getting all of them into the CRM automatically, with source and timestamp. No sticky notes, no "someone should reply to that email."

Phone is the worst offender — an unanswered call leaves no trace at all. A voice AI agent plugs that hole: it answers, notes the reason for the call and logs the enquiry in your system.

Where it breaks: forms that send to an inbox nobody owns, and enquiries entered manually "when someone has time." Capture has to be automatic, or everything downstream collapses.

Step 2: Enrich — fill in the picture automatically

A name and an email address tell you very little. Enrichment means the system automatically pulls in publicly available information: what the company does, roughly how big it is, the industry, and what the person does there. What used to be ten minutes of googling per lead happens in seconds.

Where it breaks: AI that guesses when it can't find an answer. The setup must distinguish "found" from "assumed" — and an empty field beats an invented one.

Step 3: Qualify — who deserves your time?

Now each lead is scored against your criteria: right industry? Right size? A need you actually solve? The best go straight to a salesperson with the research attached. The middle tier gets a friendly reply and a follow-up later. The ones who will never become customers get a polite no — also automatic.

Where it breaks: criteria nobody calibrates. AI gets it wrong, especially early on, and ranks the wrong leads on top. A human must be able to override it, and the criteria must be tuned against real outcomes — who actually became customers? This is not a set-and-forget system.

Step 4: Book — from interested to meeting without the back-and-forth

Once a lead is qualified, the goal is one thing: a meeting. The automation sends a booking link or proposes times, puts the meeting in the calendar, sends confirmation and a reminder, and attaches a prep note for the salesperson. Seven emails about "does Tuesday work?" become zero.

Where it breaks: booking without prioritisation. If the calendar fills up with meetings with leads who should never have got one, you've automated away your own time. Step 3 has to work first.

Step 5: Follow up — where most deals die

Very few deals die from a "no." They die from silence — the quote nobody chased, the meeting that never got a summary. The automation makes sure every lead always has a next step: AI drafts the follow-up in your tone, the salesperson reads, adjusts and sends. After a meeting, the summary and next steps are generated. After two weeks of silence, the nudge is sitting there as a draft.

Where it breaks: fully automatic sending. It's tempting to let the system send everything itself, but one tone-deaf email to the wrong person costs more than ten minutes of approvals per day. On important relationships, a human presses send.

Where it breaks — the traps across all five steps

• Messy source data. Mentioned already; breaks things most often. Clean the CRM first.
• Over-aggressive automation. The more automatic outbound, the more can go publicly wrong. Start cautious.
• Nobody owns the setup. An automated workflow isn't finished — it's operated. Someone has to look at the numbers monthly and adjust. Without ownership, it all quietly decays.
• Treating the AI as infallible. It isn't. It misqualifies, writes oddly, and occasionally misunderstands. Build in checkpoints where a human reviews — that's part of the design, not a weakness.

What does it cost?

A setup like this — mapping the process, building the workflows, training on your data and connecting to the CRM and calendar you already use — starts at 49,000 NOK as a one-off project, plus 2,300 NOK/month in tuning. As an ongoing service it starts at 5,500 NOK/month. For a typical small business this is a matter of weeks, not months. The full price list is on our pricing page.

Want to see how much time is tied up in it? Work out what the time-thieves cost you first. And to see what this would look like for your specific pipeline, book a short call and we'll walk through the steps together — and tell you honestly which ones will pay off for you and which can wait. You can also read more about lead generation and how we work with AI implementation.