How conversion from hot leads at LearnIT grew 3.3×: an anatomy of the sales pipeline

A strong lead: big numbers, a clear reason, a managed outcome

Conversion from hot leads at LearnIT no longer drifts with the team's mood: it holds steadily at 30% against a previous 9%. That is a 3.3× rise with no additional headcount. Calls that go nowhere no longer eat the advisors' day — their share drops from 60% to 20%. The quote and the contract reach the client not “by this evening” or “tomorrow” but within ten minutes of the conversation, instead of the previous day. The share of second calls, which decide most deals, rises from 25% to 55% and stays there, because a system with explicit SLAs sits behind it. This is not luck but a managed machine, where every link works towards one thing — conversion.

The “before” scene: noise, overload and lost attention

Before the project, an advisor's day at LearnIT starts the same tiring way. Applications of every kind pour into CommaCRM: someone left a phone number on a landing page, someone signed up for a webinar, someone downloaded a PDF or took a quiz, and someone simply clicked an ad and filled in a form on autopilot. The CRM carries no clear priority, so everyone gets called. The advisor opens record after record, hears a polite “I was just looking”, tries to draw the person out — but they have no reason, no time and no real intent. Hours go into cold ground, the voice tires, second contacts slip, the promised e-mail is late, “let's discuss tomorrow” becomes “let's do it after the holidays”, and then disappears altogether. At the end of the day the manager opens a report and sees the familiar mosaic: scattered notes reading “call back”, “thinking”, “asked for a quote” — but no overall picture and no answer to the one question that matters: where the funnel loses clients, and why. Fatigue builds inside the team: plenty of activity, few deals, and far too long a path from interest to decision.

The turning point: focus on the ready, and speed after every touch

LearnIT's leadership sets a clear goal: stop calling everyone and concentrate on the people showing meaning and signals of intent. We come in as the contractor, run quick interviews with the head of sales, the leading advisors and marketing, read back through lead correspondence and listen to a sample of calls to hear the real objections, pauses and forks. In a working session we formulate the hypotheses that become the architecture of the pipeline: scoring by action, so that only hot leads reach sales; AI review of conversations against the sales methodology, to see the quality and the context of a dialogue; instant personalised follow-up carrying the quote or the contract; a second-contact SLA as the holy grail of conversion; transparent dashboards for the manager and light gamification for the advisors, so the system does not depend on manual discipline.

How it works now: a sales pipeline from the first signal to the contract

The system starts not with AI but with a source of truth. At LearnIT that is CommaCRM, where every lead, status, activity and comment lives. We wire the entry points so that each signal leaves a digital trace: a webinar, a quiz, a site form, an e-mail open, a link click, a reply, a return visit, a look at the pricing page. Marketing runs quizzes and segmentation in Typeform while SendGrid delivers content and invitations; every open, click and reply flows back into the CRM and into a scoring table in Google Sheets that holds the rules and their version log. There is no magic here: every row in that table is a legible event with a transparent weight, and every change carries a comment explaining why. No guesswork, only checkable rules.

When a lead accumulates enough points, n8n picks up the trigger, pulls the current lead data, checks it against the thresholds in Google Sheets and moves the record into sales. This is the pivotal moment: the team stops cold-calling everyone, because only leads carrying traces of interest reach sales. This simplest element accounts for more than half of the conversion gain on its own, because advisor time that was previously consumed comes back to people who actually have both intent and context.

Then the conversation block kicks in. The call runs through Zadarma, is recorded, and on completion goes automatically to transcription. We use a pairing of the OpenAI API and the Gemini API: the first produces a careful, structured transcript and isolates the meaningful blocks, the second scores against the sales methodology and drafts the recommendations. What appears in the CommaCRM lead record is therefore not abstract phrasing but concrete fields: “context and task”, “selection criteria”, “risks and doubts”, “signals of readiness”, “next step proposed”. Beside them sits a methodology checklist where the AI marks, with plain ticks, what the advisor did well and what they missed. This is not scoring for the sake of scoring but the basis for learning: the manager can open a record at any moment, hear the excerpt at the relevant minute and see which move worked and which did not.

Once the first conversation ends, n8n starts the next node in the pipeline: from the transcript, the AI assembles a personal e-mail. It carries a short, accurate summary of the dialogue in human language, links to material relevant to that lead's particular problem, and a carefully worded proposal for the next step. If the context calls for a quote or a contract, the AI fills a Google Docs template with the right data, passes the draft to PandaDoc and puts the document link into the message. It goes out through SendGrid within ten minutes of the call ending — and that changes the dynamic fundamentally. The client receives the message while the reason and the feeling of the conversation are still fresh, which sharply reduces the leakage between the first touch and the decision. The e-mails use deep links that lead not to the general site but to specific pages — for instance a personal section carrying the programme and terms for that segment, which lifts both follow-through and motivation.

From there the pipeline does not release the contact into chaos. The tracking layer writes open, click and reply statuses back to the CRM and records which links the client followed, which page they lingered on, what they downloaded. If the e-mail is not opened within a reasonable window, n8n creates a task for the advisor and sends a light reminder through Twilio as a short SMS: not pressure, but a convenient next step. If the e-mail was opened and clicked but no reply came, the AI drafts a follow-up that accounts for what was read: “we can see you looked at X — here are two ways forward: a quick fifteen-minute call, or the contract straight away with room to settle details”. This is not a templated blast but live personalisation that respects the client's attention.

A separate line of the pipeline enforces the second-call SLA. We record the agreed time, set reminders for the advisor and track whether the second contact actually happened. If it did not occur inside the SLA window, the manager sees it on the dashboard and the system gently returns the contact to the agenda — the point being that the second call should stop depending on someone's memory and workload and become a rule of the process. In practice this is where the funnel used to break most often; now it is a managed part of the pipeline.

What happens in the CRM: data structure and visibility for the manager

The card view in CommaCRM looks different now. At the top sits the lead's temperature — not a vague impression but a value computed from scoring events. Beside it, the deal stage, which changes not by eye but on specific actions: took the quiz, attended the webinar, opened the e-mail with the quote, clicked the contract, confirmed the second call. Below that, a “gist of the conversation” block: a compact two-paragraph summary the AI writes from the transcript. In the right-hand column, the methodology checklist with green and grey ticks, and the AI's notes for the advisor: where a clarifying question would help, where the value could be stated more simply, where offering a choice of two paths would lower the client's cognitive load. At the bottom, “next step”: a button that creates a task and, where useful, drops in an opening script or a short message.

The manager opens the dashboard and sees not just numbers but a story. Which lead sources produce more hot leads, where replies are slow, which advisors reliably keep the second call inside the SLA, whose e-mails get opened more often. We do not stage public dressings-down — we build feedback into the culture instead. The AI flags strong moves in specific conversations, and those excerpts get used in the morning stand-ups. Learning then rests on real wins and real mistakes rather than on theory.

Why this stack, and what each tool does

CommaCRM is the warehouse of truth about leads and deals. Without a single point of truth, any automation degenerates into a chaotic zoo of data. We spend a lot of attention on clean fields, consistent statuses and simple rules for writing notes, so that the analytics does not need curing later. n8n is the connective tissue, the conductor that reads events from the CRM, the mail system and telephony, launches the AI tasks, watches the thresholds and writes the results back. Its strength is not a wow effect but observability: every process has logs, every step a version, every rule a comment, and every failure a legible alert for whoever owns it. We use the OpenAI API and the Gemini API as tools with different specialities: one assembles and structures text more carefully, the other applies methodological criteria flexibly and writes in clear business language. SendGrid is needed not only for delivery but for careful event tracking, which matters for follow-up and for gauging engagement. Twilio supplies short, well-placed SMS that arrive where e-mail can get lost. Zadarma provides reliable recording and playback and, together with transcription, turns a conversation into data. PandaDoc closes the last mile of the document: quotes and contracts are built from the record's data, signed without bureaucracy and not lost in an e-mail chain. Google Sheets and Docs are not crutches but flexibility: the scoring rules and version logs live there, and that is where we test hypotheses quickly and record what we changed and why. Typeform turns quizzes from a marketing toy into a segmentation point whose answers genuinely change a lead's route through the process. The tracking services return every open, click and reply to the system and make it possible to build scenarios on facts.

Human in the loop, quality and compliance

AI does not replace the person. Wherever the cost of a mistake is high, we keep a human in the loop — a contested classification of a conversation, say, or a non-standard commercial proposal. We hold confidence thresholds: if the model is unsure, the record goes for manual validation. All prompts and templates are versioned, and each version carries a note on why we shipped it and how we will judge the effect. The mail infrastructure runs DMARC, SPF and DKIM so that messages arrive and the domains do not take reputational damage. These boring things rarely make it into a case study, and they are precisely what makes a system reproducible and durable.

Small stories that show what actually changed

The head of sales runs the stand-up differently now. Instead of asking how everyone feels, they open a specific conversation, find the moment where an advisor gently moved the client from generalities to specifics, and name it as a team move. In the next conversation another advisor repeats it, and we watch the share of agreed next steps rise. The advisor who used to spend the morning thawing a cold list now steps into a queue of hot leads and is doing real work from the first minute. They do not compose an e-mail from scratch — the AI prepares a tidy draft, the advisor adds one or two personal details, and the message goes out on time. The client, spoken to only moments ago, receives it before they have switched to other tasks, and replies the same day, because a convenient next step is already on the table.

The results and what they mean for the business

Conversion from hot leads rising from 9% to 30% is not a pretty percentage but a direct addition to revenue at the same traffic. The team does not grow, the acquisition budget does not swell, and yet the sales function turns the same flow of attention into more contracts. Cutting calls that go nowhere from 60% to 20% gives the team back time and energy: instead of aimlessly trying to thaw random applications, advisors talk to people who are ready. Sending the quote and contract in ten minutes rather than a day removes the key leak point — when the offer is sitting in the inbox here and now, the client does not have time to cool off or move to a competitor. Raising and holding second contacts from 25% to 55% turns chance into predictable practice: once the second call is the norm, both funnel quality and forecast confidence level out. One detail deserves saying separately: scoring alone accounts for more than half of the conversion gain. The system does not try to improve everything at once; it starts by stopping the waste.

Why this story is reproducible rather than accidental

We deliberately avoid heroic explanations and build the process on simple principles. First: events matter more than opinions — if a client's action is recorded and carries a weight, priorities settle themselves. Second: speed after every touch — any promise made in conversation gets its digital continuation within minutes, not hours. Third: visibility for the manager — learning rests on facts, and changes to prompts and rules are recorded and explained. Fourth: respect for the client — e-mails are concise, humanly precise, and offer a choice rather than applying pressure. Fifth: a human wherever the AI should not decide alone. An architecture like this is not tied to any one advisor and does not collapse when they go on holiday.

What a reader can do at home without breaking the company

Start not with “which model do you use” but with the answer to “where does the funnel lose clients”. After that, go by symptom.

And do agree on a handful of metrics that will be your monitor. These four make the effect of automation visible without any elaborate BI panels:

• Conversion from hot leads

• The share of calls that go nowhere

• Average time to send the offer

• The share of second calls

The horizon: where the system scales next

LearnIT now applies the same principle to new lead channels: additional forms and webinars are not forgotten but woven into the pipeline straight away. The scoring rules get extended without turning into magic — every change is recorded and tested. Segmented e-mail scenarios appear: one attachment logic fires for people who came from a webinar, another for those who took a quiz. Wider personalisation is being considered too: when the AI sees that a lead is interested in a particular track, it offers material and a next step in the value language of that track. The whole system grows outward while the core stays the same: events, speed, visibility, respect, and a human at the critical points.

Gamification and a culture of improvement

It matters to us that automation is not experienced as an external add-on but becomes part of the culture. For that we use light gamification: the AI flags the good moves of the week, and the manager hands out small recognitions for repeatable practices that move conversion. We do not run a race for who makes the most calls; we highlight substantive wins — the best next step, the clearest statement of value, the fastest and most precise follow-up. Automation then stops being one more tool and becomes a system for learning and growth.

Behind the scenes: tests, versions, rollbacks

None of the prompt and rule logic is frozen. We keep version logs recording changes, hypotheses and success criteria. Before anything moves into n8n, we check it by hand on a limited sample. If a new e-mail version unexpectedly reduces replies, we calmly roll it back rather than argue about who was right. That takes the emotional swings out and makes the rollout safe for the business. And yes, there are no secret ingredients in this work — only discipline and observability.

In summary

The system we built for LearnIT shows how automating a sales function turns chaos into a managed pipeline. Conversion from hot leads rises from 9% to 30%, calls that go nowhere fall from 60% to 20%, the e-mail carrying the quote and contract goes out within ten minutes, and second calls — which decide most deals — become the norm and hold at 55%. The important part is that this happens without extra headcount, because the team's energy goes where meaning and intent are, and the routine and the delays fall away.

If you recognise your own symptoms in this story, start with a quick read of the funnel: where exactly clients are lost, which of that is measurable, and which loss costs the most. AI does not replace people — it gives them back time and attention. When the e-mail carrying the substance and the value lands in the client's inbox ten minutes after the conversation, the conversation is already about speed, clarity and trust rather than about who failed to call back.