What AI Does for Sales — Our Workflow at Transformery

Two reasons.

First, because the consulting industry has a bad habit of selling sales transformation programmes whose authors don't run sales themselves. I do. Transformery has a defined sales motion, a defined ICP, a defined pipeline, and a defined set of tools. Anything we recommend, we run ourselves first.

Second, because the most useful thing for a leader thinking about AI in their own sales function is not a generic framework. It is a concrete example of what a small team's workflow actually looks like once the tooling is mature. So here it is.

Our sales motion has five stages. Nothing exotic. The exotic part is what happens inside each stage.

1. ICP and account list definition 2. Account research and qualification 3. First-touch outreach 4. Conversation and follow-up 5. CRM and pipeline hygiene

Every stage now has AI doing work that, two years ago, was either done manually or not done at all.

Our ideal client profile is specific: mid-sized DACH organisations between roughly 50 and 800 employees, with an active transformation challenge in one of our five pillars (Culture, Leadership, Digital, Sales, AI Adoption), and a decision-maker who reads German or English fluently.

What changed: ICP definition used to be a workshop output that sat in a slide deck and was referred to once a quarter. Now it is a structured prompt that lives in a Claude project, alongside our positioning, our case studies, and our disqualification criteria. When we generate a new target list — say, "L&D leaders in Swiss German-speaking financial services" — the prompt does the first-pass filtering against our ICP automatically and flags accounts that don't fit, with reasoning.

This sounds trivial. It is not. The number of hours saved over a year, and more importantly the number of bad-fit accounts that never enter the pipeline, has been the single biggest productivity gain in our sales process.

This is where the most dramatic time savings live, and where most sales teams under-invest in AI even though it is the highest-leverage place to start.

Before: opening a target account meant fifteen to thirty minutes per company. LinkedIn for the org chart, their website for positioning, Google News for recent moves, Handelsregister for company structure, maybe a press release archive. Doable for one account. Painful for fifty.

Now: a single workflow does all of it in roughly ten minutes per account. Web research via a dedicated agent, LinkedIn intel pulled in parallel, recent news summarised, decision-maker mapping, and most importantly — a written point of view about what is likely going on inside that organisation and what hooks might be relevant for a first conversation.

That last part is what most teams miss. The output of account research should not be a fact dump. It should be a hypothesis: given everything we can see, here is what is probably keeping this CEO awake at night, and here is what we could helpfully bring to a first conversation. The fact dump is what the AI is good at. The hypothesis is where the human still has to lean in.

This is the stage everyone wants to automate, and the stage where automation backfires hardest.

Gartner's research finds that 73% of B2B buyers actively avoid suppliers who send irrelevant outreach. Mailshake's 2026 data shows cold email reply rates have dropped year-on-year for 69% of senders, largely due to spam filtering and generative-AI fatigue. The bar for a first touch has gone up sharply, precisely because everyone is now sending more, faster.

Our principle is simple: AI does the preparation, humans do the writing. The AI gives me a structured brief on the account, the contact, the hypothesis, and a suggested angle. I write the actual outreach.

This is not because the AI cannot draft a good email. It can. It is because at our scale (boutique consultancy, not a sales pipeline factory), the cost of a generic-sounding message is higher than the cost of writing it myself. Augenhöhe — the human, eye-level register that defines how we communicate — is hard to fake. Easier to write yourself than to brief well enough.

The exception: warm follow-ups, content delivery, and meeting confirmations. Those are templated and AI-assisted. The first email to a new human being is always my keystrokes.

This is where most CRMs go to die. Reps have the conversation, write hurried notes, never enter them into the system, and the institutional memory leaks out the side.

What changed for us: every sales call is recorded (with consent), transcribed, and summarised into structured outputs — meeting summary, key points raised by the prospect, agreed next steps, open questions to follow up, and a written-from-scratch follow-up email draft in the appropriate language and register. Total elapsed time from call end to follow-up email sitting in my drafts folder: four minutes.

That four-minute number is doing a lot of work. The old version — write notes, transfer to CRM, draft follow-up, format it, send it — was a 40-minute job that mostly happened the next day, if it happened at all. Now the follow-up goes out the same hour the call ended. The conversion impact of "follow up the same day" versus "follow up tomorrow" is not subtle.

What hasn't changed: the conversation itself. AI does nothing useful during the actual meeting. The relationship work, the careful listening, the willingness to disagree well — that remains entirely human. It always will.

Salesforce data shows that sales teams use an average of 10 tools to close deals, and 94% of organisations plan to consolidate their tech stack in the next 12 months. The story we tell ourselves about CRMs is that they are central nervous systems. The reality is that they are graveyards of half-entered data.

Our shift: the CRM is no longer the place where I enter data. It is the place where AI deposits data. Call transcripts, email exchanges, meeting summaries, and proposal versions get logged automatically. Once a week, I review what has accumulated, correct anything that is wrong, and update the stage and probability fields manually. That manual review takes about 30 minutes for the entire pipeline.

The discipline is reversed. The system catches up to reality automatically. The human's job is to make the strategic calls — what to prioritise, what to drop, what to escalate — not to be the data entry clerk.

This is the part that matters most.

AI did not change what we sell. It did not change how we think about the client's situation. It did not change the quality of the conversation we have when we are face to face. It did not replace the judgement calls about who is a fit and who is not.

What it did is collapse the time cost of being good at the rest of the work. Research that used to be possible only for high-value accounts is now possible for every account. Follow-ups that used to slip through the cracks now go out the same day. CRM data that used to be lies is now mostly true.

The 28% becomes 50%. Maybe more. Bain's 2025 analysis estimates AI could roughly double the percentage of time sellers spend actually selling. That tracks with what I see in our own numbers.

Don't automate outreach first. That is the obvious move, and it is the wrong one. The signal-to-noise ratio in B2B inboxes is collapsing fast, and adding to the noise will hurt your brand more than it helps your pipeline.

Start with research. It is the highest-leverage stage, the one most teams under-invest in, and the one where AI delivers value without any downside risk. A team that researches better will sell better — with or without sophisticated outbound.

Then fix the follow-up loop — the gap between conversation and email-in-prospect's-inbox. Cut it from days to hours. That single change has done more for our conversion rates than any tool we've added.

Then, and only then, look at outreach automation. By that point you will have learned enough about your own motion to know what to automate and what to leave alone.

The 30% number has been stable for fifteen years because the standard sales-tech playbook keeps trying to fix the symptoms. AI gives us a real chance to fix the cause. But only if we are honest about what the work actually is — and stop confusing busy with effective.