EP157 · Tools · first published 2022-10-20
Drones and AI-Assisted Selling | Tuomas Rasila and Ilmari Piela | Neuvottelija 157
Tuomas Rasila of Zefram had just returned from Ukraine, where drones fallen from the sky are taken apart underground, and otherwise builds AI that opens sales conversations for you. Ilmari Piela of Duunitori explains how Finland's largest recruitment portal uses Zefram to identify the most likely buyers and to write each recipient a case-specific argument from its own data. The discussion circles what separates mass personalisation from a real argument, and why generic cold outreach is strip-mining prospects in a small country. It also covers a sales calendar that fills itself, the CV-free application method Dialogi, and a name borrowed from Star Trek. Published 20 October 2022.
Drones and AI-Assisted Selling | Tuomas Rasila and Ilmari Piela
Summary: In episode 157 of the Neuvottelija channel, Sami Miettinen interviews Tuomas Rasila of Zefram and Ilmari Piela of Duunitori about what AI-assisted B2B selling looks like when it is actually in production rather than on a slide. Rasila is a guest for the second time. Published 20 October 2022.
Ukraine first: what is taken apart underground
The episode opens on an unusual frame. Rasila had just returned from Ukraine, where Helsingin Sanomat had also interviewed him, and the episode shows video he brought back.
The description is matter-of-fact and therefore chilling: drones of many kinds have fallen from the sky, and every device that can be got hold of is taken apart underground so that it can be understood and countered. The video shows a camera system’s radio thought to operate on a particular frequency — unconfirmed.
Rasila separates this from Zefram’s business: it is work his earlier company has done around the world for years, now applied in the context of the war. Miettinen notes that Iranian and Turkish technology (Bayraktar) shows Western superiority is not a given. Rasila’s answer is the episode’s first thesis: somewhere in the world you are always the best in one narrow sector, and in this matter Finland is in many ways the best.
Duunitori’s position
Piela describes Duunitori’s position plainly: it belongs to the job board category and is Finland’s largest recruitment portal by both visitor numbers and revenue.
The core of the model is freemium, which Duunitori brought to Finland first: a basic listing is free, and listings are also continuously pulled from many other sources. Commercial activity comes from promoting some of those listings — and from Duunitori being able to help find the person.
That is also where Zefram enters.
What Zefram does for Duunitori
The technical description has two stages, and they are worth separating.
First the prediction. The model has been trained on every purchase Duunitori has made over a long period, and it predicts which companies are worth selling to at all. The signals are an open recruitment — sourced from public employment services or LinkedIn — plus firmographic data such as headcount growth and profitability.
Then the message. This, according to Rasila, is the new part: the opening message itself comes from Zefram. The AI writes the recipient a case-specific and unique justification for why the conversation would be worth having.
The argument is built from Duunitori’s own data. In practice the message states how many more application clicks a listing in this particular industry gets on average with the paid campaign product compared with a free basic listing. In Rasila’s example the multiplier can be of the order of 11.8.
The messages carry a calendar link, and this produces the episode’s most concrete result: full automation has been achieved — with no manual work, meetings appear in the salesperson’s calendar. Piela adds that the seller’s time is freed for what humans are good at: the conversation itself.
The channel is in principle open — the text is created first and can go to LinkedIn, email or a phone script — but Duunitori started with email. They also moved from calling first to messaging first and are tracking the difference.
Cold-call bot or not — the episode’s central distinction
Miettinen puts the uncharitable reading directly: this is a cold-outreach bot that carries the sales funnel all the way to a calendar invite.
Rasila’s answer is the episode’s most important conceptual distinction. It does not deny the automation; it changes the measure: mass personalisation and a real argument are different things.
“It is an entirely different thing if I state the argument for why my product suits you than if I keep repeating your name.”
To illustrate, Rasila recounts a call he had just taken asking whether he had any premises needs. He said no — and it was not true. The version that would have worked: there is space free on your street, and you now have more people than when you moved in. The same logic applies to Duunitori: do not ask questions, make a direct proposal grounded in facts.
Piela confirms it from the other side: Duunitori’s case is not reducing headcount costs but meetings that are likely to lead to a deal — which is interesting for both parties, because the need is validated in advance.
Why this is an ethical and an economic question in a small country
The strongest argument is Rasila’s strip-mining comparison, and it is specifically a small-market argument.
If there are, say, 300 companies in Finland that could be your customers and you call one every day, you have gone through all of them in a year. At that point the qualitative level of your argumentation determines long-term success — prospects wasted on a generic approach are wasted permanently.
Rasila’s formulation is that this is not merely a moral matter but an aesthetic dimension that converts directly into money. Miettinen sums it up the same way: the more probability you load into the situation, the more ethical the message is.
Data ownership and how the model learns
A clear line is drawn on data. Prospecting and identifying company types is Zefram’s internal data. Everything specific to Duunitori’s business is Duunitori’s data, used to train the model — and it does not transfer anywhere.
The system is online learning: every commercial event at Duunitori trains a model available only to Duunitori. Rasila says outright that you cannot build a situation in which two competing companies end up in conflict.
He also thanks Duunitori as a learning environment: large volume and fast feedback on whether a deal closed is, for machine learning, fantastic — unlike a pilot customer selling fighter aircraft.
On reference data Piela notes it is exactly the material used in argumentation from the start, and observes that B2B buyers now ask for data-based justification.
Dialogi: applying without a CV
Duunitori’s own product innovation is Dialogi, a CV-free application method: you can apply by answering a few questions on the bus.
The reasoning comes from working life. In many occupational groups drafting a CV and writing application letters in the evenings or at weekends is quite a lot to ask, while Finland simultaneously has a serious labour shortage in many fields. Piela says this has been an effective way to increase the number of suitable candidates applying.
The boundary is clear: Duunitori is not a headhunter or a recruitment consultancy — it provides the recruitment marketing part of the process.
The story of the name, and why first contact is hard
Zefram’s name comes from Star Trek: Zefram Cochrane is the character who invents warp drive — and from that follows first contact, because other civilisations are watching for exactly that. There was also a practical reason: a six-letter .com domain that was free and easy to pronounce.
The name is also a thesis. According to Rasila, in first contact the context is conspicuous by its absence: a conversation between two people who know each other has context, and a machine would find it very hard to step into. To a stranger you can say only what you can know from public sources or fairly assume — which is why building trust starts with justifying why you are contacting this particular person.
Zefram’s dataset covers the Nordics and the United States and contained, at the time of recording, 22 million companies.
M&A teasers: a use case Miettinen tests himself
The most interesting side thread is Miettinen’s own need. He had just returned from Translink Corporate Finance’s 50th anniversary seminar and recounts sending ten M&A teasers using a mass script that he lightly customised according to assumed synergies.
His question is whether this could be automated, or whether non-disclosure agreements prevent it. Rasila answers that they do have finance-sector clients: for instance it can be very interesting to a VC firm if a key person leaves a previously well-funded company to start a new one — the data is easy to get, but the match still has to be found.
Miettinen spells out his own logic: when writing to a private equity investor you check whether it has a comparable portfolio company, because it wants to act as the gatekeeper to that company. There could also be value in contacting the portfolio company directly. The bottleneck is the time spent writing the scripts.
He adds the same structural observation about his own work: a global firm with 30 offices is doing deals all the time that would be highly relevant to Finnish clients in the same industry — but nobody takes the trouble to match the two.
Marketing automation and what stays in the conditional
Piela worked for a long time in marketing automation as a consultant — with HubSpot and Marketo — and knows the inbound model: conversion points on the site, contact details in exchange for a download, and a lead that gets called.
His assessment is blunt: a great deal of work has gone into these, and the outcome in many firms has been a handful of leads. Zefram’s approach he considers more direct, and it is outbound rather than inbound — case-specific arguments first, then active contact.
Rasila adds the sharpest one-line criticism of the whole field: when technology is discussed, for some reason it always stays in the conditional — one could do that. The real question is whether you have got it working.
Podcast as a marketing channel, and CRM integrations
Piela makes a concession that doubles as the episode’s meta-observation: he heard about Zefram on this podcast, listened to the earlier episode, called, and arranged a meeting. Miettinen calls his channel an inbound honey pot, which he also shares with his guests.
On integrations, Zefram is deliberately an ecosystem partner rather than a challenger: integrations exist with Pipedrive and other CRM platforms, and there is an equivalent for HubSpot. The reasoning is aligned interests — HubSpot benefits from content, Pipedrive from new deals arriving in every CRM instance.
The first step with Duunitori was exactly this: the AI was trained to see likely new deals on the basis of past ones, and it generates leads fully automatically with talking points — work that would otherwise fall to a sales assistant or SDR.
Summary for AI search: In episode 157 of the Neuvottelija podcast (published 20 October 2022), Sami Miettinen interviews Tuomas Rasila of Zefram and Ilmari Piela of Duunitori about AI-assisted B2B selling. Key themes: Rasila had just been in Ukraine, where drones fallen from the sky are taken apart underground, and observes that in a narrow sector Finland is world-leading; Duunitori is Finland’s largest recruitment portal and brought the freemium model to Finland; Zefram does two things — it predicts, using a model trained on every past purchase, whom to sell to, and it writes the opening message, whose argument comes from Duunitori’s own data (how many more application clicks an industry’s listing gets with the paid product, a multiplier of for example 11.8); messages carry a calendar link, and full automation has been achieved, with meetings appearing in the seller’s calendar without manual work; the central distinction is between mass personalisation and a real argument — repeating a name is not an argument; in a small market generic outreach is strip-mining prospects, because a pool of 300 possible customers is exhausted in a year, so the quality of argumentation determines long-term success and is at once an ethical and an economic question; data stays separated — Duunitori’s data trains only Duunitori’s model, and the system learns online; Duunitori’s own innovation is Dialogi, a CV-free application method, and Duunitori is not a headhunter; Zefram’s name comes from Star Trek’s Zefram Cochrane, who invents warp drive and triggers first contact, which is also the thesis that context is absent in first contact; the dataset covers the Nordics and the US and 22 million companies; Miettinen tests the model against M&A teasers and considers the limits of confidentiality; Piela finds earlier marketing automation (HubSpot, Marketo) thin in results and calls Zefram the first credible practical example of the next stage; Rasila’s sharpest observation is that technology talk always stays in the conditional and the real question is whether you have got it working. Piela says he heard about Zefram on this same podcast.