Automation vs productivity in outbound - do we know what we want?
One of the more interesting things we have learned while building Causo is that people ask for automation, but what they really want is productivity. Those sound like the same thing but actually aren't.
Ask someone what they want from an AI sales tool and the answer is usually:
“Just find the leads, write the emails and run the outreach for me.”
But the moment the tool actually tries to do that, the questions start: where did this company come from; why does it match my ICP, is this person still working there; where did you find this information; can I change the email; has anything been sent yet?
This is not users being difficult. It is a trust problem. People want to remove the repetitive work. They do not necessarily want to hand over the decisions that can damage their reputation. I think a lot of AI products are automating the wrong part.
In outbound, pressing send is the easiest part. The actual work is: figuring out which companies are genuinely relevant, researching what is happening inside them, finding the right person, verifying that the contact information is real, understanding why they might care right now, writing something that does not look like it came from a mail-merge factory.
Yet most “AI SDRs” focus on sending more emails with fewer humans involved. We took the opposite approach with Causo. The system does the annoying work: searching live sources, cross-checking companies, researching why they fit, finding decision-makers, verifying emails and drafting the outreach.
But you can still see the reasoning, change the targeting, edit the copy and approve what represents you. The goal is not to replace the founder or salesperson, but to help them not spend half the week jumping between Google, LinkedIn, Apollo, Hunter, spreadsheets and Gmail just to send ten decent emails.
Our current product rule is becoming: Automate the labour, not the responsibility.
I suspect this applies well beyond sales. The more public or irreversible an action is, the more control users want. Drafting something is useful. Publishing it without asking is terrifying. Researching options is useful. Making the final decision invisibly is not. Curious how other people think about this.
Do you actually want your AI tools to operate autonomously, or do you mostly want them to prepare better work for you to approve?
Replies
@dawid_baranowski When does automation start hurting your personal touch?
Causo
@dawid_baranowski @ali_rehan1 All depends on context. We try to extract as much of it as possible from the user, but it is not always easy.
@dawid_baranowski @ivan_sem hope it works well.
Causo
@dawid_baranowski @ali_rehan1 Thank you!
Causo
@ali_rehan1 to me it's not really a when, it's a 'how' you use automation. My own real life example:
The way I use automation in my own outbound emails is I write a golden standard sequence by hand, per given market segment or ICP. Then, I let my automation vary specific bits of it (eg change subject line, adapt certain parts, vary a sentence in a given range). This lets me preserve a large degree of personalisation.
At the same time, if I were to say "chatGPT pls make email", I would lose all personalisation. It's a balancing act of context provided, as well as parameters that automation is allowed to touch.
@dawid_baranowski Great thoughts!
Causo
@ali_rehan1 Appreciate you man!
Causo
This has probably been our biggest lesson building Causo. People want the research, enrichment and writing automated, but they still want to understand why a lead was chosen and control what gets sent. Automate the busywork, not the judgment.
Very interesting post.
I see the same with our users.
We had people ask for AI summaries for longer content and then complain that it was too short…
Seems like a psychological thing above everything.
Thnx for sharing!
Causo
@mazin_assaf thanks Maz!
Everyone asks for the tool to just do the whole thing, but the moment it actually does, the trust wavers, and they want to see under the hood. It's not that they changed their mind, it's that full autonomy sounds great right up until it's your name on whatever just went out.
Causo
@johnsongill Exactly. People want the work automated, not the accountability. The second it goes out under your name, you want to know why it was chosen and what it actually says.
This is the right distinction. Most users don’t want full autonomy first. They want better prepared decisions. In outbound especially, AI can do the research, filtering, verification and first draft but the person still needs to own the judgment because their reputation is the thing being sent.
Causo
@nipuntaneja Exactly. The useful part is getting someone 90% of the way there without pretending the last 10% does not matter. In outbound, that last bit is your judgment and your reputation.
The list of questions you got ("where did this come from, can I change it, has anything been sent") maps exactly to what I see building a content tool. People say they want AI to write posts for them, then reject the output the moment they can't trace where a claim came from. Trust in the pipeline is the product, generation is a commodity. I ended up building a gate that blocks any claim without a source, and that one constraint sold the tool better than the AI itself.
Causo
@vladimir_iudin Exactly. The more consequential the output, the less people care that AI made it and the more they care about why it made that choice.
We found the same thing with outbound. Generation is easy. Showing the source, the reasoning, and what will happen next is what actually gets people comfortable enough to let it run.
@ivan_sem Yeah, and content has a nasty version of this: the consequence shows up late. A made-up stat doesn't blow up at generation time, it blows up a week later when the client's audience spots it. So I can't rely on anyone rereading published posts, everything has to get caught at draft. Does Causo show the whole "what happens next" before anything actually sends?
Causo
@vladimir_iudin Agree with you massively. We've spent a ton of time on triple verification of anything we source in the product and simply open up sources of info.
Similarly I think public perception of what AI can and cannot do is a bit littered by big labs' grandstanding. People expect a magic wand; see that it is not one; and then they get very very nitpicky.
@dawid_baranowski The nitpicky part is what I keep running into. People shrug off a weak draft from a human writer, but one invented fact from AI and they never trust it again. Half my product exists because of that. What did triple verification cost you in speed though, did users notice the wait?
Causo
@vladimir_iudin Very relatable man. There is definitely a trust gap when output is coming from a machine. I'm guessing it can be partially attributed to human-weak output being a 'low writing skill', whereas with AI, it could be bad prompt, bad context, hallucination, etc. More potential failure paths = weaker trust?
Speedwise for verifying info - a deep reasoning search for companies can take a couple of minutes. We built a minigame you can play inside the product while you wait for results to populate so users aren't bored lol
This matches what I keep hearing too, just outside sales. People say they want automation, but the second something acts without them, the trust questions start immediately: why did it do that, based on what, can I undo it. That's actually why I built FounderFlow read/advise only instead of action-taking (I'm the founder, flagging since it's relevant here), the trust gap you're describing showed up for me before I'd even shipped an action layer, so I stopped building one. Founders take a slower, honest answer over a fast one they can't verify.