AI Prospecting for Agencies: From 331 Hours to 15 Minutes
One UK agency we worked with ran a single round of new-business prospecting that took 331 hours of cycle time, and under 1% of those hours actually created any value. The rest was clicking, copying, cross-referencing, and waiting. After we mapped it and built a pipeline, the same round took 15 minutes from trigger to a prospect reaching the client. That is the full story in our case study, and this page takes the prospecting half of it apart properly: what manual prospecting actually involves, what we automated, and how the same pattern is likely to play out in your agency. It is a deeper look at the second of the three leaks we wrote up on our pillar on AI for marketing agencies.
The anatomy of manual prospecting
New-business prospecting is the worst-hidden waste in a lot of agencies, because it has no clean monthly rhythm the way reporting does. It happens in bursts, between client fires, so nobody adds up what it costs. When you finally map one round end to end, you find 17 or more separate manual steps per prospect, and the shape is grimly familiar. It starts with keyword research: pulling the terms a target ranks for, the terms they miss, the terms their competitors own. Then competitor analysis, comparing the target against three or four rivals across visibility, content, and backlinks. Then the link checks, scraping who links to whom and where the gaps sit. Then location and listings research, working out where a local business shows up and where it does not.
None of that is the hard part. The hard part is that a person does all of it by hand, logging into one tool, exporting a list, pasting it into a sheet, switching to the next tool, doing it again, then reconciling figures that never quite line up because every platform counts things its own way. After the research comes the contact work: finding the right person at the target company, finding a working email, verifying it is not a dead box. Then the drafting, writing an outreach message that references what the research actually found rather than a generic template. Then the send, then logging that it went, then remembering to follow up.
Stack 17 steps on top of each other, across a list of targets, and you get the number that shocked everyone when we put it on the wall: 331 hours of cycle time for one round. Cycle time, not effort, because a lot of those hours were a person waiting on a slow export or a tool that timed out. When we isolated the genuinely value adding work, the judgement about who is worth pursuing and the sentences in the outreach that a human had to write, it came to 2.5 hours. Two and a half hours of value sitting inside 331 hours of cycle time. That is under 1% efficiency, and it is not because anyone was slacking. It is because the process was never designed. It accreted, tool by tool, until being a slow router between browser tabs became somebody's actual job. The reason nobody fixes it is the same reason nobody fixes any of these leaks: it has never been mapped, so its true cost has never been seen.
What got automated
Once you can see the 17 steps laid out, the build almost designs itself, because the steps are consistent and the volume is high, which is exactly the territory a machine is good at. For this agency we built an autonomous prospecting platform around five capabilities, and the test for each one was simple: does this step need human judgement, or is it a rule a person has been applying by hand?
- Automated analytics. The platform pulls research data from multiple sources on a trigger instead of a person opening each tool in turn. One pull replaces seven separate manual research steps, the keyword, competitor, link, and listings work that used to eat the front half of every round. The figures arrive the same way every time, so nobody is reconciling mismatched exports at the end of a long day.
- Prospect scoring. Automatic quality checks rank each target so the team is not vetting a long list by hand. The scoring applies the same criteria the agency already used, just consistently and instantly, which means the obviously weak prospects drop out before anyone spends a minute on them.
- Draft generation. The platform writes the outreach as a first draft into a Google Sheet, grounded in what the research actually found rather than a generic template. It is a draft, not the final word. It gives the team something to sharpen instead of a blank page.
- Slack notifications. When a batch of scored prospects with drafts is ready, the team gets a Slack ping. Nobody sits watching a queue. The work surfaces itself when it needs a human, and stays quiet when it does not.
- Direct client sending. Once a prospect is approved, it goes straight to the client from the platform, on the agreed channel, with the send logged. The loop closes itself apart from the one step where a person decides whether this prospect is worth the client's time.
Notice the shape. The platform does the relentless, rule-rich middle, the part that had been done the same way hundreds of times, and it stops at the edge of judgement. A human still approves who gets contacted and still reads the draft before it goes, because that is the slice that was actually creating value in the first place. We did not cut the team. We took 329 of the 331 hours off their plate and left them the 2.5 that mattered, then handed back the rest of the week for client work. After the build, the same round ran in 15 minutes from trigger to a prospect reaching the client, a 99.9% reduction in cycle time, with 0 manual steps in the pipeline itself. The platform later extended into 3 or more follow-on automation projects, because once an agency sees one build land, the next ones get easier to justify.
What this looks like in your agency
Your numbers will not be 331 hours, and we would be wary of anyone who promised they would be. The figure depends on how many targets you chase, how many tools your research touches, and how much of the work one person carries. What does carry across is the shape, and the shape is what matters. If your team does new-business research, you almost certainly have a version of the same 17-step grind, and you almost certainly have not mapped it, which means you cannot yet see how thin the value adding slice really is. The line between the work a machine should own and the work a person must keep is the same in every agency: automate the pulling, the scoring, the drafting, the routing, and the sending; keep the judgement about who to pursue and the sentences that make an outreach land like a human wrote it.
It is the same pattern we describe for the other big leak, where monthly reports eat a full day each and the commentary is the one part that stays human. If prospecting is not your worst leak, reporting might be, and the cure is built the same way: map the process, find the waste, automate the rule-rich middle, keep the judgement. We have written that one up in how agencies automate client reporting with AI. The point of both is that the saving is real, repeatable, and easy to measure, which is exactly what makes either of them a good first build rather than a science project.
Where to start
You do not start by buying a prospecting tool. You start by mapping your own prospecting process, end to end, so you can see where the hours leak before you automate anything. Read the full case study first to see what the finished pattern looks like, then book the half-day workshop. The AI Strategy Workshop puts your department heads in a room, finds and ranks the waste, and picks the one build most likely to ship cleanly, which for a lot of agencies is this exact prospecting pipeline. The fee credits toward the next step, so it is a deposit, not a sunk cost.