Between February and April 2025, a London B2B digital services agency ran a controlled 90-day experiment comparing AI-automated sales outreach against their existing manual outreach process. The agency has 14 staff including three business development executives. Before the experiment, each executive researched prospects manually and sent 12 to 18 personalised outreach emails per week. The experiment tested whether AI-assisted research and personalisation could increase volume without sacrificing the response rates that their manual approach produced. These are the exact numbers.
The three executives were split into two groups for 90 days. Group A (two executives) continued using their existing manual research and outreach process unchanged. Group B (one executive) used an AI-assisted workflow: research was conducted by an AI research agent, personalised email drafts were generated by an LLM trained on the agency's brand voice and past successful outreach, and follow-up sequencing was automated for non-responding prospects. All other variables were held constant: the same prospect list criteria, the same target sectors, and the same offer.
Group A prospected a combined 142 companies over 90 days (average 71 companies per executive). Group B prospected 341 companies over the same period (341 companies per executive). Both groups targeted companies matching the same ICP: UK businesses with 20 to 200 employees, in professional services, technology, or financial services, headquartered in London or the South East.
Group A: 142 companies prospected. Group B: 341 companies prospected. Volume advantage for AI-assisted: 140% more prospects reached with the same headcount.
Group A: 31 responses from 142 outreach sequences (21.8% response rate). Group B: 58 responses from 341 outreach sequences (17.0% response rate). Manual outreach achieved a 4.8 percentage point higher response rate. The AI-assisted personalisation was slightly less effective than manual personalisation on a per-email basis.
Group A: 18 meetings booked from 31 responses (58.1% response-to-meeting rate). Group B: 37 meetings booked from 58 responses (63.8% response-to-meeting rate). AI-assisted outreach booked meetings at a slightly higher rate from responses received, possibly because the AI research brief gave the executive more context to qualify the response and convert the conversation effectively.
Group A: 18 meetings in 90 days (9 per executive). Group B: 37 meetings in 90 days (37 for one executive). The AI-assisted executive booked 4.1 times as many meetings as each manual executive in the same period.
Group A: each executive spent an estimated 3.5 hours per week on prospect research and personalised outreach. Group B: the executive spent 45 minutes per week reviewing and approving AI-generated research briefs and email drafts. Research and outreach administration time reduced by 79%.
Group A: 18 meetings, 6 progressed to proposal stage, estimated pipeline value £284,000. Group B: 37 meetings, 11 progressed to proposal stage, estimated pipeline value £512,000. AI-assisted pipeline generated 80% more potential revenue with one executive versus two.
The agency converted all three executives to the AI-assisted workflow after the experiment concluded. They made three changes based on what they learned during the 90-day period.
First: they invested additional time in the research brief quality for mid-to-large prospects. The 4.8% response rate gap between manual and AI was analysed by company size. For prospects with under 50 employees, AI response rates matched manual. For prospects with 50 to 200 employees, manual outperformed AI by 8 to 11 percentage points, indicating that the AI research was less effective at capturing the personalisation signals that resonated with mid-market prospects. They added a human review step for all prospects above 75 employees before the email was sent.
Second: they built a feedback loop. When a prospect responded positively, the executive logged what specifically in the message they commented on or responded to. After 30 rounds of feedback, they used this data to improve the AI's personalisation prompts, lifting response rates by 3 percentage points on the subsequent campaign.
Third: they reduced the follow-up sequence from five messages over 25 days to three messages over 14 days. Analysis of the response timing showed that 84% of responses came from the first or second message. Messages four and five had a negative response (unsubscribe or negative reply) rate that was higher than their positive response rate, indicating they were creating net harm at that frequency.
B2B sales outreach to business email addresses is governed by PECR and UK GDPR. For unsolicited B2B email marketing, legitimate interests is the most commonly used basis. You must include a clear unsubscribe mechanism in every message, honour unsubscribe requests promptly, and not send to individuals who have previously asked to be removed. AI-generated personalisation does not change the legal basis requirements. Document your legitimate interests assessment and ensure your AI outreach workflow includes suppression list management.
The agency in this experiment built their AI research and outreach workflow using Clay (for prospect research enrichment), Claude API (for email personalisation), and Instantly (for email sequencing). Monthly tooling cost: approximately £800. Implementation setup cost with an external consultant: £4,500 for workflow design, prompt engineering, and integration. The experiment's results justified this investment within the first month of full deployment.
To build an AI sales outreach system for your London B2B business, see our AI Process Automation service.
More from the blog
Let us help
Talk to our London-based team about how we can build the AI software, automation, or bespoke development tailored to your needs.
Softomate Solutions Limited
Deen Dayal Yadav
Online
