The honest problem with most AI sales automation projects
If you have spoken to two or three "AI for sales" agencies in the last twelve months, you have probably seen the same pattern. A polished discovery call, a proposal landing somewhere between £30,000 and £100,000, a proprietary "agent platform" you have to keep paying for, and a vague promise that your sales team will "10x their pipeline".
Under the bonnet, most of those projects are doing the same thing: GPT-4 or Claude, plus some glue code, plus your CRM, plus your email. The hard part was never the technology. The hard part is knowing which workflows actually save time and which ones create new busywork. That judgement comes from having run sales teams, not from having watched a demo.
That is the gap I work in. I have spent 14 years in B2B sales, including four years running a sales-led business as Managing Director. Now I build the workflows for other founders and sales leaders, using tools they already pay for, and I train their team to own the work after I leave.
Who this is for
- UK B2B sales teams with one to twenty sellers
- Founders who have personally felt the time drain of CRM hygiene, follow-up and prospect research
- Sales leaders who have watched the AI hype cycle and want an ROI-led conversation, not a demo
- Teams whose existing process is roughly working, and want to make it faster, not rebuild it
- Companies that want to own the IP of what I build, with no permanent agency dependency
Who this is not for
- Teams where the underlying sales process itself is broken (start with fractional sales leadership or advisory first)
- Enterprises with twenty-plus sellers across multiple geographies needing a fully bespoke platform
- Anyone hoping AI will replace their sales team rather than augment it
What you actually get
Every engagement is shaped to your sales motion, but the workflows I build most often are these. Each one runs inside Claude (or a similar LLM), your existing CRM and your email, with your team trained on how to run and modify them.
1. Prospect research that does not steal your morning
A workflow that takes a company name and produces a structured research brief: company overview, recent funding or news, likely pain points, named buying committee, and a draft outbound angle. What currently takes a rep 20 to 40 minutes lands in under three. The judgement still lives with the rep; the typing does not.
2. Outbound personalisation that does not sound like AI spam
A drafting workflow that pulls the research, applies your voice, and produces first-touch outreach that reads like you wrote it on a good day. Critically, it learns from your *actual* sent emails (the good ones), not generic templates. The AI tell is gone.
3. Follow-up sequencing that does not stall
Most deals die in follow-up, not in the first conversation. A workflow that drafts the next-touch email based on the meeting notes, the deal stage, and the prospect's last response. Your reps decide whether to send; the cognitive load of writing it disappears.
4. CRM hygiene that runs on autopilot
Stale leads, missing fields, deals stuck in stages they have outgrown. A nightly workflow that flags the dead weight, drafts the cleanup actions, and surfaces a CRM hygiene report your managers can act on in fifteen minutes a week. The CRM finally tells the truth.
5. Deal-stage notes capture from your calls
If you use Gong, Fathom, Read.ai, Otter or any major call recorder, a workflow that turns the transcript into structured CRM fields: next steps, decision criteria, blockers, multi-thread risk, forecast confidence. Reporting accuracy goes up without your reps spending more time typing.
6. Forecasting and pipeline reporting that you actually trust
A weekly workflow that reads your CRM, scores deals on a consistent rubric, and produces a board-ready pipeline view. Your Monday morning pipeline meeting gets shorter and more honest.
How an engagement runs
- Week 1. Sales motion audit. I sit inside your CRM, your inbox and one or two ride-along calls. We agree the three to five workflows that will move the needle.
- Weeks 2–4. Build phase. I prototype each workflow with one or two reps, get it working in production, document it.
- Weeks 5–6. Training and handover. Your team runs every workflow themselves with me on standby. Written documentation lives in your wiki, not mine.
- Weeks 7–8. Before-and-after measurement. We agree the time-saved-per-rep metric at the start, and I show the delta at the end.
Tools I work with
- LLMs. Claude (preferred for long-form sales work), GPT, Gemini.
- CRMs. Salesforce, HubSpot, Pipedrive, Attio, Close. Most others on request.
- Email. Outlook (Microsoft Graph), Gmail, and the main sending platforms (Outreach, Apollo, Lemlist, Instantly).
- Call intelligence. Gong, Fathom, Read.ai, Otter, Granola.
Where this fits with my other services
If your sales process is already working and you just want to make it faster, this is a standalone build. If your underlying process needs work first, fractional sales leadership or a sales advisory project usually comes before the AI build. There is no point automating a broken motion at speed.