01 / Situation
The problem
SageWork's growth team increasingly wins by reaching the right employer before someone else does, but the default way of doing this had real limits. Personalisation, beyond the CRM's built-in mail-merge tokens, a company name or first name dropped into a template, meant a rep building the wording by hand: time-consuming, inconsistent with brand guidelines from one rep to the next, and often leaning on whatever supporting stats or industry data they had to hand rather than anything current or verified. Signals that would have made a real difference, a company update, something on their website, recent activity on LinkedIn, went unused entirely unless a rep happened to notice and write it in themselves.
Prospect lists had to be built by sector manually, and not always possible, since sector was often missing or inconsistent on the record itself. Sequences were built manually too, and inconsistently: what a prospect actually received depended on which rep set it up and how much time they had, sometimes the full multi-touch cadence, sometimes just a single email with nothing else attached. On top of that, new external data landed in the CRM regularly, and the gap between it arriving and any of this happening at all meant the moment often passed, or the same organisation got contacted twice because nobody had checked what was already there.
02 / Manual baseline
Then, no AI availableWhat I actually built
Against that default, I built a tighter discipline around it. When new firmographic data was ingested into the CRM from an external enrichment provider, I made sure it was checked against existing records before anything else happened, to avoid duplicating outreach, then had each prospect properly researched, company website, LinkedIn profile, other public information, including confirming their actual sector where the record itself didn't say.
Outreach was drafted to reference something real about the prospect, tied to a relevant case study and current, credible data on the benefits of hiring experienced, work-ready candidates, not just a name merged into a template. Done properly, that was a genuine, consistent multi-touch sequence every time: three emails, plus a LinkedIn connection request, a call, and engaging with the prospect's own content, all timed and sequenced, regardless of which rep it landed with.
Done well, it worked, consistently and on brand, regardless of who was doing it, though researching and personalising a single sequence properly could still take the better part of a day.
03 / Live system
Now, AI assisted rebuildRun the agent
Pick a synthetic record freshly ingested from an external data source, then watch the agent work through it: checking for duplicates, researching, qualifying, and drafting a personalised sequence, in seconds rather than the better part of a day.
Aldermoor Retail Group
Retail · 250 to 1,000 employeesNorthgate Facilities Services
Facilities & Support · 1,000+ employeesBellview Healthcare Trust
Healthcare · 1,000+ employees04 / What changed
The delta
This is not AI replacing the judgement. The criteria for what counts as a duplicate, what makes a prospect worth pursuing, and what a genuinely personalised sequence looks like all came from experience, not a model. What changed is speed and consistency: research and drafting that used to depend on which rep had time that day now happens the same way every time, in seconds, and fresh data stops sitting dormant.
05 / The point
What this demonstrates
This is what I mean when I say I build commercial systems: a repeatable way of turning raw data into judged, personalised action, one that knows when to stop as well as when to proceed. The tools have changed. The judgement behind them has not.
SageWork: Partnership Lead Scoring, Then and Now
Scoring and routing inbound employer interest used to depend on which rep picked it up, and how much time they had. Here's the same framework, rebuilt to score, route and draft a first response in seconds.
Open system