Four people. Four client campaigns. Twenty-four hours a week spent not doing PR — spent monitoring coverage, compiling reports, updating media lists, and writing proposals from scratch. That was Sandstorm Communications in early 2025: a sharp, well-regarded Manchester B2B tech PR agency that was selling its best asset — team time — on operational work a machine could do instead. Six months after deploying three AI agents, they're running seven campaigns with the same headcount, recovering nineteen hours weekly, and closing new business 22 percentage points faster. Here's exactly what changed.
The Hidden Cost of Manual PR Operations
Sandstorm's founder, Claire, had a clear-eyed view of the problem before we met. "We're a PR agency," she said. "But if you looked at where our time actually went, you'd think we were a data entry firm." She wasn't wrong. A time audit across the four-person team revealed 24 hours per week — six hours per person — spent on tasks that had nothing to do with strategy, relationships, or creative thinking.
The biggest drain was media monitoring. The team relied on a combination of Google Alerts and a mid-tier media database. Every morning, someone worked through overnight alerts, cross-referenced them against client briefs, and manually logged relevant hits into a shared spreadsheet. Coverage that landed on a Friday evening sometimes wasn't captured until Monday. In PR, that timing gap matters — clients want to know when their story breaks, not two days later.
Second was reporting. Sandstorm delivered monthly reports to every client — a PDF covering coverage volume, tier, reach, sentiment, and share-of-voice against competitors. With four clients, that meant four three-hour blocks of manual data collation and writing every month. Twelve hours, every month, on a task that followed the same structure every single time.
Third was new business. The agency was winning work on quality of thinking, but proposals took two to three days to produce — digging through old case studies, formatting credentials, writing tailored narrative. Two days the team didn't have.
What Sandstorm needed wasn't a new tool to add to the stack. It needed an AI strategy that automated the operational layer — so the human layer could get back to the work clients were actually paying for.
"We're a PR agency. But if you looked at where our time actually went, you'd think we were a data entry firm." — Claire, Founder, Sandstorm Communications
What Three AI Agents Changed
We designed three agents, each targeting one of the three high-volume pain points. None of the three replaced a PR professional. All three gave the PR professionals their time back.
The build used the same architecture we've deployed for client reporting automation in other service businesses — a RAG knowledge base grounded in the agency's own data, connected to the tools the team already used, with outputs that require a human review before anything reaches a client.
Agent 1 — The Media Intelligence Agent
This agent replaced the daily manual monitoring ritual. It connects to news APIs, RSS feeds, and the agency's existing media database, runs continuous checks against each client's keyword and source profiles, and classifies every hit by tier (national, trade, regional), sentiment (positive, neutral, critical), and estimated reach.
Significant hits — national coverage, competitor mentions, anything above a set reach threshold — trigger an immediate Slack alert to the relevant account manager, usually within minutes of publication. Minor hits accumulate in a daily digest delivered at 8am. The spreadsheet is gone. The Monday morning catch-up is gone. Coverage that lands at 11pm on a Thursday is in the client's inbox by 8am Friday.
Agent 2 — The Client Reporting Agent
On the last working day of each month, this agent runs automatically. It pulls the month's coverage data from the intelligence layer, retrieves reach and engagement metrics from the media database, calculates share-of-voice against each client's competitor set, and drafts the monthly report narrative — covering performance against targets, key hits, tone analysis, and recommended focus for the month ahead.
The account manager's job is now to review the draft, add their strategic commentary and relationship context, and send. What previously took three hours now takes twenty minutes. For four clients, that's ten hours recovered every month before the team has had to think strategically about any of it. The underlying build follows the same approach described in the AI client reporting guide — with a media-specific data layer on top.
Agent 3 — The New Business Development Agent
When a qualified prospect enquiry arrives — from the website, by email, or via a referral — the new business agent pulls the brief, matches it against Sandstorm's case study library and sector credentials, and drafts a capability document and proposal outline within minutes. It includes relevant prior work, a suggested approach for the prospect's specific sector, and a plain-English summary of what working with the agency looks like.
Claire and her team then add the strategic layer — the specific recommendations, the creative angle, the relationship paragraph that only a human can write. The document they're working from is already 70% complete. What previously took two days now takes an afternoon. The AI proposal writer architecture underpins this agent, with the agency's own case studies and positioning documents as the knowledge source.
The Results: Six Months On
Six months after going live, Sandstorm's numbers tell a clear story.
- Client accounts: From four to seven — a 68% increase, with no additional headcount and no reduction in service quality. The capacity that had been absorbed by admin was redirected into client work and new business development.
- Hours recovered: 19 hours per week across the four-person team. That's roughly half a person's working week, returned to strategic and creative work every single week.
- New business close rate: Up 22 percentage points. Proposals that used to take two days now turn around in an afternoon. Prospects that might have moved on have a response while they're still engaged. Speed combined with a tailored document has changed the conversion maths significantly.
- Running cost: £85 per month, covering API access, cloud compute, and the media database integration. The one-time build cost was £3,200. At the revenue generated by the three additional client accounts, the system paid for itself in the first six weeks.
- Client retention: Monthly reports now arrive on time, every month, without fail. Coverage alerts land in real time. Clients notice. Sandstorm hasn't lost a client to a competitor in the six months since deployment — a record for the agency.
Claire's assessment is direct: "We were a four-client agency that worked like a ten-client agency in terms of the admin we were carrying. Now we're a seven-client agency that works like a four-client one."
"We were a four-client agency that worked like a ten-client agency in terms of the admin we were carrying. Now we're a seven-client agency that works like a four-client one." — Claire, Founder, Sandstorm Communications
What This Means for Other PR and Communications Businesses
The Sandstorm build isn't unique to PR. The three pain points it addresses — intelligence aggregation, regular reporting, and proposal production — appear in almost every UK professional services firm handling multiple client accounts. The sector shapes the data sources and the outputs; the underlying architecture is consistent.
What makes PR and communications agencies a particularly good fit for this pattern is the volume and regularity of the underlying data. Coverage is generated daily. Reports are due monthly. New business briefs arrive unpredictably but follow a reliable structure. These characteristics make the automation reliable — there's no ambiguity about when to run, what to pull, or what format to produce.
The observability layer is critical in an agency context. Clients receive outputs derived from agent work. Every report generated, every alert triggered, and every proposal drafted is logged — so if a client ever questions a figure or a missed piece of coverage, the audit trail exists. Transparency is not optional when agents are producing client-facing material.
The Chartered Institute of Public Relations estimates AI now assists with up to 40% of routine PR tasks across the industry. The agencies reaching the top of that curve aren't buying better point tools — they're building integrated systems where each agent feeds the next. That's the difference between AI adoption and AI strategy, and it shows up immediately in the capacity numbers.
If you run a PR or communications agency — or any professional services firm where reporting, intelligence, and proposals consume hours your team shouldn't be spending on them — talk to us. We'll scope the build, show you the architecture, and give you a clear picture of what the numbers look like for your specific operation. Sandstorm went from scoping to live in three weeks. The build cost less than one month's revenue from the additional client accounts it enabled.