An assistant to the team, not a replacement for judgement.
A premier UK Georgian manor wedding venue running 200+ events a year. The brand was built on judgement and high-touch service. The risk was that AI would erode the very thing that made the venue valuable.
The situation
The venue was transitioning from dry-hire to high-value comprehensive wedding packages. The reputation, built on attention to detail and high-touch service, was the entire brand. The clientele expected it.
The complication
Data was scattered across twelve-plus systems: Outlook, HubSpot migrating to Sonas, SharePoint, Sage, Excel, Word, Google Forms, email. Booking details were re-entered manually in four-plus places.
The 150GB photo library had no metadata, no naming convention, no tagging. Finding images for marketing took hours. Sales-to-Events handovers ran two-to-three days late because of shift scheduling. Faults were logged via Google Form and waited in a spreadsheet, with no severity classification or RIDDOR alignment.
And, worth flagging clearly: staff had started using public ChatGPT, ungoverned, on high-net-worth client data. The reputational exposure was not theoretical.
The question
The General Manager said it cleanly: AI that feels like an assistant to the team, not a replacement for their judgement. Anything that put the brand at risk was off the table.
The answer
The work split into three layers: process redesign first, automation second, governance throughout.
- Process redesign workshops. Future-state workflows mapped around the Sonas CRM migration. "Sonas-First" data policy established. Shadow spreadsheets decommissioned.
- Sales-to-Events handover automation. At T-30 days, AI extracts the key details and generates a standardised handover document. Same-day delivery instead of two-to-three day delays.
- Open house follow-up campaigns. 24-hour thank-you, day 7 availability reminder, day 14 package highlight, day 21 booking invite. Human review queue before sending.
- AI photo library tagging. First 18 months and supplier showcase (~8,000 images) automatically classified by supplier, event type, season, colour palette.
- Intelligent fault logging. AI classifier assigns severity, flags RIDDOR-reportable incidents, auto-routes, escalates within SLA.
- Human-in-the-loop governance. Approvals required for client-facing proposals and responses. Automated workflows with review queues. Public AI tools restricted to non-client tasks. PII scrubbing layer deployed.
Outcomes
- 35 to 40 hours/week of administrative friction reduced
- Handover cycle: 2-3 days to same-day
- Open house follow-up completion: 60% to 100% (and conversion +12 to 15%)
- Photo library search: 10x faster content creation
- Fault response time: 35% improvement
- Compliance risk: HIGH to MODERATE
- £45,000 to £70,000 annually in reclaimed capacity (at £25-35/hr blended rate)
AI that feels like an assistant to our team, not a replacement for our judgment.— General Manager
Sound like a wrestle you recognise?
The first conversation is a discovery call. We will tell you honestly whether we can help, and what the first step would look like.