Highlights so far:
- One customer record shared by sales, marketing and service, replacing a patchwork of disconnected tools.
- More than a decade of sales history unified and reconciled, ready for reporting and marketing.
- Corporate sales on the same platform, moved over with no data lost.
- Reporting out of spreadsheets and into a structure AI can analyze for insights.
- An AI website assistant live in production.
The challenge
The brand served about 60,000 guests a year but could identify only about 10,000 of them. Its biggest growth lever, raising spend per visit, depended on knowing who had booked and when they were coming.
The information needed to do that was scattered across the business:
- Sales history was locked in an aging order system, and its planned replacement had slipped to 2027.
- Corporate sales and marketing ran in separate tools, with ~240,000 marketing contacts cut off from what those people had actually bought.
- Phone calls happened outside the CRM, so no one could see a guest's call history.
- Website chat was a dead end, with no way to hand a visitor to a service rep.
- There was no single view of a guest. One person could appear under dozens of booking IDs, and walk-in sales were recorded against cash registers rather than named guests.
The deadlines were fixed: an MVP by 30 September and Phase 1 by 15 January. And the business could not stop. Every change had to land safely on live customer data.
The approach
We treated the brand's live operations as the constraint: nothing we changed could interrupt bookings, corrupt sales history or surprise the team.
- Make the hard-to-reverse decisions first. On day one we set up a single customer model that holds both individual guests and corporate clients, so nothing built later would need rework.
- Prove every change before it touches the business. Each feature was built and tested in a mirror of the live system, then promoted. This caught a change that would have wiped 14 fields of customer information from live records.
- Keep a full audit trail. Every change is version-controlled and tied to a tracked work item (30+ items, 76 commits to date), so the brand can always see what changed, when and why.
- Deliver like a bigger team. A small team used Claude Code for building, data analysis and quality checks, under strict confirm-before-change rules, to move at the pace of a much larger one.
- Train each team on what it actually uses. Service, sales and marketing learned only their own tools, and their testing feedback shaped the final setup.
What we solved
The result is one customer record that every team works from. Sales history, phone calls and website conversations all land on it, and marketing acts on it.
“Our teams can't easily get to customer information.”
The brand had its guest data, but it was spread across systems that sales, marketing and service couldn't easily reach. A single guest could also appear under many booking IDs: 59,770 guests carried two or more. We brought it all into one guest profile that every go-to-market team works from, linking consumer and corporate relationships and recording marketing consent on every person. Marketing can now reach guests compliantly, service has the full picture on every call, and sales sees the whole relationship.
“We can't see our own sales history.”
Eleven years of orders were unified and reconciled so that every order total adds up. Along the way we recovered which add-ons belonged to which original booking, linking 52,799 add-on purchases to their parent order for the first time. New bookings, changes, cancellations and refunds now flow in continuously. Every update is logged, so staff can spot and fix a problem themselves without creating duplicate customers.
“Our reps juggle too many screens.”
Service and sales staff now see the guest, their bookings and the phone on one screen. Add-ons appear under the experience they belong to, so a rep can answer "what's on my booking?" at a glance. Calls are dialed from the record and logged automatically.
“Corporate sales lives somewhere else.”
The corporate pipeline was rebuilt around how the team actually sells, with required loss reasons so the business learns why deals slip away. About 7,700 deals and 2,454 contacts moved over with no failures, and 1,532 orders and 373 deals were reconnected to their accounts. A dedicated workspace now lets planners build multi-part group events, see bookings on a calendar and convert a lead straight into a booking.
“Our data costs are running unchecked.”
The brand's customer-matching process re-ran, and was billed, every time anything changed across 240,000 profiles. We replaced it with one controlled process that runs only when needed. Fixing a gap in how email addresses were captured also added 231,687 reachable contact points, and purchase history now connects to each profile, ready for upsell campaigns.
“Website visitors have nowhere to go.”
An AI website assistant built on Agentforce now answers questions using the brand's own website as its source of truth. It first asks whether the visitor has an existing booking or is planning a new one. It then opens a service request or creates a sales lead, and hands off to a live rep when one is available.
“We can't tell which marketing drives revenue.”
A new attribution model credits each order once to the campaigns that came before it and feeds the brand's Snowflake analytics. For group sales, a one-click quote now produces a branded proposal page with built-in e-signature, versioned per quote (currently in testing).
“Our reporting lives in spreadsheets.”
Reporting used to mean exporting data from several systems and stitching it together by hand in spreadsheets. We moved it into a consistent, governed structure across Salesforce, Data 360 and Snowflake, where every number traces back to the same customer and order records. Because the data is now structured and connected, AI tools such as Claude can analyze it directly, so leadership gets insights in hours instead of waiting on a manual report.
Results so far
With clean history in one place, leadership now gets answers in hours to questions it could not answer before, and several of those answers changed decisions already in motion.
| The question | What the data showed |
|---|---|
| How much does upselling really add? | For the first time, the brand can measure how often guests add extras and how much that lifts revenue, confirming upselling as a major growth lever. |
| Should we invest in more capacity? | Demand pressure turned out to be concentrated in a few peak windows rather than across the business, reframing a major investment decision. |
| Why did a key sales segment decline? | The data pinpointed the real cause and disproved the explanation leadership had assumed, changing where the team focused. |
| Can we trust our historical numbers? | Over a decade of history now reconciles, and data-entry errors are flagged before they can distort a report. |
Other outcomes:
- One record per guest. 908 duplicate profiles were merged with every order and activity intact, and 97% of walk-in sales now resolve to a named guest.
- Every call on the customer record. Calls are logged to the guest's timeline automatically.
- Predictable data spend. Customer matching runs only when the business asks for it, not on every change.
- A guest insights dashboard gives leadership a live view built on production data.
What's next
Phase 1 completes on 15 January 2027. With the foundation in place, the remaining work turns unified data into revenue.
- Launch group-sales quoting with built-in e-signature, now in testing.
- Activate upsell campaigns that use booking history to target guests between purchase and visit.
Lessons learned
Most of the risk sat in the data and in how the business really runs, not in the software setup.
- Understand your data before you move it. Learning how existing systems really record customers and orders keeps old problems from being rebuilt in the new platform.
- Surface problems; don't hide them. Flagging bad data openly builds more trust than quietly cleaning it away.
The takeaway
This brand didn't need more software; it needed to know its customers. By fixing the data first, it now answers revenue questions in hours, sells corporate events from one workspace and has a foundation where AI works on real customer information, not in isolated experiments.