Rebuilding PHI’s CRM into a system the team could safely operate

L’empresa

PHI is a digital strategy and growth agency based in Spain, working with teams across Europe and LATAM. Their work spans digital strategy, performance marketing, CRM and automation, and brand execution.

As their delivery became more integrated and performance-driven, their own CRM stopped being a database and started being the backbone of marketing, sales and decision-making.

The system was never designed for that role.

The real problem

Two problems sat on top of each other, and they had to be solved in that order.

The data could not be trusted. The same contacts existed several times under different emails, job titles or company names. Companies appeared under multiple variations. Associations between contacts, companies, deals and activity history were broken or incomplete. Segmentation was unreliable and reporting produced inconsistent numbers depending on which view you opened.

The data carried no decision signal. Even when a record was clean, it held a name and an email and nothing else. No fit criteria, no intent, no priority. Everyone looked identical in the CRM, so prioritization ran on intuition.

Without enrichment, scoring and clear prioritization, the CRM could not support outbound or automation. Scaling activity on top of it would have amplified noise, not results.

Phase 1: making the data safe to operate on

We built an automated deduplication system for contacts and companies in Clay, executed through HubSpot’s API.

Duplicates were identified with tier-based matching, moving from high-confidence signals down to lower-confidence ones.

For contacts: exact email matches, then name and domain combinations, then name plus company plus phone.

For companies: domain and root-domain matching, then company name normalization and comparison.

Only high-confidence tiers merged automatically. Everything below that threshold was flagged for human review rather than guessed at.

The part that made the system usable is what happens before the merge. Every critical association gets backed up first, including deals, notes, meetings and company relationships. A full restoration workflow lets PHI reverse an incorrect merge, recreate the contact and reattach its history automatically.

Nothing in the system is irreversible. That is what made it safe to run against a live CRM that the business depends on.

At the end of Phase 1, PHI could clean and deduplicate their entire CRM in under an hour, with no manual risk and no developer support.

- Fully automated — edge cases surfaced for manual review - Reversible by design — full restoration if a merge is wrong - Safe to rerun anytime — as new data enters the CRM - Owned internally — PHI’s team runs it, not an agency

Phase 2: turning records into decisions

With the foundation stable, the objective changed. The CRM had to stop storing information and start directing commercial action.

We built an automated system that enriches contacts and companies, evaluates them against PHI’s ICP, and assigns priority based on fit and intent.

The scoring model reflects PHI’s real criteria, not a generic point system. Company fit, recent activity, lifecycle stage and commercial context combine into one interpretable outcome, and contacts are labelled Hot, Warm or Cold directly inside HubSpot.

The CRM now answers one question consistently: who should we talk to next, and why.

What changed

We now have a much healthier database, with a personalized scoring system and a real capacity to prioritize leads on real criteria.

Guillermo Peláez, Senior AE & Marketing Manager at PHI

The system outcomes are measurable: a deduplicated CRM, enriched and validated contact and company data, documented fit and intent scoring, automation-ready workflows.

The team impact is harder to put a number on and matters more. Less time spent questioning data. Faster prioritization. Workflows that can be automated without fear of breaking something downstream. A shared definition of what a good lead actually is.

Why it worked

The project treated data problems as system design problems rather than as a backlog of cleanup tasks.

Most CRMs fail not because of the tool but because there is no explicit logic for how data should be trusted, enriched and used. PHI’s system works because data is validated before it is used, scoring reflects real business context, and the decisions are encoded in the CRM instead of living in someone’s judgment.

They now have a CRM that scales with growth instead of breaking under it, the ability to layer outbound, attribution and routing on top of it with confidence, and full ownership of the infrastructure.

This was not a cleanup project. It was the installation of a revenue-ready data system.

Per a qui és això

Companies where the CRM has been running long enough to accumulate the damage.

Where segmentation cannot be trusted, reporting produces different numbers depending on the view, and nobody wants to touch the database because the risk of breaking something is real.

If your team has decided that doing nothing is safer than making changes, the problem is no longer the data. It is that there is no safe way to operate on it.