A HUBSPOT IMPACT AWARDS SUBMISSION, TECHNICAL EXCELLENCE

In short

Spinfluence, a HubSpot Solutions Partner, set out to solve a deceptively hard problem for Q Report, Australia's pioneering standalone jewellery insurance provider. Q Report's growth depends on a national network of jeweller partners who introduce their customers, yet the team running that network had almost no visibility into how individual jewellers were performing.

We bridged Q Report's bespoke insurance CRM, Q Management, into HubSpot, and rebuilt the jeweller partner record from the ground up as a live, calculated, 360-degree view we call Jeweller 360°. The technical heart of the build is a layer of custom calculated properties and a custom-code calculation engine that turns raw policy and referral data into per-jeweller intelligence in near real time.

What makes this submission unusual is how it was built. The HubSpot data architecture, every custom property, the deal pipeline, the integration objects, and the calculation logic that drives all the metrics, was deployed through the HubSpot API by a custom AI infrastructure agent that Spinfluence designed, built and trained. The agent is powered by Claude, operated from the command line, and version-controlled in a GitHub repository like any other software project. This is not a standard implementation. It is a custom integration between two systems, a custom CRM architecture, and a piece of custom AI tooling that did the engineering work.

This is also a living system. It is being extended continuously as Q Report's team discovers more value in it.

The challenge

Q Report has been a name in Australian fine-jewellery insurance for over twenty years. Its single most valuable growth engine is its jeweller partner network: independent boutiques and premium jewellers who introduce their customers to Q Report at the point of sale. This B2B2C channel is the backbone of the business and delivers the overwhelming majority of new customers at a fraction of the cost of direct acquisition.

There was one serious problem. The people responsible for that network, a sales team that spends most of its week on the road visiting jewellers, were effectively working blind.

When a sales lead opened a jeweller's record, they couldn't answer the questions that mattered. How many policies has this jeweller actually generated this month, this quarter, this year? Are their referrals trending up or quietly falling away? What proportion of their referrals are converting into active policies? When were they last visited, and are they overdue? Is this relationship healthy, slipping, or already at risk of being poached by a competitor with a larger network?

None of that lived in HubSpot in any usable form. The policy and referral data sat in Q Management, Q Report's bespoke, in-house insurance CRM, the system of record for every policy. HubSpot, where the sales relationship was managed, had no structured connection to it. The two systems didn't talk.

There was also a regulatory barrier. As an Australian financial services business, Q Report operates under strict data-handling expectations, and there had historically been real limits on what policy data could be pushed into a third-party platform. The unlock came when HubSpot stood up an Australian data centre. With data residency in Australia, it finally became defensible to bring policy-derived data into HubSpot, and that opened the door to building something genuinely technical on top of it.

So the challenge was threefold: bridge two systems that had never been connected, design a CRM architecture rich enough to turn raw policy events into meaningful partner intelligence, and do it all in a way that respected Australian compliance from the first line of code.

The solution

Bridging two systems

The foundation is a custom integration between Q Management and HubSpot, built directly against HubSpot's CRM REST API. Q Management pushes raw deal data, referrals and policies, into HubSpot as events occur. Each jeweller in Q Management is matched to its HubSpot company record on a shared jeweller_id key, so every policy and referral lands on exactly the right partner.

The integration is built to behave well in production: it searches for an existing deal before creating one so status changes update the right record rather than spawning duplicates, it handles historical backfill through batch endpoints, and it respects HubSpot's rate limits. Crucially, it is compliance-first by design. Deals are named by policy or referral number only, never by customer name, in line with ASIC requirements, and the entire data model was built to keep customer-identifying information out of places it shouldn't be.

Raw deal data is only half the story, though. The real value is in what HubSpot does with it once it arrives.

The calculated properties, the engine of the whole system

This is the technical core of Jeweller 360°, and it is the part the AI agent built.

A jeweller's record is not a static profile. It is a set of custom calculated properties that continuously translate raw deal activity into the numbers the sales team actually uses. The agent created this entire property architecture in HubSpot via the API, dozens of custom company and deal properties, organised into clear groups, every one of them purpose-built for this network. The metric layer includes:

Policies sold, counted across five rolling windows: this month, last month, last 90 days, last 12 months, and the previous 12 months, so year-on-year movement is visible at a glance.

Total referrals, counted across the same five windows, so the team can see not just conversions but raw introduction activity and whether a jeweller is engaging at all.

Referral conversion rate, calculated on a rolling three-month cohort basis, taking every referral from the three full calendar months before the current month and measuring how many of those specific referrals converted to an active policy. A cohort approach was chosen deliberately so that one unusually quiet or busy month can't distort the picture.

Average policy value over the last 90 days, days since last referral, and a referral trend indicator that flags each jeweller as growing, stable, or declining based on a year-on-year comparison.

A composite health score and status, a single 0-to-100 figure, surfaced as a simple green / amber / red indicator, built from six weighted inputs: referral trend, conversion rate, referral recency, visit recency, whether they've produced a policy this month, and tenure. It answers "how is this relationship doing?" in about three seconds.

Visit cadence properties, a visit-frequency target derived from the jeweller's tier and an overdue indicator, so the team can see at a glance who's due, who's overdue, and how often each partner should be seen.

Every one of these is a calculated property. Nobody types these numbers in. They are derived from the underlying deal data, which means they never go stale and never depend on someone remembering to update a field. That design choice, calculated over manual everywhere it was possible, is what makes the whole system trustworthy enough to run a sales team on.

Behind the metric layer sits a custom-code calculation engine. The agent authored the logic that powers it: an event-driven routine that, whenever a deal is created or changes stage, recalculates every performance metric for only the affected jeweller, paginating through hundreds of associated deals, filtering to the right pipeline, bucketing each one into the correct time windows, and writing the results back to the company record in near real time. It replaces the obvious-but-wasteful alternative of recalculating every partner in the network on a nightly batch.

From data to a decision-ready record

On top of the property and calculation layer, the team configured the HubSpot workflows and the reporting dashboard that put this intelligence in front of users. These were assembled in-app, by hand, because HubSpot does not yet expose workflow or report creation through its API, an honest division of labour between what the agent could deploy via API and what still requires the HubSpot UI. The workflows keep visit cadence, health status and tier-upgrade prompts current; the dashboard rolls the calculated properties up across the whole network into reports on referrals, policies, conversion, pipeline, trends, health and visit status.

The result is a company record where opening a jeweller tells you everything: who they are, how they're performing, whether they're healthy, when they were last seen, and where the opportunity is, all driven by properties the agent built.

A genuine 360-degree view

Put together, Jeweller 360° gives the sales team something they simply did not have before: complete visibility and transparency across the entire partner network, from a single record.

They can now see how many policies each jeweller is generating and how many customers each one is introducing. They can see conversion quality, not just raw volume. They can see visit cadence, who's on track and who's slipped through the cracks. They can see jeweller performance and sales-team performance side by side. And because the same calculated properties roll up to a dashboard, leadership can see the health of the whole network at once.

Most importantly, it surfaces opportunity. The team can now see, clearly, every place there's room to grow a jeweller, scale up a strong performer, or recover one whose activity is fading, the recover-grow-scale picture that was invisible when the data was trapped in a separate system. All of it was built compliance-first, in Australian English, on Australian-resident data, with the language and data rules of a regulated financial services business baked in from the start.

Results and impact

This is an evolving system rather than a finished project, and that's the point. Q Report's team keeps finding new value in it, and we keep building on it as they do.

The clearest signal so far is adoption. Usage of the jeweller records has climbed sharply as the team has come to trust the numbers. When the data updates itself and is genuinely reliable, people start opening the record before every call and every visit, which is exactly the behaviour change we were after. For the first time, the team can see the full set of opportunities in front of them to grow, scale and recover jeweller performance, and act on them deliberately instead of by gut feel.

Because the metric layer is entirely calculated and event-driven, that intelligence stays current on its own. There's no manual maintenance burden, no stale fields, and no overnight wait to see the effect of new activity. The system gets more useful the more the network is worked, and the roadmap (multi-location partner structures, commission tracking, deeper onboarding, and extending the same model to the JewelCover brand) keeps expanding as the team asks for more.

AI impact

The most distinctive part of this build is who did the engineering.

Spinfluence designed, built and trained a custom AI infrastructure agent. We didn't use AI to assist a developer at the margins. We built a piece of custom AI tooling and gave it the job of constructing the HubSpot data architecture itself. The agent is powered by Claude and operated from the command line, with its definition, configuration and prompts version-controlled in a GitHub repository, treated as a real software project rather than a throwaway script.

Critically, the agent was trained for this work. It was grounded in HubSpot's full knowledge base, the platform's objects, properties, association model, API behaviour and limits, and on everything it needed to know about Q Report and the JewelCover partner programme: the partner model, the tier and visit logic, the compliance constraints, and the data rules of a regulated Australian financial services business. That combination of platform expertise and deep customer context is what let it produce correct, production-grade configuration rather than generic boilerplate.

Working through the HubSpot API, the agent deployed the custom property model, the deal pipeline and objects, the integration scaffolding, and the calculation logic that drives every metric described above. The calculated-property engine that is the technical heart of Jeweller 360° is, quite literally, the agent's work. This is CRM context architecture for AI in practice: a trained agent reasoning about a real CRM and building against it.

Reports and dashboards can't be created through HubSpot's API at all, so those were built in-app by hand. The workflows were also assembled in HubSpot directly, the agent authored the custom-code calculation logic that runs inside them, while the v4 workflows API remained in beta.

For a Solutions Partner, this is a new delivery model. A trained, API-driven agent can stand up complex, customer-specific CRM architecture faster and more consistently than hand-configuration, while a human stays firmly in the loop for design, judgement, compliance and the parts the API can't reach. Jeweller 360° is the proof that the model works on a real, regulated, multi-system build, and it's the foundation we'll keep extending for Q Report.

What's next

Jeweller 360° is live and in daily use, and it is still growing. Planned phases include parent/child structures for multi-location jewellers, commission tracking once the structure is confirmed across both brands, a richer onboarding and enablement layer, and extending the same calculated-property model to Q Report's sister brand, JewelCover. As the team sees more value, the system grows with them, which is exactly how the best technical solutions earn their place.