Useful AI inside a company depends less on which model you pick and more on whether the operational data underneath is structured, connected, and trustworthy. Most companies aren't there. Getting them there is mostly plumbing work: connecting systems, consolidating data, and building a single source of truth. Our work with Knoxx Foods is a case study in exactly that.

Knoxx Foods is a Sydney-based B2B importer and supplier of premium bulk food ingredients. With over 15 years in the industry, they supply processed tomatoes, rice, dehydrated vegetables, coconut products, and more to food processors, ready-meal manufacturers, and foodservice distributors across Australia and New Zealand. Their products are sourced from trusted producers across India, China, Italy, Indonesia, and Thailand. They serve customers spanning quick-service restaurants, catering providers, bakeries, aged care institutions, and wholesale distributors.

Knoxx was already further along than most when we came in. For a small business in a traditionally non-tech industry, they'd built impressive foundations: a data warehouse, dashboards for reporting, and a set of Custom GPTs already deployed against specific operational tasks. Just as importantly, leadership was actively looking for places automation and AI could accelerate the team, treating it as something to embrace. That mindset, more than anything else, is what made the partnership work: this was never a zero-to-one story.

What had grown faster than everything else was the operational layer underneath. A diverse B2B customer base, international sourcing, multi-channel distribution, all run by a lean team using a collection of tools that had each been adopted for good reason but had never been wired together. The team was spending hours each week manually copying data between platforms just to keep the lights on.

This is the story of how we untangled it: by building bridges where none existed and consolidating operations into a single hub that's now ready for AI.

Knoxx's operations were spread across a patchwork of tools that had each been adopted for good reasons but had never been wired together. Orders, warehouse logistics, customer relationships, and accounting each lived in separate systems with no connections between them. And the gaps between all of them were filled by spreadsheets, manual exports, and a lot of copy-and-pasting.

The consequences were predictable. Data entered in one system had to be manually re-entered in another. Records were duplicated across systems with no common identifier. And nobody had a single view of what was actually happening across the business.

None of this was unusual. In fact, it's the norm for lean teams that have grown quickly. You adopt tools as you need them, and by the time you realise they don't talk to each other, you're too busy to fix it.

This was the first piece of work we took on together. The Knoxx team regularly produces presentation decks for a variety of purposes, drawing on data from their existing BI setup. The data was already there, but turning it into polished decks was still manual: team members would screenshot charts from dashboards and paste them into slides, one deck at a time. So while the data itself was automated, the overall process was still a significant recurring time sink.

This kind of 'last mile' manual bottleneck is exactly what we tackled in our work with NTUC, where a team was doing similar copy-and-paste work to turn raw social media exports into AI-powered insight reports. In both cases, the thinking was handled; it was the plumbing in between that needed automating.

We automated the rest of the pipeline. A team member makes a couple of selections in a simple table interface, clicks a button, and a Zapier workflow pulls the relevant data from the data warehouse and sends it to Gamma, which generates a polished, branded presentation deck automatically. What used to involve screenshotting dashboards and assembling slides now takes a single click.

Knoxx's order management platform had no API at all. No export function, webhooks, or documented way to programmatically access the data. For most consultancies, this would be a dead end. We approached it as a fun challenge worth tackling. We built a custom bridge to extract and sync order data reliably through a system that, on paper, had no programmatic interface at all.

The warehouse logistics platform was a different challenge. It had a public API, but the off-the-shelf integration options didn't cover what Knoxx needed. So we built a custom Zapier integration from the ground up using the platform's API documentation.

With both platforms now properly connected, the real goal came into reach: making them work together as if they were one system, with no manual re-entry or copy-pasting between browser tabs.

All of this worked, and at first, we thought that building these integrations was the whole job. But over time, a bigger problem emerged: the order management system was full of features the team didn't really need, and the way it worked didn't match how Knoxx's customers actually preferred to do business. We needed a deeper rethink that would balance ease of use for Knoxx's customers with time and efficiency gains for their own team.

That platform was one example of a common pattern we see: the price on the invoice is rarely a tool's full cost. Paying for capabilities that don't fit how a team works is a hidden cost in itself, but it's not the only one. You adopt tools one at a time, each for a good reason, and collectively they accumulate into tool sprawl: the visible cost is the SaaS bill, with per-seat licences for CRMs, project trackers, reporting dashboards, and communication tools. As teams grow, those bills grow linearly, and the natural response is to ration access. Only sales gets CRM seats. Only managers get the reporting tool. Only finance touches the accounting platform.

That rationing in turn creates the bigger, less-visible cost. When access is gated by licence, knowledge is gated with it. The salesperson can't see whether finance has invoiced the customer. Ops doesn't know what the salesperson promised. Transparency and shared understanding erode, and decisions get made on partial information. Then there's the cognitive overhead: remembering which tool holds what, context-switching across half a dozen UIs all day, working around the data silos that form between systems that don't talk to each other. The true cost of sprawl is several multiples of the line items on the SaaS invoice.

Consolidation is about efficiency, but it's every bit as much about transparency, accuracy, and paying only for the capabilities you actually use.

At Knoxx, the answer was consolidation: rather than continuing to manage each function in its own tool, the core pieces of Knoxx's operational data are being brought together in a single workspace configured as the central source of truth: Notion.

It's important to call out that we're not trying to replace every specialist tool with Notion. There are some jobs that genuinely warrant a dedicated tool that Notion can't replace. I would never, for example, advocate replacing QuickBooks or Xero with Notion, and warehouse inventory management requires a whole set of specialised operations that continue to justify a purpose-built tool. But other pieces of the operational layer can move into one place, where every team member sees the same data and updates propagate instantly.

We started with the order management system, which has now been fully replaced by a custom system built natively in Notion, living alongside the rest of Knoxx's operational data. Besides removing a technical headache, this migration took a significant annual platform cost off the table.

Here's what that foundation makes possible. With orders, products, customers, and pricing all modelled cleanly in one place, we could finally put an agent on top of the data, and the first one we built tackles the most repetitive job in the business: order entry.

A Notion Custom Agent now watches the orders inbox. When a customer emails an order, exactly as they always have, the agent reads the message and creates the order directly in Notion. If anything is ambiguous, it doesn't guess; it flags a person. Customers changed nothing. The data entry just stopped being someone's job.

What makes this even more powerful is combining it with Zapier workflows. Once the agent creates an order and it's ready for processing, the downstream fulfilment, invoicing, and record-keeping follow automatically.

It also reflects a deliberate architectural choice. Business logic that benefits from transparency and team ownership lives in Notion in natural language the entire team can read and understand. The deterministic plumbing that moves data between systems stays in Zapier. Agents are applied where judgement matters, deterministic automation where it just needs to run.

Consolidating your operations into a structured, well-organised hub pays off twice: in efficiency today, and in what becomes possible tomorrow.

AI agents that can draft customer communications, surface insights from order patterns, or prepare meeting briefs from historical data need structured, accessible data to work with. They can't help you if your customer history is split across five platforms and three spreadsheets.

By building a clean, relational data model in Notion, where customers link to orders, orders link to invoices, and communications link back to contacts, we've created the foundation for AI-powered workflows. The company can now deploy AI skills and agents that actually understand the business context, because the data is structured in a way that makes sense.

Read this as a story about a team that was already thinking the right way; the tooling is secondary.

Knoxx's leadership came to every conversation with their own ideas about where automation and AI could help. They weren't waiting to be sold a solution; they were already mapping the surface area themselves, and our job was as often to refine and sequence their thinking as it was to introduce new ideas. When we recommended something, decisions came back fast. When something didn't work the first time, they were curious about why rather than nervous about the cost.

That posture matters more than the particular tool stack or data infrastructure that you start with. The best automation and AI enablement work is made possible only when the client treats it as an investment in how their team operates. A lot of what's in this case study wouldn't have been possible without leadership that was genuinely leaning into the work.

This kind of work doesn't have a finish line. It's a continuous process of removing the friction that costs the team hours every week, and each system we connect surfaces the next bottleneck worth tackling.

The change is already tangible. Orders that were once re-keyed by hand from a customer's email are now created automatically, and proof of delivery is verified and filed without anyone lifting a finger.

And the more of the business that lives in one connected workspace, the more useful the whole thing becomes: the team can pull up the full picture in seconds, instead of stitching it together across tools by hand.

For lean teams, especially in the APAC region where we work with startups and growing businesses every day, the lesson is straightforward. You need someone to connect your existing tools, consolidate the data that matters, and build the foundation for what comes next.

That's what we do at workFlowers.

If your operations feel like they're held together by spreadsheets and good intentions, let's talk.