# Case study: Productify.ai Product Data Platform

> Productify.ai is our own AI product data platform. Teams send it product names, spreadsheets, or photos of packaging, and it returns e-commerce copy, taxonomy categories, translations, and structured pack data through a web portal, an API, or Zapier and Make. Clients in Europe, Singapore, and South Africa each run more than 8,000 operations a month.

- URL: https://www.happen.software/case-studies/productify-product-data-platform
- Publisher: Happen Software (https://www.happen.software/)

## At a glance

- **Client**: Happen Software (our own product)
- **Category**: Retail & eCommerce · AI & automation
- **Running**: Since 2023 · 3 years
- **Built for**: Teams preparing product content for e-commerce, including agencies working across several workspaces
- **Work types**: E-commerce copy, text transforms, and image digitisation: 13 AI operations in all
- **Data in**: Excel imports, single products in the portal, a REST API, Zapier, and Make
- **Data out**: Results in the portal, Excel exports, API results, and webhooks when batches finish
- **Billing**: Prepaid credits bought by card, with bonus tiers, expiring bundles, and optional monthly credits
- **Clients**: In Belgium and elsewhere in Europe, Singapore, and South Africa

## Outcomes

- **8,000+**: AI operations a month for each client, with larger batches before peaks such as Easter and Christmas
- **3 continents**: Clients in Belgium and elsewhere in Europe, Singapore, and South Africa
- **47**: Languages that product copy and text can be translated into
- **1,083**: GS1 nutrient codes that every digitised nutrition table is checked against

## Context

Every product in an online catalogue needs copy: a description, a headline, features and benefits, keywords, and a category. Food and grocery products also need what's printed on the pack, from ingredients and allergens to the nutrition table, in a structured form. Done by hand, that work is slow, it gets slower as catalogues grow, and it peaks before events such as Easter and Christmas, when catalogues change fastest.

Productify.ai is our answer to that, and a product we own end to end. A team signs in to its workspace, chooses the work it needs, and uploads a spreadsheet or calls the API. Background queues run each AI step in order, the results are checked and tidied, and they come back in the portal, as an Excel export, over the API, or to a webhook. We designed it, we build it, and we run it, the same way we work on systems for our clients. We tailor and fine-tune its output with e-commerce copywriting and product data experts, so the copy, categories, and pack data read the way retailers and shoppers expect.

## The challenge

- **Copy for every product**: Each product needs descriptions, a headline, features and benefits, keywords, and a category, written consistently in the right tone and language.
- **Data printed on the pack**: Ingredients, allergens, certifications, and nutrition tables only exist on packaging, and nutrition data needs standard codes before other systems can use it.
- **Fitting into existing systems**: Teams already work in spreadsheets and their own tools, so the results have to go in and come out without copying and pasting.

## Before and after

| Area | Without it | With Productify.ai |
|---|---|---|
| Product copy | Descriptions, headlines, and keywords written one product at a time. | Written for a whole spreadsheet in one batch, with three marketing descriptions per product to choose from. |
| Categories | Each product placed in a taxonomy by hand. | The closest Google or GS1 categories found, and the most specific one returned with its full path. |
| Packaging data | Ingredients, allergens, and nutrition tables typed out from the pack. | Read from photos into structured fields, with barcodes decoded from the same images. |
| Nutrient codes | A standard code looked up for each nutrient. | Assigned automatically and checked against all 1,083 codes in the GS1 T4073 list. |
| Languages | Copy rewritten for each market. | Translated into any of 47 languages, with UK spelling and metric or imperial units set per workspace. |
| Other systems | Results copied and pasted between tools. | Sent through the API, to a webhook when the batch finishes, or into other apps with Zapier or Make. |

## A catalogue, from spreadsheet to results

1. **Choose**: The team picks the work, such as marketing descriptions, headlines, and categories, with tone, length, and translation settings. Anything a choice depends on, like the marketing description a headline is written from, has to be included.
2. **Try**: A single product can be run straight away in the portal to check the output before committing a catalogue. If a test run fails, its credits are refunded automatically.
3. **Import**: The catalogue goes in from the Excel template or over the API, up to 10,000 products a batch. The cost is estimated and checked against the workspace's credits before anything starts.
4. **Process**: Each AI step has its own background queue, checked every 20 seconds and working on up to 10 products at once. Steps run in order, and each is tried up to 3 times before it's flagged for attention.
5. **Check**: Results are tidied before they're stored: UK spelling and units applied where the workspace asks for them, category paths built from the taxonomy, and nutrient codes checked against the GS1 list.
6. **Deliver**: When the batch finishes, portal users are emailed, any webhook is called, and the results are ready in the portal, as an Excel export, or from the API.

## What we built

- **E-commerce copy**: Marketing descriptions, short descriptions, headlines, features, attributes and benefits, and SEO keywords. Each product gets three alternative marketing descriptions, in one of five tones, at a shorter, default, or longer length.
- **Taxonomy categorisation**: Products are matched to the Google or GS1 taxonomy. The ten closest categories are found by meaning, AI chooses the most specific, and the full category path comes from the taxonomy itself.
- **Image digitisation**: Text extraction, table digitisation, and nutrition table digitisation from photos and scans, returned as editable text and structured tables.
- **Back-of-pack extraction**: From one or more photos of a pack: brand, net quantity, serving size, ingredients, allergens, storage and usage instructions, origin, claims, and 11 dietary and certification flags such as halal, kosher, and vegan. Barcodes are decoded from the images too.
- **Standard nutrient codes**: Each nutrition row gets a GS1 T4073 nutrient code, checked against all 1,083 codes in the list. Known mistakes are corrected, unlisted nutrients get a consistent code, and units are normalised.
- **Rewrite & translate**: Rephrase in a chosen tone, summarise to a target length, or translate into 47 languages. E-commerce batches can be translated as they're written, and English copy can follow UK spelling and metric or imperial units.
- **API, webhooks, Zapier & Make**: A REST API with an interactive explorer, and single-product and batch endpoints for each type of work. Batches can call a webhook when they finish, a Zapier app offers 16 actions, and it connects to Make too.
- **Credits & billing**: Each batch is costed before it starts and checked against the balance. Credits are bought by card with bonus tiers, bundles expire, and workspaces can have monthly credits and their own pricing.

## What each team gets

### E-commerce & content teams

- Three marketing description drafts per product
- Tone, length, and language chosen per batch
- Results to review in the portal or export to Excel

### Product data teams

- Nutrition tables with GS1 nutrient codes
- Ingredients, allergens, and certifications from pack photos
- Barcodes decoded from the same images

### Developers & integrators

- A documented REST API with an interactive explorer
- A webhook when each batch finishes
- 16 Zapier actions, plus Make, for no-code workflows

### Workspace owners

- Members invited, removed, or made owner
- Work types switched on or off for the workspace
- Credit purchases, invoices, and utilisation charts

### Agencies & multi-brand teams

- Several workspaces under one sign-in
- Separate credits, API keys, and members per workspace
- Switching between workspaces without signing out

### Our support team

- Email alerts for any step that fails after its retries
- Credit pricing and monthly credits set per workspace
- AI token usage tracked for every result

## How it was delivered

- **2022, Foundation**: Workspaces with Google and email sign-in, the first AI product descriptions, batches, and the start of the client API, from October.
- **2023, Launch**: Image digitisation and text transforms, credits bought by card, the public API and its documentation, a production deployment pipeline, then nutrition tables and translation.
- **2024–2025, Integrations & packaging**: Table digitisation moved to AI, an AI test suite on real packaging, completion webhooks, the Zapier app, Google and GS1 categorisation, and back-of-pack extraction with barcodes.
- **2026, Modernised**: A container-based deployment path with health checks, and nutrient codes validated against the full GS1 list.

## Engineering

- Every AI task has its own background queue, so a slow step doesn't hold up the others, and the large batches clients send before Easter and Christmas are worked through in order. Steps that depend on another wait for it, and if a prerequisite fails, the steps after it are flagged rather than run on missing input.
- AI output is checked rather than trusted: categories are chosen from candidates found in the real taxonomy, nutrient codes are validated against the GS1 list, and units and acronyms are normalised.
- Every AI call is stored with its request, raw response, and token count, so any result can be traced and its cost understood.
- Credits are reserved when a batch starts, refunded automatically when a test run fails, and can be refunded when a batch is retried.
- Over 100 automated tests, including AI tests run against 54 real label, table, and packaging images, with every nutrition row checked.
- The AI model for each task is swapped as better ones appear, without changing the API.
- Kept current with regular technology, library, and framework updates, along with ongoing performance and security improvements.

Stack: .NET, ASP.NET Core, SQL Server, Angular, Python, Zapier, Make

## Get in touch

Need product data written, checked, and digitised at scale? Productify.ai may already do what you need. If you'd rather have something similar built around your own data and systems, we can do that too. Book a free 30-minute call with one of our engineers.

Contact: https://www.happen.software/contact or sales@happen.software