The old way
Six vendors
$1M–$10M a year · 7 contracts
Catalog and governance, data quality, modeling and ontology, pipelines, AI and context generation.
Time to live
9+ months
The Ziggy intelligence platform
Own the platform, own the code, plug and play with your own LLM
Ziggy UI
0:00.0 / 0:00.0
Ontology
Build an ontology for Brightfield from its existing data. Use the warehouse's mart tables as the authoritative source, and map each concept back to the source system it starts in.
Here is a first ontology for Brightfield: 5 domains, 33 concepts, 359 properties and 49 relationships. Every mart table is the authoritative source of at least one concept, and 26 concepts also point back to the source table they start in.
| Domain | Concepts | Steward |
|---|---|---|
| Commercial | Product, Brand, Category, Customer, Store, Order, Invoice, Sell-out, Sell-through, Promotion, Promotion Plan, Promotion Result, Price Agreement | Luis Ortega |
| Supply Chain | Site, Plant, Distribution Center, Shipment, Inventory, Forecast, Service Level, OTIF, Fill Rate | Hannah Brooks |
| Finance | GL Account, GL Amount, Cost Center, Fiscal Calendar, Standard Cost, Trade Spend | Omar Haddad |
| Consumer | Consumer, Consumer Case | Grace Liu |
| People | Employee, Sales Region, Sales Territory | Grace Liu |
Two choices worth checking:
It's all in a changeset, so nothing is live yet. Anything to change before it goes to the stewards?
Add three business terms that have no data behind them: TPR, Sell-in and KAM.
Added as Business Terms: each has a definition and synonyms but is never mapped to data.
| Term | Abbreviation | Domain | Definition |
|---|---|---|---|
| Temporary Price Reduction | TPR | Commercial | A promotion mechanic: the shelf price is cut for a limited time, funded by Brightfield |
| Sell-in | Commercial | What Brightfield sells to its customers: orders, shipments and invoices | |
| Key Account Manager | KAM | People | The Brightfield employee who owns the relationship with a customer and plans its promotions |
Fill rate is also called case fill. And give the fiscal calendar the abbreviation 4-4-5.
Done.
| Concept | Change |
|---|---|
| Fill Rate | Synonyms "case fill" and "case fill rate" added |
| Fiscal Calendar | Abbreviation 4-4-5, and "4-4-5 calendar" added as a synonym |
A search for "case fill" or "4-4-5" now finds them, in the glossary and when you ask me.
Looks good. Send it for approval.
Sent. Each steward reviews the domains they own: Luis has Commercial, Hannah Supply Chain, Omar Finance, and Grace Consumer and People. Once they approve, the ontology is live: it shows in Explorer and the glossary, and I use it to answer questions about Brightfield's data.
Ask anything about your data
0:00.0 / 0:25.3
Natural language
Ask anything about your data
0:00.0 / 1:08.0
Three ways to get AI-ready data
The old way
$1M–$10M a year · 7 contracts
Catalog and governance, data quality, modeling and ontology, pipelines, AI and context generation.
Time to live
9+ months
In-House build
$1M · a team of 5 FDEs
Requires deep domain expertise from every data practice. You get sucked into early delivery before it’s ready. And will it really work in the end?
Time to live
18 months
Ziggy
From $50K
Starter Kit
Yearly
From $100K
Source Code
One-time fee
Complete AI data readiness in one AI-native solution, represented logically in an ontology graph, with the source code.
Time to live
Days
Speed
0:00.0 / 0:00.0
Compliance
We are US consumer packaged goods (CPG) food and beverage manufacturer. Which US regulatory compliance reports could we produce from Brightfield's existing data? Give me two tables: one for reports the current data fully supports, and one for reports it doesn't, saying what data is missing for each.
Brightfield's current data fully supports 3 US compliance reports. Another 19 need data it doesn't hold: 4 small gaps, 5 medium and 10 serious.
Two dates are close. The FDA food facility registration renewal window closes on 31 December 2026, and the data has the plant and warehouse addresses but not the other details the form asks for. California's food additive ban starts on 1 January 2027, and with no ingredient lists you can't check which products it hits.
Fully supported (3)
| Report | For | Built from |
|---|---|---|
| Price and promotion fairness self-check (Robinson-Patman Act) | FTC | Discounts, promotion funding and display fees for each chain, by channel |
| Answers to requests to see or delete personal data | State privacy laws (California and others) | Consumer contacts, complaints and rebate claims; HR records for California staff |
| Sales by destination state, for state income-tax apportionment | State tax departments | Invoices by delivery store, 43 states |
Not supported (5 of 19)
| Report | Gap | What's missing |
|---|---|---|
| State additive bans and warnings (California, West Virginia, Texas); FDA Red No. 3 ban | Serious | Ingredient list for each product |
| Nutrition Facts, allergen and "healthy" claim checks | Serious | Recipes, ingredients, nutrient values and allergens |
| Supplier traceability records, including FSMA 204 for the peanut-butter pretzels | Serious | Ingredient and packaging suppliers, and what each delivered, by lot |
| Recall list of stores that got a lot | Medium | Lot code on store deliveries (lots can only be traced as far as the warehouse today) |
| FDA food facility registration renewal | Small | DUNS numbers, owner and emergency contacts, FDA food category codes |
I'd close the four small gaps first: each needs only one or two new fields, and together they unlock four reports.
Yes, list them
Closing the four small gaps takes 8 new fields: 2 on drink products, 1 on stores, 3 on plants and warehouses, 1 for the company, and 1 new list of declared emergencies.
| Report | Field to add | Added to | Where the answer comes from |
|---|---|---|---|
| Bottle deposits | Container: aluminum can, plastic bottle or glass bottle | Each drink product | Pack specification |
| Sweetened-drink taxes | Sweetener: sugar, artificial, both or none | Each drink product | Ingredient label |
| Sweetened-drink taxes | Inside the taxing city's limits: yes or no | The 33 stores in Seattle, Philadelphia, San Francisco and Oakland | Each store's location checked against public city boundaries |
| Price-gouging check | Declared emergencies: state, counties, start and end dates, allowed price rise | A new list | Public emergency declarations |
| FDA registration | FDA registration number and DUNS number | Each plant and warehouse | Facilities team; Dun & Bradstreet |
| FDA registration | Emergency contact: name, phone and email | Each plant and warehouse | Facilities team |
| FDA registration | FDA food categories | Each plant and warehouse | Worked out from what each site makes or stores |
| FDA registration | Owner or operator: legal name, address and contact | The company, once | Legal team |
Everything else these reports need is already there, such as drink sizes, pack counts, store states, prices over time and site addresses. I'd add these fields to the Brightfield data so all four reports can be built and shown in Ziggy.
Shall I add them?
Ask anything about your data
For in-house teams
Security
Review the two fixes to the read-only SQL check before we merge them.
The review found that the two fixes weakened the read-only check: one Serious problem and four Medium ones. The new comment-and-quote reader now decides what counts as code, but each database's rules for it were written on the premise that they could only ever make the check stricter.
| Security finding | How bad |
|---|---|
| On MySQL and MariaDB, a hidden second statement with a DELETE now passes the whole check. Only the database driver's default setting stops it running. | Serious |
| On Postgres and SQL Server, nested comments can hide a second statement from the text check; only the parser still catches it. | Medium |
On SQL Server, ]] inside a bracketed name can hide code; only a parse error catches it. | Medium |
| Some per-database comment and line-ending rules are guesses that now let text through rather than refusing more. | Medium |
| A metric measure was promised to refuse any semicolon, but now accepts one inside a quoted value or comment. | Medium |
| On Databricks, words inside an optimizer hint are no longer checked. Nothing is reachable through it today. | Small |
| Product finding | How bad |
|---|---|
SQL ending in ; -- a comment is still refused | Small |
| Two efficiency tidy-ups (the same text read several times; one list repeated) | Small |
Four test gaps: nested comments, decoy /*! text, a bare carriage-return line end, and ]].
Patching these one by one would keep trusting a reader that can disagree with the database. My recommendation is a safer design for the one fix round:
If the reader ever misjudges a comment, the worst case is that text is dropped or refused. Nothing unchecked can reach the database. That fixes the ESG pipeline steps and the comment-word problem. A semicolon or write word inside a quoted value would stay refused, as before.
Shall I do the fix round that way?
Ask anything about your data
Replace
Book a demo, start a trial, or talk to us about partnering.