top of page

Introducing the Datitude AI Gateway: From Trusted Data to Trustworthy Answers

Writer: Gavin Wilkinson
Gavin Wilkinson
4 days ago
8 min read

Ask an AI tool that’s connected to your data how much stock you're holding, and it will answer in seconds. Clear, confident, nicely formatted. In one of our tests, it was also 35 times too high.


That's the type of challenge we've solved with the AI Gateway, now live on the Datitude platform. It connects retailers’ AI tools to their unified data and helps AI agents understand that data before they're allowed to query it.


In this article, we explain why pointing AI at data isn't enough, why the quick fixes fall short, how the Datitude AI Gateway works, and what retailers like Loaf are already doing with it.


Pointing AI at data isn't enough

AI makes it possible for anyone to ask questions of their data in plain English. That's hugely appealing for retailers. But a quick answer is only useful if it's right. So before opening the Datitude platform up to AI tools, we wanted to know whether their answers could be trusted.


We connected a leading LLM (large language model) to the platform using the most sensible set-up we could think of. Secure access. Read-only, so it could query data but never change it. No personally identifiable data within reach. Tables and columns with clear names.


Then we asked it everyday retail questions about stock, weeks of cover and customer spend, and checked its answers against figures we knew were correct.


The speed was impressive: plain-English questions answered in seconds, across all of a retailer's data. The accuracy wasn't. Some answers were badly wrong, and delivered with exactly the same confidence as the right ones!


That's the bit that matters. A wrong answer that looks wrong gets caught. A wrong answer that looks right ends up in a board pack.


Why AI gets data wrong

LLMs are brilliant with language. But they’re probabilistic - they predict likely words, rather than recalling stored facts. Give one a table and a question, and it will make a reasonable-sounding guess about what each column means and how to add it up.


Retail data is full of traps for reasonable guesses. In one test using demo data, each customer's total spend from personal shopping appointments was stored as customer-level data against every one of their order lines. The AI added it up line by line, counting each customer's total several times over. It reported £2.42m. The real figure was £250k. And average spend per customer was reported as £3,876 versus £400 actual.


The data itself was fine. What was missing was understanding in the form of metadata: the data about the data that explains the overall data-layer concepts, what each table and each field means, the level it's recorded at, how it should be counted, and the gotchas to avoid.


The quick fixes, and why they fall short

There are two common fixes, and most retailers will reach for them first. We tried them too.


Better prompts. 

You can tell an AI a lot in the instructions you give it. But prompts get long quickly, and they're even less repeatable than a copy-and-paste Excel formula. Two people asking the same question slightly differently - or using different models - can still get different answers.


AI skills.

Skills are reusable instructions that tell an AI tool how to behave, for example as a merchandising specialist. They're powerful, but they sit with each user or organisation. Once shared, how they're used and kept up to date varies from one business to the next, and from one person to the next.


Both help, up to a point. Neither guarantees the AI understands the data. It’s why we believe quality control has to live in the unified data platform itself, where it applies to every question, from every user, in every AI tool.


Introducing the Datitude AI Gateway and what it does

AI Gateway is the result. It's an MCP (Model Context Protocol) server, built to connect AI tools to the Datitude platform. MCP is an open standard that lets AI tools connect to data and systems. It's just the plug, though. What matters is what it’s being plugged into.


Before an AI tool can query anything, AI Gateway gives it Datitude's semantic layer: tested, trusted metadata for every table and column, together with overall ‘how to use’ instructions. That means plain-English definitions, how the data relates, and advice on which table to use depending on the question. Each definition is managed once, in the central data dictionary, so the same field means the same thing wherever it appears. It works in three steps:


  1. Discover. The AI sees which tables are available and what each one is for.

  2. Understand. It reads the columns and definitions for the table it needs.

  3. Query. It asks a structured question, which AI Gateway turns into a fast, safe query. Results come back with the context needed to read them properly.


Access is secure and read-only throughout. We’ve also chosen to keep personally identifiable data out of reach for now.


Diagram of the Datitude AI Gateway: an LLM connects securely to Datitude’s MCP server, which uses the semantic layer to discover tables, read table schemas and run safe queries on the unified data lakehouse, returning data with semantic context.
The Datitude AI Gateway: an LLM connects securely to Datitude’s MCP server, which uses the semantic layer to discover tables, read table schemas and run safe queries on the unified data lakehouse, returning data with semantic context.

Same question, two answers

We ran another test. Both with the same AI model, the same data and the same question: "How much stock are we holding, by department, and how many weeks will it last?" The only change was whether it had the semantic layer.


Table showing closing stock and weeks of cover by department (trading week 40): without and with the semantic layer

Closing stock and weeks of cover by department (trading week 40):

without and with the semantic layer

Department

Stock units without semantic layer

Stock units with semantic layer

Weeks cover without semantic layer

Weeks cover with semantic layer

Womens Clothing

2,211,229

52,266

130.2

104.3

Shoes

1,623,084

40,886

21.1

23.6

Mens Clothing

912,910

34,915

26.6

34.6

Gifting Edit

983,070

33,301

28.7

33.3

Tabletop

738,659

21,235

12.4

11.6

Jewellery and Watches

494,200

16,064

1,280.3

1,529.9

Other

643,964

15,928

11.6

9.7

Kids Clothing

356,104

8,144

n/a (no sales)

n/a (no sales)

Unknown

20,872

3,980

5.2

56.9

Garden

187,236

3,546

32.5

29.1

Total

8,171,328

without semantic layer

230,265 with semantic layer

28.4 without semantic layer

29.1

with semantic layer

Figures are from demo data. The ratio is real.


Without metadata, the AI added up closing stock for every week of the year, rather than taking the latest week. Stock is a snapshot, not a running total, so the answer came out about 35 times too high.


Here's the sneaky part. Overall cover looked about right, because the errors cancelled each other out. By department, they didn't. One department appeared to have five weeks' cover when it actually had more than a year's worth. Markdown, intake and buying decisions get made at that level.


With the semantic layer, the AI used the right week, worked out cover from recent sales rates, and showed how it got there. That’s because it guides the AI to the right data, and the right calculation.


What Datitude’s AI Gateway makes possible

With the right context in place, AI becomes a useful analyst. Not just for data teams, but for anyone who needs a trusted answer they can act on. We’ve also developed specific Datitude Skills in Claude (a general data analyst skill and merchandising analyst skill being examples) to build on the semantic layer. They’re designed to add an extra layer for checking definitions and validating figures and are maintained by Datitude.


Loaf has already built a business-specific contextual layer alongside the Datitude AI Gateway. It brings together Loaf’s business context with its unified data, so commercial decisions are based on hard evidence rather than opinion and gut feel.


Here’s what Louis Adamou, former Technology Director, had to say:

"I think we’re really on the bleeding edge of what’s possible.

When we started working with Datitude in 2025, our goal was properly governed and cleaned data, stored in one place in Datitude, with a series of traditional BI dashboards available through the platform. Eighteen months ago, we thought that was as good as you could aspire to as a business.


Where we’ve landed with Datitude’s AI Gateway, so that Loaf can interrogate its own clean data that Datitude’s storing, governing and defining, has been quite revolutionary. We’re just scratching the surface, but we’re already getting quick answers using our chosen LLM and building out dashboards. By combining that with contextual information about the business, we can start to infer meaning out of the numbers and understand the options for actions we could take.

It’s been eye-opening. And it’s shown the real value of Datitude is in the infrastructure, governance and the active partnership."

The AI Gateway changes everyday work too. In our demo, a single sentence produced a weekly sales, stock and intake (WSSI) dashboard, both in Claude and in Excel, without anyone needing to know how the data is modelled. The semantic layer’s metadata does that job for them.


Weekly sales, stock and intake (WSSI) dashboard built by Claude from a single plain-English request using Datitude’s AI Gateway, showing KPI cards and a weekly trend chart. Demo data.
Weekly sales, stock and intake (WSSI) dashboard built by Claude from a single plain-English request using Datitude’s AI Gateway, showing KPI cards and a weekly trend chart. Demo data.

In another example we asked Claude to create a full price discipline scorecard showing how much of sales is made at full price and what markdown and promotion are costing, compared with the previous year.


Full-price discipline scorecard built by Claude using Datitude’s AI Gateway, showing the share of sales made at full price and the cost of markdown and promotions compared with last year. Demo data.
Full-price discipline scorecard built by Claude using Datitude’s AI Gateway, showing the share of sales made at full price and the cost of markdown and promotions compared with last year. Demo data.

The AI Gateway has been live since September. We've shown it to clients one to one over the past couple of months and presented it to clients at a roundtable event at the beginning of October.


"Building the Datitude AI Gateway was exciting and, at times, a bit scary. The first answers looked brilliant, until we checked them. That taught us more than any early success would have: AI needs to understand data, not just see it. Connecting AI to databases is very much the quick and easy bit; testing answers, rebuilding and testing again until the answers were right is the hard bit.

The reaction from clients tells us it was worth getting right. As the CEO of one of our retail clients put it: 'you have just given me everything I need on a silver plate'." Gavin Wilkinson, Co-founder & Director, Datitude


Included with the Datitude platform

Datitude AI Gateway is included in the Datitude platform at no extra cost. Retailers on the platform can connect their unified data in the platform to any AI tool that supports MCP, such as Claude, so they can choose their preferred AI tool.


Every tool connects to one governed source of data, rather than many separate systems. So, access, guardrails and governance are managed in one place. And because quality control lives in the platform, not in individual prompts, everyone works from the same definitions and gets consistent answers.


Diagram of Datitude’s unified retail data platform, from integration, data and semantic layers up to the intelligence layer, which feeds BI and reports, the AI Gateway and data feeds to third-party systems.
Diagram of Datitude’s unified retail data platform, from integration, data and semantic layers up to the intelligence layer, which feeds BI and reports, the AI Gateway and data feeds to third-party systems.

Strong foundations first

AI makes everyone a speedy analyst. Not necessarily an accurate one. The difference isn't the AI tool. It's the data, and the metadata, behind it. With the semantic layer built into the Datitude platform, AI turns trusted data into trustworthy answers.


That's always been the point of a unified data platform: timely, consistent, correct data. AI Gateway extends that to the AI tools people already use.


Whoever looks after your data, three questions are worth asking before you let AI loose on it:


  1. Is the data unified, or will the AI be piecing answers together from separate systems?

  2. Does your business even agree on definitions of key metrics across departments, or are these frequently argued over?

  3. Is quality control built into the data, or does it depend on every user writing the perfect prompt?


Datitude clients using AI for data analytics and business intelligence are in safer hands with AI Gateway. If you're not a Datitude client and want to see what AI-ready data looks like in practice, get in touch.



About Datitude

Datitude's award-winning, unified retail data platform provides the unifying framework to enable retailers and brands to get what they need from their data.


We connect, integrate and unify data into the single source of the truth.


All your data, in one place, AI-ready. No silos. Data where it needs to be, on time, in the right format.


Trusted by mighty retailers and fast-scaling brands to provide a strong, AI-ready, data foundation for clearer insight, faster decisions and more profitable growth.


Data by Datitude | Ready for AI

bottom of page