Step-by-step guide · Microsoft Foundry
Use your organisation's Microsoft Foundry for the AI tools
Organisations that already use Microsoft Azure can run DampApp Pro's AI tools on their own Microsoft Foundry resource. Surveyors' notes, photos and recordings then go only to that resource, under your organisation's agreement with Microsoft, and never to our shared AI, Google or OpenAI.
This guide is for whoever looks after your organisation's Azure subscription. It takes about 20 minutes.
Checked against the providers' own pages on 3 November 2026.
- Time needed
- About 20 minutes in the Azure and Foundry portals.
- You will need
- An Azure subscription you can create resources in, and access to the B2B portal (or the app on each device).
- What it costs
- Microsoft bills your Azure subscription for what the AI tools use. There is no extra charge from us.
- Is Copilot enough?
- No. Microsoft 365 Copilot has no way for another app to use it. Foundry is the Microsoft service that does.
Step by step
Step 1: Create a Foundry resource
In the Foundry portal, create a project with a new Foundry resource, or in the Azure portal create a “Microsoft Foundry” resource. Choose a region in the geography your organisation keeps its data in, such as UK South.
Open the Foundry portal (opens in a new tab)Step 2: Deploy a chat model that can read photos
In your project, open “Models” (or the model catalogue), choose a chat model that accepts images, such as gpt-4.1-mini, and select “Deploy”.
- Deployment type: “Standard” keeps processing in the resource's geography. “Data Zone Standard” keeps it in a data zone such as the EU. “Global Standard” may process requests in any region. Pick what your data policy allows.
- Deployment name: note it exactly as you type it, for example gpt-4-1-mini. The app needs this name, not the model's.
Step 3: Optional: deploy a transcription model for AI Transcribe
AI Transcribe writes down a surveyor's recording. To offer it, deploy gpt-4o-transcribe, gpt-4o-mini-transcribe or whisper in the same resource and note its deployment name. Without one, AI Transcribe stays hidden and the phone's own dictation still works.
Step 4: Copy the endpoint and a key
In the Azure portal, open the resource, then “Resource Management” → “Keys and Endpoint”. Copy the “Endpoint” (it looks like https://your-resource.openai.azure.com) and “KEY 1”.
Keep KEY 2 for changing keys later
To change the key without a gap, put KEY 2 into the portal, then regenerate KEY 1.
Step 5: Set it for every device from the B2B portal
Sign in to the B2B portal and open “App Settings” → “Organisation AI & Weather Keys”. Choose “Microsoft Foundry”, then fill in:
Microsoft Foundry settings in the B2B portal Box What goes in it “Microsoft Foundry Endpoint” The endpoint from step 4. A copy with /openai/... after it works too. “Microsoft Foundry Key” KEY 1 (or KEY 2) from step 4. “Foundry Deployment” The deployment name from step 2. “Foundry Transcription Deployment” Optional: the deployment name from step 3. Devices pick the settings up at their next check, usually within a few minutes of opening the app. Surveyors see Using your organisation's key and can't change them.
No B2B portal? Each surveyor can enter the same four values in the app under SettingsIntegrationsAI ProviderMicrosoft Foundry.
Step 6: Check it works
On a device, open a survey, type a rough note in a room's Observations and tap “AI Polish”. A tidied note means Foundry is answering. If something is missing, the app says what and sends nothing.
Where the data goes
- Only to your Foundry resource. With Microsoft Foundry chosen, the AI tools never use our shared AI, Google or OpenAI, not even when Foundry fails or isn't set up: the tool says what's wrong instead.
- Straight from the device. Requests go from the phone or tablet to your resource. They don't pass through our servers, and we can't see them.
- Microsoft's commitments. Microsoft says prompts and completions sent to models sold by Azure are not available to other customers or to OpenAI, and are not used to train or improve the models. With a Standard deployment they are processed in the resource's geography.
- Abuse monitoring. Microsoft checks requests for misuse. Requests its system flags may be stored in the resource's geography for review by authorised Microsoft staff, unless your organisation is approved for modified abuse monitoring.
Read Microsoft's own page for the full terms: Data, privacy and security for Foundry Models sold by Azure (opens in a new tab).
What each AI tool sends
| AI tool | What is sent |
|---|---|
| AI Polish | The text in that box. |
| AI photo analysis | The photos of that room or area, its name, and the region and age of the property. |
| Talk Through a Room | What the surveyor said about the room, as text, and the room name. |
| AI Transcribe | The recording the surveyor makes (up to three minutes), which is not kept by the app. |
| AI executive summary and translation | The client's name, the address and the survey's findings. |
Each tool can be turned off for every device, or one device, in the B2B portal under “AI Tools”. Smart Build and the phone's own dictation don't use AI and stay.
If something goes wrong
| What the app says | What it usually means | What to do |
|---|---|---|
| Microsoft Foundry isn't fully set up | The endpoint, key or deployment name is missing. | Fill in the boxes named in the message (step 5). |
| That isn't a Foundry address | The endpoint is not an Azure address. | Copy the Endpoint from Keys and Endpoint (step 4). |
| 401 or an access denied message | The key is wrong, or was regenerated. | Copy KEY 1 or KEY 2 again. |
| 404 or DeploymentNotFound | The deployment name does not match. | Copy the deployment name exactly as Foundry shows it, not the model name. |
| 429 or a rate limit message | The deployment's tokens-per-minute quota is used up. | Wait a minute, or raise the deployment's quota in Foundry. |
| The request was blocked by a content filter | Foundry's guardrails stopped the request. | Reword the note. Photos of people can also trigger it. |
Questions people ask
Can we use Microsoft 365 Copilot instead?
No. Copilot works inside Microsoft's own apps and doesn't let other apps send it work. Foundry is Microsoft's service for that, and your Copilot licences don't cover it.
Which model should we deploy?
A small chat model that reads images, such as gpt-4.1-mini, is fast and low-cost and handles survey wording well. Bigger models cost more and are rarely better for this.
What will it cost?
Microsoft bills by the text and images processed. Each AI action is a small request, typically a fraction of a penny on a mini model. Your figures are in Azure Cost Management.
Can we turn the AI tools off instead?
Yes. In the B2B portal, open “AI Tools” and turn “All AI Tools” off for every device, or turn off just the tools you don't want.
Official help from the provider
These are the provider's own pages. They are the final word if anything here looks different on your screen, because providers change their websites from time to time.
- Microsoft Foundry portal (opens in a new tab)
- Microsoft Learn: create model deployments (opens in a new tab)
- Microsoft Learn: data, privacy and security for Foundry Models sold by Azure (opens in a new tab)
- Microsoft Learn: deployment types (Standard, Data Zone, Global) (opens in a new tab)
- Microsoft Learn: abuse monitoring (opens in a new tab)
Related guides
Still stuck?
Tell us which step you reached and what you can see on screen. A screenshot helps a lot. We answer every message ourselves.