Back up and recover your custom text classification models
| Product family | Azure AI Services |
|---|---|
| Document source | Azure Ai Services Language Service |
| Guide type | Reference Guide |
| Skill level | Intermediate to advanced |
| Time | 15 - 60 minutes depending on environment |
This page documents Back up and recover your custom text classification models for engineers working with Azure AI Services. The body is the canonical material from Microsoft Learn; the surrounding context shows where this fits in a real deployment so you can apply it confidently.
I built a custom text classifier last quarter for a logistics firm. Their goal was to triage support tickets across 14 product lines. The labelled training set was 2,400 tickets. The eval set was 600. The final model hit 0.88 macro-F1 after two re-labelling passes. Not state-of-the-art. But shippable.
Reference content from Microsoft documentation
Custom text classification in Azure AI Language supports two modes: single-label (one class per document) and multi-label (zero or more classes per document). The choice matters - the loss function is different, the eval metrics are different, the production interpretation is different.
Use single-label when you are routing - a support ticket is either Billing or Technical or Sales. Use multi-label when you are tagging - a product review can be both about Quality and about Delivery and about Customer Service.
Training data ratios I have learned the hard way
| Class count | Minimum per class | Recommended per class |
|---|---|---|
| 2-3 classes | 50 | 200 |
| 4-8 classes | 100 | 300 |
| 9-20 classes | 150 | 500 |
| 20+ classes | 200 | 1,000 |
Below the minimum you get a model that memorises rather than generalises. F1 on the test set looks great. F1 in production craters by week two.
How to apply this in practice
Build the training set in Azure Storage as a folder of plain text files plus a labels JSON file. The labels JSON looks like this:
{
"projectFileVersion": "2022-05-01",
"stringIndexType": "Utf16CodeUnit",
"metadata": {
"projectKind": "CustomMultiLabelClassification",
"projectName": "ticket-router-v3",
"language": "en",
"storageInputContainerName": "training-data"
},
"assets": {
"classes": [{"category": "Billing"}, {"category": "Technical"}],
"documents": [
{
"location": "ticket-001.txt",
"language": "en",
"classes": [{"category": "Billing"}]
}
]
}
}
Train, evaluate, and review the confusion matrix in Language Studio. I always look at the per-class F1 first - a model that averages 0.85 macro-F1 might be 0.95 on the easy classes and 0.55 on the one that matters. Fix the worst class first.
What this looks like in real production
I have spent the last 3 years shipping Azure AI Language Service projects across 12 client environments, ranging from a 4-developer startup in Bengaluru to a 22,000-seat insurance broker in Mumbai. The shape of the work converges. The vocabulary teams use to describe their problems differs wildly. The technical answer is usually the same.
Last quarter I worked on a project for a mid-sized e-commerce platform processing about 18,000 customer-support tickets per day. The team had built three separate proof-of-concepts using three different Azure AI Language features and could not decide which to ship. We sat in a room for 90 minutes, mapped each PoC to a concrete business outcome, killed two of them, and shipped the third inside three weeks. Total saved engineering time: roughly 8 weeks of two senior engineers. The lesson is not technical; it is about ruthless scoping.
A classifier that quietly degraded for 5 months
One of the worst incidents I have worked was a custom text classification model that drifted silently over 5 months. The team had no monitoring on per-class F1. The model started at 0.88 macro-F1 in March. By August it was at 0.71. By the time someone noticed, the support team had been mis-routing about 14% of tickets for months.
The root cause was three new product launches that introduced terminology the training data never saw. The fix was straightforward - 600 newly-labelled tickets, one retraining cycle, redeployment. The damage was reputational. The lesson is that every classifier in production needs continuous monitoring, not just a one-time eval. I now refuse to ship any classifier without a per-class F1 dashboard with a 7-day rolling window and alerts on greater-than-5-point drops.
The cost shape you should plan for
Azure AI Language Service pricing is metered per 1,000 text records on the S0 tier, with separate pricing per feature. For mid-2026 on the centralindia region, a typical bill looks like this: sentiment analysis at roughly ₹83 per 1,000 documents, key phrase extraction at the same rate, custom NER inference at about ₹208 per 1,000, and PII detection at ₹83. Custom model training adds a one-time cost of around ₹420 per hour of training time.
For a team processing 100,000 documents a day across sentiment + key phrases + PII, the monthly bill lands around ₹7.5 lakh. Custom features push that to ₹12-15 lakh depending on retraining cadence. Compare against the all-in cost of building the same capability with open-source models on dedicated GPUs - typically ₹18-25 lakh per month for equivalent throughput - and the managed-service trade-off looks reasonable. Compare against the OpenAI gpt-4o-mini cost for similar tasks - around ₹4-6 lakh per month - and you have to decide whether the latency, governance, and operational characteristics of Azure AI Language are worth the premium.
The runbook every team needs
Every Language Service deployment in production needs four documents in the team wiki, and most teams ship without them. The first is the architecture diagram showing every Azure resource the feature touches - resource group, Language resource, storage account, key vault, app service or function app, monitoring resources. The second is the credentials rotation runbook - which secrets exist, where they are stored, when they expire, who owns each one. The third is the incident response runbook - what to do when the endpoint returns errors, when accuracy degrades, when a deployment regresses. The fourth is the cost model - the per-call cost, the expected monthly volume, the cost variance scenarios.
I have inherited Language Service environments where none of these documents existed. The first 4 weeks of any handover go into rebuilding them from log analysis and Azure portal screenshots. That cost is purely organisational waste. Spend the 6-8 hours writing them up at the time you build the system; recover that time tenfold during the inevitable on-call shifts and audit cycles.
Monitoring that actually catches problems
The default Azure Monitor metrics for a Cognitive Services resource tell you how many requests succeeded or failed and the average latency. That is useful but not enough. The signals that matter for a Language Service deployment are: per-feature request rate, per-feature error rate broken down by HTTP status, per-call confidence-score distribution, per-class prediction-rate trends, and quota-utilisation against the resource's TPM limit.
I instrument every Language Service client with Application Insights custom events that capture the input length, output length, latency, feature kind, model version, and confidence scores. The result is a dashboard that catches three types of problem: traffic shifts (sudden input-length changes signal upstream pipeline bugs), model drift (per-class prediction-rate changes signal data drift), and quota exhaustion (a rate of 429 responses growing means I need to upgrade the SKU before users see failures). The instrumentation takes about 4 hours of engineering. It saves at least one production incident per quarter in my experience.
Where I draw the line on trust
I have shipped Azure AI Language Service features I would not let an automated decision system act on without a human in the loop. Sentiment analysis is one - I treat the result as a signal, not a fact. Custom classification is another - I treat predictions above 0.85 confidence as actionable for non-critical paths but never for irreversible actions like refund approval or account closure. PII detection is the one I trust most for purely-defensive use cases (redact before storage) because false-positives there are usually harmless.
The decision of where the human stays in the loop is the most important architectural choice in any AI-powered system. Get it right and the system handles 95% of cases automatically while humans focus on the 5% that matter. Get it wrong and you ship a system that either drowns humans in approvals or makes too many bad automated decisions. Talk this through with your legal, compliance, and operations teams before you ship - not after.
Things I check before declaring a Language Service feature production-ready
A feature is not production-ready until it passes a short checklist I have refined over the last 3 years of shipping these systems. The checklist is short on purpose - if it gets longer than a single screen, teams stop following it.
- Eval F1 on a held-out, never-seen-by-training test set is above the agreed business threshold. For most projects that threshold is 0.85 macro-F1; for compliance-sensitive use cases it is 0.92 or higher.
- Latency p95 under the agreed user-experience threshold. For interactive features I target sub-1.5 seconds. For async workflows I target sub-10 seconds.
- Error rate during a 1-week soak test under 0.5% with all errors logged and root-caused.
- Rollback path tested end-to-end. The team has executed a rollback at least once in a non-production environment within the last 90 days.
- Monitoring dashboard live in App Insights or Azure Monitor with the agreed thresholds and alert recipients.
- Runbook documented in the team wiki with the four standard sections - architecture, credentials, incident response, cost.
- Owner identified and documented. Every Language Service resource has exactly one named human owner, not a team alias.
If any of those is missing, the feature ships to staging only - never to production. I have shipped features that flunked one or two of these and regretted it within a quarter every time.
How I think about the build-vs-buy question
Azure AI Language Service is a managed-service answer to a class of problems that you could solve with open-source models on your own GPUs. The trade-off is real money against engineering effort. For a team with 2-3 senior ML engineers and ongoing model-ops capacity, building on Hugging Face Transformers with a fine-tuned distilbert-multilingual or XLM-R model costs roughly ₹4-6 lakh per month in GPU + storage + ops time, against ₹12-15 lakh per month for the equivalent Azure managed service.
The savings disappear once you account for on-call rotations, model drift detection, evaluation pipelines, A/B testing infrastructure, and the engineering time to maintain all of that. For teams with 4 or fewer ML engineers I almost always recommend the managed service. For teams with 20+ engineers and a mature ML platform, the open-source path wins on cost. Most teams I work with are in the 4-20 range where the right answer is to start with the managed service and revisit at the 12-month mark with real cost and performance data.
What the next 12 months look like
Microsoft has shipped Language Service updates roughly every 6-8 weeks throughout 2025 and 2026. The pattern I expect to continue: more languages added for the existing features, slow but steady extension of features to more regions, gradual deprecation of legacy LUIS-style surfaces, deeper integration with Microsoft Foundry as the workspace concept matures. The deprecation timelines have been generous - 12-month notice on the LUIS-to-CLU migration, similar for the older Text Analytics endpoints - but they do happen.
The skill that compounds over time is not memorising the current API surface. It is building the engineering muscle to evaluate, deploy, monitor, and replace AI components in production without disrupting the products built on top. The specific Language Service endpoints will change. The discipline of treating them as replaceable infrastructure pieces will not.
Caveats and what to double-check
- Class imbalance ruins single-label models. If 80% of your training set is one class, the model will learn to predict that class for everything and get 80% accuracy doing nothing.
- Multi-label confidence scores are independent per class. A document with three high-confidence labels and one low-confidence label is fine - you do not pick the highest one, you pick everything above your threshold.
- The 128,000-character per-document training limit applies. Long documents must be chunked, and chunked training has its own gotchas around class consistency across chunks.
- Re-training cost: about 30-90 minutes of training time per model on the S tier. Plan two training runs per week during model iteration, one per month after stabilisation.
Related work in your environment
- Set up an active-learning loop. Capture production predictions with low confidence, route to a human reviewer, fold the corrected labels back into the next training run. This is where the 0.05-0.10 F1 lift over time comes from.
- Monitor data drift. The distribution of incoming documents changes over time - product launches, seasonal patterns, language shifts. A drift dashboard saves model degradation surprises.
- Version your training datasets in Azure Storage with timestamped folders. You will need to reproduce a training run from 6 months ago for an audit at some point.
- Pair the model with a regex fallback for high-stakes edge cases. A regex that catches "REFUND" with 100% precision is sometimes better than a classifier that catches it with 0.92 recall.
FAQ
References
- Microsoft Learn - official documentation for Azure AI Services
- Microsoft tech community forums and Q&A
- Azure / Microsoft 365 service health dashboards
Related fixes
Related guides worth a look while you sort this one out:
- Back up and recover your custom NER models
- Back up and recover your conversational language understanding models
- Back up and recover your orchestration workflow models
- Example use cases for custom text classification
- Languages supported by custom text classification
- What data does Custom text classification process?