Buyer's guide for operators, ISPs and regulators
How to evaluate DPI vendors: criteria for telecom operators
Compare deep packet inspection (DPI) vendors by testing each one on your own traffic against criteria you write down first. The criteria that separate platforms are encrypted traffic classification, accuracy measured per class (including false positives), deployment modes, scale, policy scope, reporting, integration, data governance, local support and a structured proof of concept. This guide sets out each criterion, the question to ask and the evidence to expect, then shows how Interkey Intelligent DPI answers them.
A Saudi product, built by Interkey
For network, procurement and regulatory teams at telecom operators, ISPs and regulators preparing a DPI evaluation or tender. Reviewed November 2026.
- Write the criteria first:agree what is tested, how it is measured and what counts as a pass before any vendor deploys anything.
- Test on your traffic:a classifier tuned on one operator's traffic will not behave identically on another's.
- Count false positives:a false positive means a policy acts on traffic it was never meant to touch.
- Ask every vendor the same questions in the same form,and run every proof of concept the same way.
Before you compare vendors
Start with the job DPI has to do on your network, because it decides which criteria carry the most weight: capacity planning, policy enforcement (fair use, tiered plans, quality-of-service commitments), regulatory reporting or content governance. Then describe your network: fixed, mobile or both, the traffic volumes involved, and where the platform would sit.
Check that DPI is the right tool. On an enterprise or campus network, flow records and firewall logs may already answer the question more cheaply. DPI earns its place on operator-scale networks, where you need to know what the traffic is, not only how much of it there is.
Ten criteria for evaluating DPI vendors
| Criterion | The question to ask | Evidence to expect |
|---|---|---|
| 1. Encrypted and tunnelled traffic | How does the platform classify traffic it cannot decrypt? | A clear method based on observable characteristics, shown on your own traffic sample |
| 2. Accuracy | What are detection and false-positive rates per application class on our traffic? | Per-class results from your proof of concept, not a single headline figure |
| 3. Keeping up with applications | How does classification keep pace with applications that change to avoid detection? | A described update process that enters your network through your change control |
| 4. Deployment modes | Can it run out-of-band first, and what does inline add? | Out-of-band classification and reporting; inline enforcement as a later, separate decision |
| 5. Scale and resilience | Does the architecture match our traffic volumes and availability needs? | An architecture description and a load test scoped to your traffic |
| 6. Policy scope | At what granularity can policies act, and how fast can they change? | Policy per application, subscriber group or cell, with report-only testing |
| 7. Reporting | What do capacity, policy and regulatory teams see, and how close to real time? | Sample reports and dashboards your teams can check against their needs |
| 8. Integration | How does it connect to our policy and charging architecture, 5G core and reporting systems? | A written integration plan that states what each side provides |
| 9. Data governance | What data is produced, where does it stay, and who sets retention and access? | Metadata-only output, in-Kingdom operation, retention and access set by your rules and the contract |
| 10. Local support and commercial basis | Who supports it, from where, in which languages, and how is it priced? | Named support arrangements and a quote tied to deployment, capacity and scope |
Get the RFP checklist as an editable file
The same checklist as this page, ready to paste into your RFP.
Send me the fileLeave your work email below and we send it to you.
Questions to put to every vendor in the same form
- Do you decrypt traffic to classify it? If not, what do you classify from?
- Which accuracy figures do you publish, how were they measured, and have they been independently benchmarked?
- Can you report detection and false-positive rates per application class during our proof of concept?
- What does the platform need to run out-of-band, and what changes when it moves inline?
- What does the platform output, where is it stored, and who controls retention and access?
- Can the platform run entirely in the Kingdom, on our own infrastructure?
- How are classifier and policy updates delivered, and how do they pass through our change control?
- Who supports the platform after go-live, from where and in which languages?
- What does your price depend on?
Run every proof of concept the same way
A proof of concept is only a fair comparison if every vendor goes through the same stages, on the same traffic, against the same written criteria. Interkey recommends 3 stages:
1. Written acceptance criteria.What is tested, on which traffic, how it is measured and what counts as a pass, agreed before deployment.
2. Out-of-band classification against verified traffic.Each platform classifies without touching the forwarding path, and false positives are measured as carefully as detection.
3. Report-only policy, then the inline decision.Policies show what they would have done before they act; inline enforcement follows only after classification has proven itself.
Durations are agreed during scoping.
How Interkey Intelligent DPI answers these criteria
| Criterion | Interkey Intelligent DPI |
|---|---|
| Encrypted and tunnelled traffic | Classified from handshake structure, packet sizing and timing, and endpoint behaviour, without decryption |
| Accuracy | Product material states 99% recognition accuracy, including for VPN and other adversarial encrypted applications. The figure has not been independently benchmarked and is not a specification for any particular network: measure it in the proof of concept |
| Keeping up with applications | An AI traffic model built into the engine core, with automated feature extraction from live traffic, automated model training and closed-loop feedback from Live View |
| Deployment modes | Out-of-band (classification and reporting) or inline (enforcement); most operators start out-of-band |
| Scale and resilience | Designed for carrier-scale traffic across fixed and mobile networks, where volumes run from Tbps up to dozens of Tbps; throughput is sized for each deployment; full-mesh control and user planes; multi-level disaster recovery |
| Policy scope | Per application, subscriber group or cell; real-time deployment and adjustment; service-chain control; blocking where your policy and applicable regulation require it; report-only mode |
| Reporting | Live View with drill-down to application and flow level, trends, policy execution feedback and latency measured in seconds; a reporting layer for capacity and regulatory teams |
| Integration | Policy and charging architecture and 5G core integration scoped with your team and documented in the integration plan |
| Data governance | Metadata output; runs on your infrastructure and can operate entirely in the Kingdom; retention and access set by your rules and the contract |
| Local support and commercial basis | An Interkey product, localised, integrated and supported by Interkey from Riyadh in Arabic and English on the Saudi working week; quote-based pricing, no published price list |
Common evaluation mistakes
- Comparing headline accuracy figures.Each vendor measures differently and on different traffic. Compare per-class results on your own sample.
- Ignoring false positives.A platform that detects more but also misclassifies more will apply policies to the wrong traffic.
- Starting inline.Testing in the forwarding path before classification is proven lets a classification error reach live traffic, and it mixes two decisions.
- Leaving data questions to the end.Where data stays and who controls it should be settled before deployment, not during contract negotiation.
- Scoping integration late.Policy and charging and 5G core integration shape the deployment; bring your enforcement points to the first scoping meeting.
Questions buyers ask
Can you give us a list of the top DPI vendors?
Are vendors' published accuracy figures comparable?
What should a DPI tender specify?
How long does a proof of concept take?
How is Interkey Intelligent DPI priced?
Proof of concept in three phases
Written acceptance criteria.Agree which rules are tested, on which traffic and subscriber groups, how results are measured and what counts as a pass.
Out-of-band classification against verified traffic.Measure false positives as carefully as detection for every class the rules depend on.
Report-only policy, then the inline decision.Each rule shows what it would have done; enforcement follows only after classification has proven itself.
Discuss a DPI proof of concept
Tell us what DPI needs to do on your network, whether you are preparing a tender, and your timeline. Our team in Riyadh replies in Arabic or English.
- Our engineers read your request
- We agree the scope of a technical session
- You see the product on a use case you choose