Computer vision and applied AI in India, explained
By Abha Lohia · Startup Decoded
Computer vision is AI that understands images and video, and applied AI is the wider practice of using AI on a specific business problem. In India these companies sell to factories, banks, retailers and governments, and they usually earn by saving a customer money or time.
What is computer vision and applied AI?
Computer vision teaches software to read pictures and video. It can spot a crack on a part, count people in a store, read a number plate, check a document photo or watch a field from a drone. Applied AI is a broader label for AI built for one clear job in one industry, such as predicting machine failures or scoring credit risk, instead of a general tool.
This area is older than generative AI. Companies were doing image recognition and prediction long before chatbots. Fractal Analytics is a well-known Indian example of an analytics and AI firm that helps large companies use data. Because the customer pays for a result, not for a model, these businesses look more like industrial technology than software labs.
Newer approaches mix vision with language, so a system can look at a picture and answer questions about it. That widens what the tools can do without changing the basic idea.
A simple way to see the difference: a vision tool answers what is in this picture, while an applied AI product answers what should we do about it. The second needs the first, plus business rules, integration with the customer systems, and someone responsible for acting on the result.
Where is it used in India?
Manufacturing is a major use: cameras and models check product quality on a line, catch defects and track safety rules. Banking and insurance use vision to read identity documents and check claims photos. Retail uses it for stock and store traffic. Farming uses images from phones and drones to spot crop disease or estimate yield. Cities and transport use it for traffic and safety.
Robots are a close cousin. ANSCER Robotics builds autonomous robots for factories, and such machines rely on vision to move around and handle goods. In the StopDown data, a recent headline says Inner Sky Labs launched physical AI foundation models, which refers to AI aimed at machines acting in the real world, not just on screens.
India has a large amount of varied visual data and many low-cost cameras, which helps, but images from real sites are messy. Poor light, dust and odd angles are normal, so models must be tested on site.
How do these startups make money?
Most sell a subscription or a per-use fee tied to volume, such as per document checked or per camera monitored. Others charge a project fee to design and install a system, then a yearly fee to maintain it. Some tie pricing to outcomes, for instance a share of savings.
Sales are usually business to business and slow. A factory will run a pilot on one line, measure results for weeks, then expand to more lines or plants. Revenue grows as the customer trusts the system with more of its work.
Margins depend on whether the company sells software only or also installs hardware such as cameras and sensors. Software-only deals scale faster. Mixed deals carry hardware cost and field work but can lock in a customer.
What does it cost and who funds it?
Costs include data labelling, which means people marking images so models can learn; computing for training; and field teams who install and tune systems. These companies usually need less capital than model labs, since they train smaller models for narrow tasks.
Investors who were active across AI and deep tech in the last 12 months in the StopDown data include Peak XV Partners, Accel, Lightspeed, Bessemer Venture Partners and Khosla Ventures. Corporate investors with a stake in the customer industry sometimes join too, because they want early access to the technology.
Revenue quality matters more than hype here. Investors look at how many pilots turn into paid rollouts, how long customers stay and whether a model that works at one site works at the next without heavy rework.
What rules and risks apply?
Images of people are personal data. If a system identifies faces or tracks individuals, the DPDP Act, 2023 applies, and its Rules, notified in November 2025, bring most company duties into force by about May 2027. Facial recognition by public bodies is also a sensitive policy area, so check the rules for your use. Banks and insurers add RBI and IRDAI requirements.
Risks include models that work in a demo but fail on a messy shop floor, bias if training images do not reflect real users, and wrong decisions that harm people, such as a false fraud flag. Customers also worry about who owns the data a system collects.
Competition comes from large IT firms, global vendors and in-house teams. This is general information and not legal advice.
How do you judge a vision or applied AI company?
Ask first about the paid customer base. A company with several plants or branches paying yearly fees tells you more than one with many free pilots. Ask how long a new site takes to set up, because quick set-up is what lets a business scale without a large field team.
Ask about accuracy in plain terms: how often the system misses a problem, and how often it raises a false alarm. Both matter, and the right balance depends on the job. A missed defect may cost a customer more than a false alert, while for a bank a false fraud flag annoys a good customer.
Also look at data rights. Who owns the images and the improved model after a project? Clear terms avoid disputes later. Finally, check how the company is placed against big IT firms that bundle similar services, since price and trust shape many buying decisions.
What is changing next?
Cheaper cameras and smaller models let vision run on the device itself, without sending every image to the cloud, which helps with cost and privacy. Systems that combine vision with language and with robots, sometimes called physical AI, are gaining attention, and Inner Sky Labs launched physical AI foundation models in October 2026.
The breakdown
Business models
| Model | How it makes money | Who uses it |
|---|---|---|
| Per-use fee | Charge per document, image or camera processed | Document and identity checking tools |
| Subscription plus set-up | Install fee followed by a yearly licence | Factory and retail vision systems |
| Project and maintenance | Fee to build, then support contract | Applied AI and analytics firms |
| Outcome-linked fee | Share of savings or gains the system delivers | Process improvement and prediction tools |
The numbers that matter
- Pilot-to-rollout conversion decides growth, since most deals start with one site or line.
- Labelling and field tuning costs can eat margin if every site needs rework.
- Hardware content (cameras, sensors) lowers margin but raises switching costs.
- Customer retention is high once a system is part of daily work.
Rules and regulators
| Regulator or law | What it means |
|---|---|
| DPDP Act, 2023 and Rules, 2025 | Images of people are personal data; most duties due by about May 2027. |
| RBI and IRDAI rules | Vendors of identity and claims checks must meet the regulated customer's standards. |
| Sector safety rules | Factory and transport use may need to meet workplace and safety standards. |
Risks
- Models failing on real-world conditions.
- Bias and wrong decisions affecting people.
- Slow pilots and long payment cycles.
- Competition from large IT firms and in-house teams.
- Privacy concerns about cameras and stored images.
Computer vision & applied AI: latest on StopDown
- Jaipur Robotics raises €4.3 Mn seed round 7 September 2026
- ProactAI builds India's first vision foundational model 27 July 2026
- SwitchOn raises $8M Pre-Series B round 16 July 2026
- Hakimo raises $12M growth funding from Zigg Capital 9 July 2026
- Constems AI raises $2M pre-Series A round 11 March 2026
Every Computer vision & applied AI story →
Most active investors here
- AIF (1 round)
- Cognify (1 round)
- Defy.vc (1 round)
- Finvolve (1 round)
- IIM Lucknow Enterprise Incubation Centre (1 round)
- India Accelerator (1 round)
- Inflection Point Ventures (IPV) (1 round)
- IvyCap Ventures (1 round)
Rounds StopDown covered in the last 12 months. Activity is not a measure of quality.
Questions people ask
What is computer vision?
Computer vision is AI that understands images and video, for tasks such as spotting product defects, reading documents or counting people and vehicles.
What is applied AI?
Applied AI means using AI on one specific business problem in one industry, such as quality checks in a factory or risk scoring in a bank, instead of building a general-purpose tool.
How do computer vision startups earn money?
They usually charge a subscription or a per-use fee, often after a paid pilot. Some also charge for installation and maintenance, and a few link fees to savings achieved.
Does computer vision raise privacy issues in India?
Yes. Images of people are personal data under the DPDP Act, 2023, and most company duties under its Rules are due by about May 2027. Check your use case with a professional.
Which Computer vision & applied AI startups in India raised money recently?
Jaipur Robotics (€4.3 Mn, seed); SwitchOn ($8M (₹78 Cr), Pre-Series B); Hakimo ($12M, growth); Constems AI Systems ($2M, pre-Series A).
Who invests in Computer vision & applied AI startups in India?
Among the most active backers in StopDown's coverage over the last year: AIF, Cognify, Defy.vc, Finvolve, IIM Lucknow Enterprise Incubation Centre.
Which Computer vision & applied AI companies are in the news?
Recent stories on StopDown cover Jaipur Robotics, ProactAI, SwitchOn, Hakimo, Constems AI Systems.
More in AI & Deep Tech
Startup Decoded · Glossary · Sectors explained · Investor directory · FAQs