AI Readiness for Indian Businesses
By Akkul Dhand · Published 26 October 2026
Choose a useful first AI pilot and put basic controls around it before expanding.
Who this is for: Business owners and functional leaders considering AI for everyday work, without assuming a specialist technical team.
Start with a real workflow and a measurable problem. Buying access to a tool is not an adoption strategy.
1. Diagnose before acting
Answer each question with a concrete example or piece of evidence. Mark unknowns openly. This is a discussion aid, not a validated scoring instrument.
- What specific task is slow, costly or inconsistent today?
- Is the input information accurate and accessible with permission?
- What happens if the output is wrong?
- Who is responsible for checking and accepting the result?
- Do you know where information goes and how the chosen service handles it?
2. Put the method to work
1. Choose a bounded use case
Select a repeatable task with observable inputs and outputs. Favour a first pilot where errors are detectable and reversible. Define what the system may assist with and what people must continue to decide. Do not start with autonomous high-consequence decisions.
2. Check information readiness
Map the information needed, its source, quality, owner and permitted use. Remove unnecessary personal or confidential data from test material. Confirm applicable obligations and provider terms with the appropriate specialists. Do not assume a generic account is suitable for business data.
3. Build a baseline and test set
Measure current time, quality and rework. Create a representative test set covering normal work and difficult exceptions. Define acceptance criteria before testing so that attractive demonstrations do not move the goalposts.
4. Run a controlled pilot
Name the business owner and human reviewer. Restrict access and uses. Record errors, corrections and total handling time, including checking. Test misleading inputs, missing information and unsupported answers. Keep a manual process available.
5. Decide from the evidence
Compare total cost and accepted quality against the baseline. Include subscriptions, integration, training, review and ongoing maintenance. Expand only when controls work and the improvement is repeatable. Re-test after material changes to the service, instructions or source information.
3. A worked example
Illustrative example: a team pilots drafting routine supplier emails using sanitised examples. Staff check recipients, facts and commitments before sending. The pilot measures drafting plus review time and the number of material corrections. A quick first draft is not treated as a saving if checking takes longer than writing the email manually.
4. Working sheet
The first row shows how to fill this in. Use the blank rows for your own business. Add evidence links and review dates to your working copy.
| Use case and allowed scope | Data and permission | Human reviewer | Acceptance / stop rule | Baseline and result |
|---|---|---|---|---|
| Draft routine supplier emails; never auto-send | Sanitised examples; business-approved service | Responsible employee | Stop for data exposure or repeated factual errors | Total handling time and accepted quality |
5. Action plan
- Week 1: select one use case and name its accountable owner.
- Week 2: review information, terms and controls; build the test set.
- Week 3: pilot with trained users and required human review.
- Week 4: compare results and decide to expand, redesign or stop.
6. Check whether it worked
- Accepted outputs / outputs tested, with material errors recorded separately.
- Total time including prompting, review, correction and rework.
- Full pilot cost, incidents and control exceptions.
This is an operating framework, not a certification or legal assessment. AI can produce fluent but unsupported answers. Human review must be meaningful, especially where mistakes affect people, money or obligations.
What you should leave with: A use-case brief, representative test set, pilot register and documented go/no-go decision.