All insights

State Modernisation

The 90-Day Pilot That Doesn't Become Shelfware

· 10 minute read

Ninety days is enough to prove a pin, an export and a metric. It is not enough to transform the department. Write a DPA, keep production PII out until the checklist is green, and put the exit in the work order.

The pilot VM was still on. Nobody had logged in for eleven months. The completion report said successful POC. The vendor asked for an AMC. Finance asked for the metric. The officer who had signed the POC had been transferred. That is shelfware: a ribbon, a ghost machine, and a contract that wants to live forever.

Ninety days is a good public-sector clock. It fits a quarter, a budget conversation, and a secretary's attention. It is also how vendors lock a department into a production tenant because the data is already there. The difference is four papers: a Data Processing Agreement, a sandbox rule that keeps production PII out until a checklist is green, a metric with a baseline, and an exit clause that works without a fight.

This playbook is for CIOs, societies and departmental secretaries. It is written on 17 August 2026. It is not legal advice. GFR and state manuals still govern how you buy the ninety days. DPDP still governs what you put in the sandbox.

What ninety days can prove — and what it cannot

It can prove that a signed pack retrieves, that officers will use a draft, that an export works without the vendor, that a quality sample can be scored, and that isolation survives a pcap. It cannot prove department-wide transformation, cash ROI that needs a staffing cycle, or safety of a write tool against a twenty-year core. Those claims need a later project. If the work order's success line cannot be observed inside ninety days, you have written a production project and called it a pilot. Rename it, or cut the success line.

Pilot papers. Missing any one is how the VM becomes furniture.
PaperGreenShelfware pattern
DPA / processor termsPurpose, training ban, sub-processors, residency, deletionMoU or standard SaaS terms
Sandbox ruleNo production PII until checklist green; synthetic or redacted firstA restored production dump just for realism
Metric and baselineOne count you already trustUser delight or vendor telemetry only
Exit clauseData return/wipe, 30-day run without vendor, no hostage AMCAuto-renew into production seats
Owner and sunsetNamed post; day-90 decision date in the diaryVendor CSM as the only owner

The PII checklist before production data

Production personal data enters the pilot only when all of these are green: fiduciary named, purpose written, DPA signed, training banned, access list named, logs in Indian jurisdiction, retention for the pilot transcripts written, kill-switch tested, and counsel has initialled the data class (especially category certificates, health, children, grievance narratives). Until then, use synthetic rows, or redacted historical tickets with identifiers hashed, or circulars only. The model needs real messy data is often true for a later evaluation set. It is not a reason to skip the checklist on day four. A sandbox that is a copy of production with a different URL is not a sandbox. It is production with worse passwords.

The day-90 decision

Three allowed outcomes, written on day 1: exit and wipe; remain in a narrow production of the same workflow with a new work order; or redesign (different pack, different wrap) as a new ninety days. Continue informally is not an outcome. It is how AMCs start without a file. The decision meeting includes finance, the records officer, the CIO, and not only the vendor. The metric owner speaks first. The vendor speaks last. If the owner has been transferred, the pilot pauses. Continuity of the named post is a control.

Procurement shape

Buy ninety days as a fixed-fee service with deliverables, not as discounted production seats. Seats want to survive. If GFR or your state rule makes you use a portal, use it. A free POC from a vendor who already has your data is not free. It is an unpriced processor relationship. DPIIT recognition may matter for eligibility. It does not replace the four papers.

  • Hardware the department retains is easier to exit than a tenant.
  • Prompt and pack live in the department's repo from week one.
  • Weekly minutes, even six lines, beat a day-90 novel.
  • Do not schedule go-live on a public deadline.

Two rooms you can walk into

Same ninety days. One file ended. One file evaporated.

Objections you will hear — and what to do with them

These are the lines that stall the file. Answer them in the room, then put the answer in the note.

Ninety days is too short for government process.

Then write a 180-day pilot with the same four papers and two decision gates. Do not write an open-ended POC. Length is not the enemy. Missing papers are.

We cannot get a DPA signed that fast.

Then you cannot get production PII. Stay on circulars. A DPA that will follow is how dumps travel.

The vendor says exit clauses kill their pricing.

Good. You want the bidder who can price an exit. Hostage discounts are not savings.

A successful pilot should automatically convert.

Only if the work order said so and the day-90 papers are green. Automatic conversion is how shelfware gets a budget line.

Day 0 to day 90, week by week

If day 0 does not produce the four papers, do not start the clock. Starting the clock is how vendors claim you already agreed.

  1. Week 0 (before clock): DPA, sandbox rule, metric, exit, named owner, day-90 meeting in the diary.
  2. Weeks 1–2: circulars or synthetic only. Isolation pcap. Export drill. No production PII.
  3. Weeks 3–4: checklist review. If green, a tightly scoped production class may enter. If not, stay synthetic.
  4. Weeks 5–8: weekly sample, weekly six-line minute, no new tools.
  5. Weeks 9–10: freeze. Repeat export with vendor accounts off.
  6. Weeks 11–12: owner writes the three-way decision. Vendor last. Wipe or new work order. No informal remainder.

How this shows up in the file

Subject: 90-day agent pilot — conditions. Duration 90 days from a dated start, decision meeting dated. DPA attached. Production PII is forbidden until the annex checklist is green. Metric named. Exit: wipe/return, 30-day department-only run, no auto-conversion. Owner named. Success is a proved pin/export/sample, not a transformation claim. Informal continuation is not authorised.

What we will and will not claim

Prcept AI will sign the four papers, stay off production PII until you tick the list, and leave when you say wipe. We would rather lose an AMC than keep a ghost VM in your SDC. DPIIT recognition does not entitle us to your dump.

This article is informational field guidance for Indian universities and public institutions, not legal, procurement, audit or engineering advice. Confirm against the live Gazette, GFR, state financial rules, GeM terms, UGC text, GIGW, DPDP commencement, departmental manual and your counsel before you file it.

How to sequence this in a state, not a slide

“The 90-Day Pilot That Doesn't Become Shelfware” is a department problem. A P1 CIO/CTO should name the legacy system, the officer who owns the file, and the citizen charter clock before buying “government AI pilot”.

Ninety days is enough to prove a pin, an export and a metric. It is not enough to transform the department. Write a DPA, keep production PII out until the checklist is green, and put the exit in the work order. Do not invent league tables of states. Read tenders and policies. Election Model Code of Conduct can freeze a rollout. NIC is a partner, not a villain. SDC readiness is GPU, power, ops and egress — not a logo.

  • Audit the legacy store first.
  • Keep mutation and money as officer actions.
  • Map SLAs to the citizen charter.
  • Budget change requests after go-live.

Close this loop before the next CAB

Put “The 90-Day Pilot That Doesn't Become Shelfware” on the next change-advisory or bid-opening agenda as a single line item with an owner. If it cannot earn a line item, it will not earn a control. The owner should be a P1 CIO/CTO, not “the vendor.”

Revisit the item when the model, the GeM term, the region, or the SI changes. “government AI pilot” is not a one-time workshop. It is a watch item. Date the last check. Unsigned watch items are souvenirs.

What the next file must contain

“The 90-Day Pilot That Doesn't Become Shelfware” earns a line in the noting only if a P1 CIO/CTO can attach proof of “government AI pilot.” A heading is not proof. A vendor slide is not proof. A workshop photograph is not proof.

Write three dated sentences: what was decided, who owns it after the next posting order, and when it will be re-checked. If you cannot write the three sentences, you are not ready to buy, to sell, or to go live.

Leave unsourced percentages out of the note. DPDP is not a blanket localisation statute. The November 2025 AI governance text is guidance, not an Act. CERT-In’s 28 April 2022 directions still set specified incident and log clocks. A PAC, when lawful, lives in GFR Rule 166.

  • Name the designation that owns “government AI pilot.”
  • Attach one artefact a stranger can open next year.
  • Record the instrument you are actually using.
  • Revisit when the model, the SI, the notice or the posting changes.

Questions this usually raises

What makes a government AI pilot turn into shelfware?
No metric, no owner, no exit, and production data in an undead tenant. A ribbon without a day-90 decision is how ghost VMs sit for years.
Can we use production PII in a 90-day sandbox?
Only after a checklist is green: DPA, purpose, training ban, access, logs, retention, kill-switch, counsel on the data class. Otherwise use synthetic or redacted data, or circulars only.
Should a successful pilot auto-convert to an AMC?
Only if the work order said so and the day-90 papers are green. Prefer a new work order. Auto-convert is how you skip finance.
Is a free POC safer than a paid pilot?
Usually less safe. Free POCs skip DPAs and treat dumps as demos. Pay for ninety days with papers.
What should we keep after an exit?
The pack, the scoring sheet, the minutes, and any hardware you bought. Not the vendor's tenant and not an unofficial copy of citizen data on a leftover VM.

Sources