Nine Thousand Books, One Desktop: Choosing and Surviving a Library Automation System

Decide on support and data ownership, not on price alone. Both options fit the money; the risk is different. The proprietary package is smoother for a small staff but locks the records in the vendor's format and charges again at renewal. Koha is free and exports MARC and SQL, but it needs a Linux server, upgrades and someone local who understands it; the trainer's ₹25,000 buys installation and training, not a support contract.

Recommendation, conditional: take Koha if the trainer will sign a one-year support arrangement (response within 48 hours, two on-site visits) and the school buys a small server machine or dedicates the desktop, plus a UPS. Cost: ₹25,000 installation and training, ₹30,000 server-class mini PC, ₹6,000 UPS, ₹6,000 labels, ₹5,000 for two scanners, ₹15,000-20,000 for part-time data entry help - about ₹90,000, inside the sanction. If no support arrangement can be had, take the proprietary package instead, because unsupported software fails silently in exactly the way that loses a year's circulation data.

Whatever is chosen, the contract must state: the library owns the data; full export in MARC and CSV on demand, including at termination; no charge for export; training for at least two staff, not one; and a written disaster-recovery procedure. Without an export clause, the school is renting its own catalogue.

Do not attempt full catalogue records for 9,000 volumes first. Circulation is the immediate problem, and circulation needs only a short record: accession number, barcode, author, title, class number, year, price, status. Full description, subject headings and added entries come later for high-use titles.

Sequence:

1. Clean the Excel file first. Standardise author names (surname first, initials with full stops), remove duplicates, and check about 50 rows against the accession register to see how trustworthy it is. If it is only partly reliable, treat it as a typing aid, not as data.
2. Reconcile, then convert. Work shelf by shelf against both the accession register and the shelf list, so a book not on the shelf is detected at once rather than after conversion.
3. Capacity arithmetic. Two people at about 20 books an hour each is 40 an hour; 9,000 volumes is about 225 hours, roughly six to eight weeks part-time. So convert the 800-1,000 titles that actually circulate first, then the reference and rare sections, then the rest.
4. Barcode and label as you go, one book touched once, not twice.
5. Run parallel for one month: software record plus the ruled register, compared daily. Only then retire the register.

Power: a UPS on the server and the circulation desk gives 20-30 minutes, enough to finish a transaction and shut down. Beyond that, keep a manual fallback: a printed pad of issue slips by accession number, entered into the system the same day. Staff must be trained in the fallback, because a system that stops at every outage will simply be bypassed.

Data safety:

• Automatic daily backup after closing.
• Three copies in two places: the server, an external hard disk kept in the principal's office, and an encrypted cloud copy when internet is available.
• One restore test each term, with the date and the person's name written in a register. An untested backup is a hope, not a backup.
• A written log of who changed what, and a rule that only the librarian deletes records.

Inspection report: the report is trustworthy only if the figures are reconcilable. Every statistic - volumes, members, issues, fines - must be derivable from the system and must agree with the accession register and the cash book. Run a stock check on a sample of 300 volumes before the inspection and record the discrepancy rate; if it is above about 2 per cent, say so and give the reason rather than presenting a tidy number. An inspector who finds the reported totals reproducing exactly from the records will trust the rest; one who finds a figure that cannot be traced will doubt everything.