Migration checklist

Forty-four questions to ask before you move.

Software moves rarely fail at go-live, but on a question nobody asked in week one. Ask these of yourself first, then of every vendor, including us.

Request an invitation

Last reviewed 11 October 2026. The answers in the last section describe how Ordence works in practice, not contractual obligations; get the ones that matter to you in writing.

Vendor-neutral: Ordence answers several with a no. Print it or save it as a PDF for the meeting.

How to use it

Which groups to answer yourself and which to take into vendor calls.

In short: answer groups 1 and 2 yourself; ask 3 to 6 in vendor calls.

Do not send it as a questionnaire: half is about your business, and the rest is worth more asked live. Do groups 1 and 2 with the people who do the work.

Expect to fail some. Most businesses have duplicate customers, three names for one product and an undocumented spreadsheet. That is not a reason to stop; it is the scope, found before you pay, not in week nine.

1. Your data, before anyone touches it

In short: problems show in the data months early. Nine checks for this afternoon.

Where is the master customer list, and who may change it? Two places means two lists.
How many duplicate customers do you have? Sort by phone, not name. Most businesses this size find 5 to 15%.
Are phone numbers in one format? Country codes, spaces, landlines in the mobile column. Finding duplicates depends on it.
Do you hold a GSTIN for every business customer, and does each pass the 15-character checksum? An invalid one blocks your buyer's credit (the GST they claim back). A checksum only catches typos; a cancelled or wrong-business registration still passes. Ask vendors if they check the live registration.
How many item codes describe one physical thing? Gold foil 640mm, GOLD 640 MM and FOIL-AU-640: three rows, one product.
Does everyone use the same unit per item? Rolls or metres, kilograms or square metres, the conversion in one head.
Can someone tie opening balances to a trial balance (the summary of all account balances) and defend them? If nobody knows total receivables, that is the finding.
How much history do you need: all, three years, or open balances plus last year? Eleven years takes weeks, and nobody reads it.
What will not move? Emails, WhatsApp chats, shared-drive photos, a costing sheet of formulas. Get the list in writing.

2. Your process, as it really runs

In short: eight questions on how work really happens, not how it is described.

What does one person know that is written nowhere? What happens during their two weeks at a wedding?
Which approvals are real and which are for show? One never refused is a delay, not a control.
What is broken and what is just annoying? Broken loses money or customers; mixing them up doubles the scope.
Which spreadsheet does everything depend on? There is always one. Who keeps it, and what stops when they are away?
What is still on paper, and why? Habit, or the only thing that works at the gate at seven in the morning.
Where is the same number typed twice? Each is a cost and a future mismatch.
How many person-days does month-end take? Measure now, or you can never prove it worked.
Which rules come from the law, and which from your current software? Few people know.

3. The vendor, asked out loud

In short: ask these live. Listen for specifics, and for any no at all.

What happens to your data if they close? Ask how. Source-code escrow (their code held by a third party) is no answer; few could run it.
Can you export everything yourself, in open formats, without asking? Even while you owe them an overdue invoice?
Who answers at six in the evening on month-end day? A name and a channel, not a support tier.
How many staff, and how many can fix your problem? A big helpdesk is not the team that built it.
What do they refuse to do? One who agrees to everything has not understood you, or will bill for it later.
Which certifications do they hold, and does your largest customer need any? SOC 2 and ISO 27001 come with a report and date, or not at all.
What connects to other software today, not on a roadmap? Is there a public API? A clear never beats a vague soon.
Will they introduce you to a business like yours? A famous logo says nothing about a bad Tuesday.

4. The money, all of it

In short: the quote is rarely the real cost. Seven questions to find the rest before you sign.

What is in the price and what is not? Migration, setup, training, visits, integrations, custom reports, year two.
Is it charged by user, module, company or scope? Add three people and a branch, and ask again.
What raises the price after signing? Get the triggers in writing: entities, GST registrations, locations, data volume, custom fields.
What is the yearly increase, and is it capped? Get the figure in the contract.
What does support cover, and when do changes become billable? A limit is fair; discovering it in month four is not.
What does leaving cost? Notice, export fees, and paying someone to map data to the next system.
What do you still pay others for? Your CA, Tally licence, e-invoicing, online payments, hosting. Add these to both quotes.

5. The rollout

In short: six decisions to make now, not when they become urgent.

Which team goes first? The one with the worst pain and a willing manager, not the easiest-looking one.
Which module first? In the order money hurts, not what looks neat on a diagram.
Will you run both systems side by side, and until when? It feels safe but kills rollouts. Set a switch-off date. If you are leaving Tally, the parallel run has a defensible length and a switch-over sequence to match.
What is the go-live test? Matching the trial balance is one. Everyone has been trained is not.
What happens to the old system next morning: read-only, archived, or still edited by two people who missed the memo?
When? Not in peak season, not across March, not the week the auditor is in.

6. The first ninety days

In short: adoption decides the outcome, and it turns on six things that are not software.

What does failure look like? The old WhatsApp group returns, or someone rebuilds the old spreadsheet from new exports.
Who owns adoption? Not the vendor. Name someone with authority to stop the old way when it gets busy.
Do managers use it? One who asks for numbers on WhatsApp tells the team it is optional.
What is still in spreadsheets in week six? The most useful question, if asked without blame.
Which three numbers prove it worked? Measure them before go-live: month-end days, enquiries with no owner, receivables past ninety days.
When do you change course, and who may say so? A project nobody may halt costs a quarter.

And if it fails

Ordence's own answers to the seven things a failed migration turns on. Ask every vendor the same seven.

In short: agree how you would undo the move before you start, and compare each vendor's specifics. It sounds like planning to lose, so few ask. Not part of the forty-four.

The dry run comes before you are committed. Masters, open orders and stock are loaded and reconciled before go-live. Tally data comes in as XML or CSV, and Ordence lists where it disagrees with the Tally import. A dry run that does not reconcile is a finding, not something to push through. Full history usually stays in the old system, readable.
Acceptance is three numbers agreed before the load. Closing stock value, total receivables, and the trial balance your accountant signs, each with an agreed tolerance. Most differences are explainable (a rounding rule, an unposted credit note); arguing is cheap before, costly after.
Opening balances are reconciled against something you already defend. Not the new system's own total, which always agrees. A balance that will not tie to the trial balance becomes a named exception with an owner, not hidden in suspense.
Decide where the CA's books live. Ordence keeps its own double-entry books. Many keep Tally for their CA, fed by Tally-ready XML or the optional Tally connector. Then a failed migration costs time and a reload, not your ledger or filings.
Rolling back means going back to the old system, so do not switch it off. Keep it running and licensed until you accept, then read-only for a while. Rollback costs more each day, as new entries must be re-typed into the old system, so keep the parallel period short and dated.
Run in parallel deliberately, briefly, with a date on the switch-off. One cycle, one month-end, one dispatch week. Left open, people slip back in the first busy week and you get double entry. Set the date, and who may move it, first.
Agree how your data leaves before you sign. Any register exports any time as CSV, Excel or JSON, books as Tally-ready XML. If you leave, or Ordence stops, you get a full export of your records. Ask every vendor for its exit method in writing; if you cannot run it yourself, that is the finding.

Ask us all forty-four.

Put every one to us on the fit call. Several answers are no: nothing filed directly with the GST portal or the IRP, no SOC 2 or ISO 27001, no ledger ever created in Tally by Ordence, and three of the seven industry packs with no customer yet. Better heard now.