Skip down to the research record

2 backup sources for a rep's forecast file

Quota Forecast Editorial Team · Published · 12 min read

The two backup sources are cloud object storage, which gives you a single off-site anchor version, and a local backup tool, which builds a chain of restore points on the machine that holds the file.

  • Cloud object copy: one stored version, replaced when you push again; lives off the rep's machine; retrievable from any machine with credentials; can go stale silently if nobody refreshes it.
  • Local backup chain: multiple restore points across days; runs unattended when the machine is on; only grows on days the job actually runs; retention rules trim obsolete points.
  • File size: standard uploads stop at 5 GB; large files run 5 MB to 10 TB in parts, with the storage charge tracking bytes stored, not the number of parts.
  • Recovery order: try the backup chain first because it is local and fast; if the machine itself is gone, pull the most recent dated anchor from the cloud.
  • Setup checklist: confirm the file path, schedule, retention window, anchor cadence with dated file names, credentials held by two people, size headroom, and a quarterly recovery test.

Why this decision lands on you

A rep's forecast file is one of the most quietly load-bearing objects in your sales org. It holds commitments, pipeline shape, and the math behind next quarter's plan. If it disappears, you do not just lose a spreadsheet. You lose the conversation you were going to have on Monday.

You have two realistic homes for a second copy. The first is cloud object storage, where a file is uploaded and kept as a stored object. The second is a backup tool running on the machine that already holds the file, where restore points accumulate on a schedule. Both keep a copy. They fail in different ways. They cost different amounts of attention.

This post weighs those two sources against each other so you can pick one, or run both in different roles, and know why.

How to read the two options

Think of option one as an off-site locker. You push the file up to a cloud provider. The provider stores the object. You can pull it back when you need it. The file sits there until you delete it.

Think of option two as a rewind dial on the machine itself. A backup job runs on a schedule. Each run leaves a restore point. You can step back to an earlier point when something goes wrong. Retention rules decide how many points survive.

The difference matters most when a rep overwrites a tab, a formula breaks, or a whole sheet is deleted by accident. A static cloud copy gives you one known-good version, the one you last uploaded. A backup tool gives you a chain of versions from different days, and the tool itself decides which ones stick around.

There is one more angle worth naming early: how long you are supposed to keep the record at all. The IRS explains that the right retention length depends on what the document records, and that the period of limitations is the window in which a return can be amended for a credit or refund, or in which the agency can assess additional tax, as stated in IRS recordkeeping guidance. Your forecast file is a working document, not a tax record, but the habit of matching retention to purpose transfers cleanly. Keep each version as long as it still answers a question you might ask.

The question to answer first: what would you need to recover, and from which day?

Where a cloud object copy fits

A forecast file you export weekly and push up fits that shape well. You keep the live file on the rep's machine or in a shared drive, and you push a snapshot to the cloud on a rhythm you choose.

Backblaze documents how large files are handled in this model. Files can range from 5 MB to 10 TB, which is well beyond the 5 GB limit on normal files, according to Backblaze's large-file documentation. That ceiling matters if your forecast file carries embedded history, prior-year tabs, and a growing append-only log of commits.

The 5 GB ceiling on standard files exists for a practical reason: it is roughly how much can be reliably uploaded with a single HTTP request in a reasonable amount of time. Large-file upload splits the work. Each part can be anywhere from 5 MB to 5 GB, and you pick the size that suits your tooling. For best upload performance, Backblaze recommends using the recommendedPartSize parameter returned by the b2_authorize_account operation. When you call the b2_finish_large_file operation, the parts combine into one large file, and the byte count stored stays the same, so the storage charge does not change because of the split.

For a sales leader, the takeaway is simple. If your forecast export is a few megabytes, standard upload is fine. If someone starts archiving every revision with full history embedded, you will cross into large-file territory, and the part-splitting handles it. The cost does not spike just because the file arrives in pieces.

Strengths of the cloud copy

  • The file lives off the rep's machine, so a laptop failure or a spilled coffee does not take the only copy with it.
  • You control when a new version replaces the old one. The copy is whatever you last pushed.
  • Storage scales with the object. Storage scales with the object, and the storage charge tracks bytes stored.

Weaknesses of the cloud copy

  • It is one version unless you deliberately keep many.
  • You must know it exists, where it lives, and how to pull it down.
  • If the rep stops pushing, the copy silently goes stale. Stale looks identical to fresh until you open it.
  • Access control is yours to maintain. Shared links and broad permissions create their own risk.

The right role for the cloud copy is the known-good anchor. Push it after each close, after each forecast call, after each committed change. Give it a name that includes a date so nobody has to guess which version they are holding.

Where a local backup chain fits

A backup tool runs on the machine that holds the file and writes restore points on a schedule. Veeam describes this pattern in its retention documentation. The backup job has successfully run 3 times and created 3 restore points in the backup chain, and after a stretch where the machine is off, the chain reflects only the days the job actually ran. If a restore point is obsolete, it is removed from the chain, and the tool ignores restore points created on the day the retention policy is applied, as set out in Veeam's short-term retention policy.

That behavior has a sharp edge worth understanding. The chain is only as deep as the days the machine was on and the job ran. A rep who travels for a week and closes the laptop does not accumulate versions during that week. When they return, the next run starts a new point, and older points age out according to the retention rule.

The upside is genuine recovery depth. If a rep destroys a formula on Tuesday, you can step back to Monday's point. That is a different kind of safety than a single uploaded snapshot.

Strengths of the backup chain

  • Multiple restore points across days, so a mistake is undoable without hunting through folders.
  • Check how restore works in your tool before relying on it.
  • Confirm the job's configured scope includes the folder where the forecast file lives.

Weaknesses of the backup chain

  • It lives on or near the source machine. A fire, theft, or ransomware that reaches the backup store can take both copies.
  • Retention rules trim the chain. Obsolete points are removed, so depth is finite by design.
  • The chain only grows on days the job runs. A long period offline leaves a gap.
  • Tuning is real work. Schedules and retention windows all deserve a look before you trust them.

For your forecast file, the backup chain is the undo button. The cloud copy is the off-site anchor. They solve different problems, and the strongest setup often uses both.

The comparison at a glance

Here is the core of it in one view.

Point of comparisonCloud object copyLocal backup chain
What you getOne stored version, replaced when you push againA chain of restore points across run days
File size handlingStandard files up to 5 GB; large files 5 MB to 10 TB in partsGoverned by job settings and local storage
Who decides what survivesYou, when you overwrite or deleteThe retention rule; obsolete points are removed
Failure mode to plan forStale copy that nobody refreshedGaps on days the job never ran
Best first useOff-site anchor after each closeDay-to-day undo for a broken file

That table is the argument you take to your ops lead. Anchor and undo are separate jobs. Collapsing them into one tool usually means doing both badly.

Picking the roles for your team

A split usually works well here. The backup chain handles the day-to-day. The cloud copy handles the quarterly and the catastrophic.

Set the daily layer. Confirm the backup job on each rep's machine covers the folder where the forecast file lives. Confirm the schedule runs on days people actually work. Confirm the retention window covers at least the span of a forecast cycle, so a bad edit last Thursday is still reachable this Thursday.

Set the anchor layer. Decide who pushes the cloud copy and when. Common moments are the night before the forecast call, the day after commitments are logged, and the day the quarter closes. Name each upload with a date. Do not let a file called simply "forecast.xlsx" be your only clue about which week it holds.

Set the access layer. Decide who can read the cloud copy and who can overwrite it. A rep should be able to push. A rep should not be able to silently replace an anchor without anyone noticing. Two sets of credentials, or a write-once folder, keeps the anchor honest.

Test the recovery. Both layers are theories until you pull a file back. Do that on a quiet afternoon, not during a fire drill. Restore last week's anchor from the cloud. Restore yesterday's point from the chain. Time it. If either takes longer than your patience, you have found a problem while it is still cheap.

When you work the example back into the week: a rep commits numbers Tuesday night, pushes the cloud anchor Wednesday morning, and breaks a roll-up formula Thursday. The chain gives you Thursday morning's file back in one action. The anchor gives you Wednesday's committed version if the whole drive is lost. Neither alone covers both moments.

What to check before you trust either one

Use this list on each rep's setup. It takes an afternoon and prevents the bad surprise.

  • Confirm the file path. The forecast file lives where the backup job looks, and not only in a sync folder that behaves differently.
  • Confirm the schedule. The job runs on workdays, not only overnight when the laptop is asleep.
  • Confirm the retention. The window is long enough to reach back across a forecast cycle, with obsolete points trimming from the far end.
  • Confirm the anchor cadence. Someone pushes the cloud copy at the moments you named, and the file name carries a date.
  • Confirm credentials. Two people can restore, so a single departure does not strand the archive.
  • Confirm size headroom. If the file is growing, check whether you are still in standard-upload territory or have crossed into parts.
  • Confirm recovery once a quarter. Pull one file from each layer and open it. Verify the numbers are the ones you expect.

A note on scope, in plain words: your sources here cover file size limits, retention behavior, and general recordkeeping purpose, and they do not spell out a rule for how long a private forecast file must be kept. That call belongs to your own policy owner or counsel, who can align it with your contracts and any discovery obligations you carry.

FAQ

What if a rep only ever uses one of the two? They get one failure mode covered and one left open. A chain without an off-site anchor survives a bad formula but not a stolen laptop. An anchor without a chain survives the laptop but not Thursday's mistake. One layer is better than none. Two layers are what you actually want.

Does a larger forecast file change the choice? Not really. It changes the upload path. Standard files stop at 5 GB because that is roughly what one HTTP request handles reliably in a reasonable amount of time. Above that, the file goes up in parts, each between 5 MB and 5 GB, and the storage charge tracks the bytes stored, not the number of parts. The role of the copy stays the same.

How deep should the restore chain run? Deep enough to reach back to the last point where you would still trust the numbers. The chain only grows on days the job actually ran, and obsolete points are removed when retention applies. So the real depth is the run days inside your window, minus the points the rule has trimmed.

Who should own this on the team? One named owner for the anchor cadence, and one for the backup job settings. Shared ownership sounds efficient and usually means neither gets checked. The owner's job is not to run the tools. It is to notice when the last anchor is older than your last forecast call.

Can the cloud copy go stale without anyone noticing? Yes, and this is a quiet failure to watch for. The upload works, the file sits there, and three months later you pull it during an outage and find last quarter's numbers. Date the file name, and check the anchor age on the same day you check the chain depth.

What should the first hour look like after a file is lost? Try the backup chain first, because it is local and fast. Step back to the last point from before the damage. If the machine itself is gone, go to the anchor, pull the most recent dated upload, and treat it as the new floor. Then rebuild forward from whatever the rep can recall, and push a fresh anchor once the file is trustworthy again.

The choice is not which tool is better. It is which failure you refuse to live with. Build for both, and keep the names on the files honest.

Sources