Thred
All posts

Why accounting automation projects fail, and how to make yours stick

Most accounting automation does not fail at go-live. It fails quietly a few months later, when the team stops trusting it. Here is why, and a rollout that holds.

Mark Holleman6 min read

Working on this yourself? Talk to Mark, our co-founder. We’re here to solve it with you.

Book a demo
On this page

Accounting automation projects usually fail after go-live, not before it. The software runs, yet the team keeps a shadow spreadsheet next to it and rechecks every result by hand, because nobody ever decided which results to trust and who owns the exceptions.

If you are a finance lead or founder who has bought a tool that promised to take bookkeeping off your plate, you probably recognise this. Nothing broke. It just never became the way the work gets done.

What does a failed automation project look like?

A failed project rarely looks like a failure. There is no outage and no angry email. Instead you see small signs:

  • The tool matches bank lines, but someone still ticks every match in Excel before the close
  • Supplier invoices are captured automatically, but approvals still happen over WhatsApp
  • The exception list grows every week and nobody knows whose job it is
  • The person who championed the project left, and the configuration stayed as it was on day one
  • At month-end the team exports everything and works from the export

The cost is double. You pay for the software and you still pay for the manual work, plus the time spent checking one against the other.

Which signs show a rollout is heading for the shadow spreadsheet?

In our experience at Thred, the causes are almost never technical. They come down to five patterns, and most stuck projects show at least three of them.

1. The process was never written down

Automation needs rules. Which costs go to which account, when a customer payment counts as matched, which invoices need a second approval. In many small companies those rules live in one person's head. When the software makes a choice that differs from that person's habit, it looks like an error, even when it is a perfectly defensible booking.

2. The first process was the hardest one

Teams often start with the process that hurts most, such as a messy marketplace payout or a multi-entity intercompany flow. Those are exactly the cases with the most edge cases. A project that starts with a 60% success rate never earns trust.

3. The remainder lands in a shared inbox

Every automated process produces a remainder that needs a person. If that remainder lands in a shared inbox or a dashboard nobody opens, it piles up. Within a month the backlog is bigger than the original manual workload and the project gets blamed.

4. The switch was all or nothing

Turning off the old way of working on a fixed date, before the new way has proved itself, makes people defensive. They keep the old way running in secret, which is where the shadow spreadsheet comes from.

5. Bookings arrive without their evidence

If a booking appears without a link to the bank line, the document and the rule that produced it, a careful accountant has no choice but to redo the work to check it. Opaque automation creates more review work, not less.

How do you make accounting automation stick?

The fix is not a better tool on its own. It is a rollout that earns trust step by step, so the team chooses to stop checking because the evidence tells them they can.

Start with one process that has volume and clear rules

Pick something that happens every day, has one right answer and can be checked against a document or a bank line. If a wrong result would take an expert to spot, it is not your first process. Intercompany flows and revenue on complex contracts can wait.

Put the booking rules into plain sentences first

Spend an afternoon with whoever does the work today and write the rules in plain language. Which supplier goes to which cost account, what tolerance you accept on a payment difference, which amounts need a second approver. Rules that are written down can be applied consistently by software and checked by a reviewer.

Run in shadow mode first

For the first few weeks, let the automation propose and let a person decide. Nothing is written to the ledger without approval. This gives the team a direct comparison between what the software would have done and what they would have done, item by item.

Name an owner for the exceptions

One person, not a team, owns the exception queue for each process. They clear it daily and they feed every recurring exception back as a new rule. The exception rate should fall every week. If it does not, the rules are wrong, not the people.

Widen the permissions only when the numbers justify it

Once a category of work has been proposed correctly for a few weeks in a row, let it run without asking. Keep a person in the loop for anything above an amount you choose, anything new and anything unusual. Moving from "ask me first" to autonomous should be a decision per type of work, not a single switch.

Keep your ledger, at least at first

A lot of projects fail because automation arrives bundled with a migration. New chart of accounts, new reports, new screens for the accountant. That is too much change at once. Automation that works on top of the accounting software you already use, such as Exact Online, Xero or Moneybird, lets you change how the work gets done without changing where it lives. That is the idea behind Thred Layer, where agents work in a synced copy of your ledger and push finished entries back.

A worked example of measuring progress

Here is an illustrative example of how a finance lead at a 40-person company might track a bank reconciliation rollout. Assume 400 bank lines a month.

  • Month 1, shadow mode. The software proposes a match for 300 lines. The team accepts 280 and corrects 20. The other 100 lines get no proposal. So 280 of 400 lines (70%) are correct without changes, and 120 lines need a person.
  • Month 2, ask first. Ten new rules came out of the month 1 corrections. Now 352 lines are proposed and accepted unchanged (88%), and 48 need a person.
  • Month 3, autonomous below €2,500. Lines below the threshold that match a known rule are booked without asking. 372 lines are handled without a change (93%), and 28 go to the exception owner.

The number that matters is the one going down, from 120 to 48 to 28 items that need a person. That is the workload you actually took off the team. Track it per process and share it monthly, and the project keeps improving instead of freezing at its go-live state.

What should you ask a vendor before you start?

A few questions separate tools that will stick from tools that will stall:

  • Can I see, for every entry, the source document, the bank line and the rule that produced it?
  • Can I set different permission levels per type of work and per amount?
  • What happens to an item the software cannot handle, and who sees it?
  • Does it learn from corrections, and can I read the rules it learned?
  • Do I have to migrate my ledger, or does it work with the one I have?
  • How do I measure the share of work done without a person, month by month?

If the answers are vague, expect the shadow spreadsheet to come back.

Where does Thred fit?

Thred was built around the rollout described above. Agents propose first, you decide per agent how much freedom it gets, and every entry is traceable from source document to journal line. Exceptions land in one place with a clear owner, and corrections become rules the agents apply next time. If you are a founder or finance lead who wants the recurring work off your desk without a risky migration, start with Thred for businesses.

Book a demo and we will walk through a rollout plan for your first process.

Frequently asked questions

How long does it take to automate bank reconciliation?

For a small company with live bank feeds, a few weeks of running alongside the current process is usually enough to see which matches can be trusted. Most of the work is writing down rules and clearing the first exceptions.

What is shadow mode in accounting automation?

Shadow mode means the software proposes bookings or matches but a person approves each one before anything is written to the ledger. It lets the team compare the software's choices with their own before giving it more freedom.

Should I change accounting software when I automate?

Not necessarily. Changing ledger and process at the same time doubles the risk. Tools that work on top of your existing ledger, such as Exact Online or Xero, let you change how work is done first.

How do you measure the success of finance automation?

Track the number of items that still need a person each month, per process, and the length of the close. A falling exception count is a better measure than a go-live date.

Which accounting process should you automate first?

Choose a daily, high-volume task with one right answer that can be checked against a document or bank line, so the team can verify results quickly. Leave work that needs expert judgement, such as revenue on complex contracts or intercompany flows, for a later stage.

Sources

  1. Keeping business records (business.gov.nl)
  2. Company and accounting records (GOV.UK)
  3. VAT in the Digital Age (European Commission)

Read next

Where Thred comes in

Thred Layer

Finance agents running on top of your existing ledger. You set the rules, approve the work and handle the exceptions. No migration.

Keep reading

Finance, handled.

Book a demo, or request a callback and we will call you.

Book a demo