When automation projects stall, it's rarely the technology's fault. It's the approach: the wrong process, the wrong expectations or the wrong people in the room. These are the eight patterns we see most often, and what to do instead.
1. Starting with a tool, not a problem
Someone sees a demo, signs up for a licence and announces the firm is "doing AI". Then everyone looks for problems the tool can solve.
Instead: start with where the hours go. List the recurring admin, estimate the time, and only then ask what would fix it. Sometimes the answer is a workflow, sometimes AI, and sometimes just a better process.
2. Automating a broken process
If nobody agrees on the steps, automation makes the disagreement faster. The results are unreliable and the team loses faith.
Instead: map the process with the person who actually does it, agree the rules and the exceptions, then automate. Mapping is half the value of an Operations Audit.
3. Underestimating integration
Demos run on clean sample data. Real firms have data spread across a CRM, a finance system, shared drives, spreadsheets and inboxes. Connecting them is usually most of the work.
Instead: choose first projects where the systems involved have proper APIs, and treat integration as the main job, not an afterthought. It's also where the biggest wins hide: the admin between systems is usually the expensive part.
4. Ignoring confidentiality
Staff pasting client information into consumer chatbots is more common than most partners realise.
Instead: set a clear policy on which tools can be used with which data, use AI through business services that don't train on your inputs, and keep automated processing in accounts you control.
5. Going big first
Firm-wide rollouts look decisive. In practice they bake in assumptions nobody has tested.
Instead: build one workflow for one team, measure it against the estimate, then extend. The first workflow is typically live in 2–4 weeks, quick enough to learn before you commit more.
6. Nobody owns adoption
The build gets signed off, but nobody makes sure people actually use it. Old habits win.
Instead: name an owner before launch, involve the people who do the work in testing, and make the new way easier than the old one.
7. Removing the human checkpoint
Full autonomy is tempting, until the first wrong email goes to a client or the first wrong figure goes into the ledger.
Instead: design the checkpoints in from the start. AI reads, sorts and drafts. Fixed rules handle the routine. A person approves anything client-facing, financial or uncertain. Every project we build works this way.
8. Not measuring
Without a baseline, nobody can say whether it worked, and the project quietly loses support.
Instead: write down the starting point (hours per month, turnaround time, software cost) and measure against it. Where a figure is an estimate, say so. It keeps everyone honest, including us.
The common thread
All eight come from treating automation as a product to buy rather than a capability to build. The firms that get results pick one painful process, fix it properly, measure the result and then move on to the next.
If you recognise your firm in any of these, you're not alone. The free Operations Audit is a straightforward way to find the right first step, and if there's nothing worth automating, we'll tell you.