The most common mistake firms make with automation isn't picking the wrong tool. It's picking the wrong process. Every demo makes its use case look urgent, so the useful question isn't "what could we automate?" but "what's worth automating first?"
This is the framework we use in every Operations Audit. It takes an afternoon, needs no specialist knowledge, and usually produces a clear first choice.
1. Start with where the hours go, not with tools
List the recurring admin in your firm and, for each item, write down roughly how often it happens, how long it takes and who does it. Rough numbers are fine. You're looking for pools of time, not precision.
The best candidates tend to sit between systems: copying deal details from the CRM into a contract, chasing invoices that should have arrived, re-keying approved figures into Xero, reading every inbound email to work out who should deal with it.
2. Score each candidate on four things
- Volume. How often does it happen? Something done fifty times a week beats something done twice a quarter.
- Time per instance. Saving thirty minutes on a task done fifty times a week gives back 25 hours. Saving two hours on a task done twice a month gives back four.
- Predictability. Does it follow the same steps each time? Rule-based work is quicker and cheaper to automate. Variable inputs, like free-text emails, are where AI earns its place.
- Cost of a mistake. Internal drafts and research summaries are safe places to start. Anything that moves money or goes to a client needs a human checkpoint, which is fine, but should be designed in from day one.
3. Do the back-of-envelope maths
You don't need a financial model. You need four lines:
- Hours saved per month = time per task × times per month × share that can be automated
- Annual value = hours saved × 12 × loaded hourly cost
- Plus any software you can retire
- Payback = build cost ÷ monthly value
Our admin cost calculator does the first three for you. For a first project, look for payback inside a year. In practice the good candidates pay back much faster. One consultancy we worked with cut its monthly invoicing run from 8 days to about 4 hours and switched off £18,000 a year of legacy software in the same project (read the case study).
4. Check it's actually feasible
A great score on paper means little if the work can't be built. Before you commit, check:
- Is the data available, and in a system with an API, or locked in someone's head?
- Is the process stable, or still changing every month?
- Will the people who do the work today use the new version?
- Are there regulatory or client constraints on automating this step?
Feasibility usually eliminates more candidates than the maths does. That's healthy.
5. Pick one and prove it
Plot what's left on two axes: value and feasibility. High-value, high-feasibility items are your first project. High-value, low-feasibility items go on the roadmap, often after a simpler project has connected the systems they depend on.
Then build one workflow, measure it against the estimate and only then move on. The first workflow is typically live in 2–4 weeks. That's quick enough to see whether the numbers hold before you spend more.
Patterns that score well in professional services
Across consultancies, investment firms and financial services operations, these come up again and again:
- Invoicing and month-end. High volume, rule-based, easy to measure.
- Contracts and statements of work. Generated from CRM data instead of drafted by hand.
- Inbound request triage. AI reads, classifies and routes emails so skilled people stop sorting the inbox. For one fund data provider this cut request turnaround from 6 weeks to 4 days.
- Recurring client and management reporting. Pulled from source systems automatically.
- Research and preparation. Gathered and summarised by AI, reviewed by a person.
What to avoid
- Tool-first decisions. "We bought Copilot, now what?" is the wrong way round.
- Automating everything at once. One process, done properly, beats ten half-finished.
- Ignoring adoption. If the people doing the work weren't involved, they won't trust the result.
- Over-engineering. Not everything needs an AI agent. Often a well-built workflow is the answer.
If you'd like help running this exercise on your own processes, that's exactly what the free Operations Audit is for. And if nothing comes out worth automating, we'll tell you.