Workflow automation or AI: which does your problem need?
By the ChamberOne team 4 minute read
Many firms buy AI to fix a problem that a simple rule would solve, and a few try to write rules for a problem that is really about reading messy text. The two tools are different. Knowing which is which saves money and avoids a certain kind of embarrassment, where a clever system gets a routine step wrong because nobody told it a rule that would have fitted on one line.
Key points
- Use rules for predictable steps and AI for reading messy text.
- Keep deadlines, conflicts and money under rules and records, not model judgement.
- Count the supervision time AI needs, not just the time it saves.
- Trial one task at a time with a named reviewer and a simple error log.
Two different kinds of tool
Workflow automation follows rules you write. When a matter of type X is opened, create these tasks. When an invoice is approved, send it. When a deadline is three days away, remind the fee earner. The behaviour is predictable. You can explain exactly why it did what it did, and it does the same thing every time.
AI works on input that does not fit neat fields: an email chain, a contract, a client's rambling description of what went wrong. It reads text and produces a draft, a summary or a set of extracted fields. The output varies and can be wrong. It needs review.
- Follows rules you write
- Behaves predictably and does the same thing every time
- Takes thought to set up, cheap to run
- Works on text that does not fit neat fields
- Output varies and can be wrong
- Needs review; cheap to start, costly to supervise
A quick test
Ask four questions about the step you want to improve.
- 1
Could you write the rule on a card? If yes, use automation.
- 2
Is the input already in a form a computer can sort, such as a drop-down, a date or an amount? If yes, automation again.
- 3
Does the step require reading unstructured text and picking out meaning? That points to AI.
- 4
What does a mistake cost, and who would notice? The higher the cost, the more you want a rule or a human check, whichever tool you choose.
Examples from a typical firm
Suppose a five-lawyer firm handles employment, property and family work. Here is how its routine tasks might divide.
- Opening a matter and creating the standard task list for a residential purchase: automation. The steps are known.
- Sending a reminder when a limitation date is approaching: automation. It should never depend on a model's judgement.
- Sorting incoming enquiries by practice area from free-text descriptions: AI can suggest, a person confirms.
- Pulling parties, dates and obligations from a lease: AI extraction, with a lawyer checking against the document.
- Chasing unpaid invoices at fixed intervals: automation.
- Summarising a long email thread for a colleague taking over: AI, as a draft.
- Deciding whether to accept an instruction: neither. That is a lawyer's decision, though checks can feed into it.
The costs people forget
Automation is cheap to run but takes thought to set up. Someone must decide the rules, test them and update them when practice changes. A rule nobody owns becomes the thing that quietly sends the wrong letter two years later.
AI is cheap to start and costly to supervise. Each output needs checking, and the time spent checking should be counted. If reviewing the draft takes as long as writing it, you have gained nothing on that task.
Ask both of any proposal: who maintains it, and how will we know when it is wrong?
Combining the two
The best designs put each tool where it is strong. A common pattern is for AI to read and propose, and for a rule to route. An enquiry arrives; AI suggests the practice area and flags key facts; a rule sends it to the right partner's queue; a person decides whether to act. If the AI suggestion is wrong, the cost is a quick reassignment, not a missed deadline.
Another pattern is to keep deadlines, conflicts and money out of AI's hands entirely. Calendar dates, billing calculations and trust-account movements should come from rules and records, with AI used, at most, to explain or summarise them afterwards.
- Let AI read and propose, and a rule route the result
- Keep deadlines, conflicts and money under rules and records
- Have a person decide whether to act
- Leave calendar dates to a model's judgement
- Leave billing calculations to a model
- Leave trust-account movements to a model
How to start without overspending
Pick the three most repeated tasks in the firm. Write down the steps, who does them and how long they take. Mark each step as rule-shaped or text-shaped. Automate the rule-shaped ones first. They pay back quickly and teach you how your firm really works.
Then choose one text-shaped task and trial AI on it for a few weeks, with a named reviewer and a simple log of how often the output needed correction. Decide on the evidence. If the log shows the tool saves time with acceptable accuracy, extend it. If not, stop.
Whatever you choose, check the vendor's terms on where data goes and how long it is kept, and check the guidance from your regulator or bar on supervision and confidentiality.
Where neither helps
Some problems are about people, not process. If fee earners do not record time because they distrust the system, neither rules nor AI will fix it. If clients are unhappy because they never hear from you, a template might help a little, but the real fix is a conversation about who calls whom and when. Tools make a good process faster. They rarely rescue a bad one.
Signs you have picked the wrong tool
- People keep overriding the automation because its rules no longer match how the firm works. The rules need an owner and a review date.
- AI output is corrected in the same way every time. That pattern is a rule waiting to be written, or an instruction waiting to be improved.
- Nobody can say who checks the output before it reaches a client.
- The tool is used for a task because it is available, not because the task needed it.
Review the list every quarter. Tools and firms both change, and a decision that was right last year may not be right now.
Questions to put to a vendor
If a vendor describes a product as 'AI-powered', ask which parts use AI and which are simple rules. Ask what happens when the AI part is unsure, and whether a person sees the output before it takes effect. Ask who maintains the rules and how you change them. A vendor who cannot separate the two is selling a label, not a design.
This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.
Read next
Onboarding a new client matter: an AI-assisted step-by-step checklist
Walk through every step of opening a new matter — from conflict check to task assignment — showing where AI saves time and where only a lawyer can decide.
6 minute readUsing AI matter briefs to pick up an unfamiliar file quickly
How a generated brief speeds up handovers and cover, and how to check it against the file before acting.
5 minute readAI task suggestions and matter phases: standardising how work starts
How matter-type templates and suggested tasks help new files start consistently, and why each suggestion needs review.
5 minute read