Why use criteria at all?
Without them, the loudest request wins. A partner who had a bad week with a proposal pushes for proposals. Someone who has just seen a demonstration pushes for whatever it showed. Neither is necessarily wrong, but neither has been compared with the alternatives. A short set of criteria, applied the same way to every candidate, turns a debate into a comparison and makes the eventual choice easy to explain.
This guide gives the method. If you want the typical answer for a professional services firm, what to automate first sets out the shortlist most firms end up with.
What are the six criteria?
- Time cost. Total hours a week across everyone who does the job. Estimate hours per person in a typical week and multiply by the number of people. Calculating unbillable hours explains how to estimate this without false precision.
- Frequency. How often a single instance happens. Weekly beats monthly, and monthly beats quarterly, because frequent jobs pay back sooner and become habit faster.
- Consistency of structure. Does the output follow the same shape each time, with recognisable sections and a known order? It does not need to be written down; an experienced person just needs to be able to describe it.
- Inputs already exist. Is the information the job needs already held somewhere: in time records, case files, notes, spreadsheets or previous documents? The more it has to be gathered from scratch, the less a rebuild can do.
- Seniority of the people doing it. The more senior the person, the more their time is worth and the more valuable the freed hours.
- A clear owner. Is there one person who would be responsible for the new way of working, and do they have time to see it through?
How do you score them?
Score each candidate from 1 to 5 on each criterion, where 5 is the strongest case for automating. Use plain anchors so different people score the same way:
| Criterion | Scores 1 | Scores 5 |
|---|---|---|
| Time cost | A few hours a week firm-wide | The largest block of hours you can find |
| Frequency | Once or twice a year | Every week or more often |
| Structure | Different every time | Same sections, same order, every time |
| Inputs exist | Gathered fresh each time | All held in records the firm already keeps |
| Seniority | Mostly support staff | Mostly partners, directors or senior managers |
| Owner | Nobody obvious | A named person with time |
Then weight them. Double the time cost score, because it is the reason for doing any of this. Leave the other five at single weight. The maximum total is 35. Scores are for comparing candidates within your firm, not a pass mark.
What rules a candidate out regardless of score?
Some problems cannot be outweighed by strong scores elsewhere. Check each high scorer against these before choosing.
- The standard is disputed. If senior people disagree on what a good output looks like, settle that first.
- It depends on a system change. If the job cannot improve until a core system is replaced, it is not a process rebuild.
- The output is mainly judgement. Advice and decisions are the product. Automate the assembly around them, not the judgement itself.
- Review cannot be defined. If you cannot say who checks the output and against what, anything going to a client carries unmanaged risk. A named person should review and sign off client work, as explained in reviewing AI output.
What does a scored comparison look like?
An illustration of the method with invented candidates and scores, not a description of any real firm.
| Criterion | Monthly client reports | Fee proposals | Annual compliance return |
|---|---|---|---|
| Time cost (×2) | 5 (10) | 4 (8) | 2 (4) |
| Frequency | 4 | 3 | 1 |
| Structure | 5 | 3 | 5 |
| Inputs exist | 4 | 3 | 4 |
| Seniority | 3 | 5 | 3 |
| Owner | 4 | 2 | 3 |
| Total | 30 | 24 | 18 |
The reports win clearly. The proposals are painful and fall on senior people, but they vary more from one to the next and nobody has claimed them. That second point is fixable: find an owner and they may overtake the reports on a rescore. The compliance return is highly structured but too infrequent to go first. Nothing hits a disqualifier here, so the reports would be the first rebuild.
How do you run the exercise?
- List every candidate anyone suggests, without debate. Aim for six to ten.
- Estimate time cost for each using the people who do the work.
- Score the other five criteria together, using the anchors, in one sitting.
- Apply the weighting and add up the totals.
- Check the top three against the disqualifiers.
- Pick the top remaining candidate and name its owner before you leave the room.
The process priority scorer is a quicker version of this method for comparing candidates side by side without a spreadsheet. It uses five ratings instead of six, counts time double in the same way, and asks about the cost of a mistake in place of frequency and ownership, so check those two by hand before you commit. Once you have chosen, documenting the process is the natural next step, and the audit gives an independent estimate of where your fee earners' hours go.