;

How to Use AI Productively, Profitably, and Safely

Picking AI Tools Without Sacrificing Security or Accuracy

AI can deliver real value when deployed wisely. Evan Zimmerman, CEO of legal tech company Edge, explains how the right AI product streamlines legal workflows, reduces cost and error, and boosts profitability—without compromising accuracy or security.

Start by Breaking the Work Down

When looking at a workflow where you might want to use AI, start by breaking it up into its smallest component parts. Then, take note of how and by whom this work is done.

If you’re feeling particularly ambitious, then you should try to assess the time spent or cost of each component. The results of this last part might surprise you!

This kind of granular analysis may feel unnatural at first. Lawyers are trained to think in terms of outcomes—opinions, filings, advice—not in terms of discrete steps.

But this shift in perspective is critical if you want to understand where AI can add value, where it introduces risk, and where it should not be involved at all.

A Simple Illustration: Making Guacamole

To see what this kind of decomposition looks like in practice, it can be helpful to step outside the legal context entirely. If you were making guacamole, the first four steps would look something like the following:

1. Create a shopping list (done by a chef, or looking at a cookbook);

2. Go to the store (done by you or your spouse);

3. Halve the avocado and scoop out the flesh (done by you, using a knife and spoon); and

4. Discard the pit and outer skin (done by you, by hand).

And so on.

The point of this exercise is not food preparation—it is clarity. Each step is distinct, is performed by a different “actor,” uses different tools, and carries different consequences if done poorly. Some steps require judgment; others are purely mechanical. Some steps create value; others simply enable later work.

Why This Exercise Is Worth the Effort

This kind of mapping is tedious work, but it is very worthwhile. Not only will you have far better visibility into your process than you’ve had before, but you will also have a spec against which you can compare new tools, documentation against which you can build new processes, and the pieces to build the financial case for making go/no-go decisions.

Just as importantly, this will give you the raw material you need to ask good questions of AI tools.

Applying the Same Thinking to Trademark Clearance

When doing a trademark clearance search, it is very important, for example, for every mark that is identified by the AI system to actually exist. We wouldn’t want the equivalent of hallucinated case law in a brief. So how do you guarantee that?

You start the same way—by breaking the work down.

A Step‑by‑Step Trademark Clearance Workflow

Let’s do a step-by-step analysis:

1. Create a search strategy (done by a paralegal, reviewed by an attorney);

2. Execute the search (done by a database);

3. Evaluate the results (done by the searcher); and

4. Write the opinion (done by the attorney).

In this process, a strong AI system will do steps 1 and 3. But there is no reason that AI needs to replace the database in step 2. Furthermore, all that the AI actually needs to do is produce a ranking; there is no reason why AI needs to produce the mark itself, which would introduce a hallucination.

Any good AI system should work this way—and any responsible evaluation should include questions of this type.

Turning Process Insight into Business Insight

Once you’ve broken things down, you can start to truly evaluate where you expect to get value from the process.

Most lawyers have no idea what the cost of the searcher or the search database is. When you write it down, you might be surprised. They also have often not rigorously tracked the total time spent.

Breaking down workflows into steps also allows you to see where problems actually lie.

Search strategy construction is often inconsistent. Evaluation is frequently rushed or artificially constrained. This is where well-designed AI tools can make a meaningful difference.

A Step‑by‑Step Trademark Clearance Workflow

So, if you are looking at an AI trademark clearance search system, you might conclude something like this:

· We’ve broken down our search process into 12 steps.

· Ten of those steps involve a human being.

· Steps 2, 5, and 7 are highly inconsistent and slow, so fixing these would be a good value proposition.

· Step 10 accounts for 30 percent of the total cost, so nearly eliminating this will double our profit.

And so on.

You can then construct a set of test cases to quickly evaluate an AI tool. It will also make the value much clearer: A US $200 AI search may be worthwhile if it saves you US $600 of cost. A 5 percent error rate might be acceptable if your human error rate is 10 percent.

Ultimately, no system—human or machine—is perfect. What matters is measuring and calculating the trade-offs so that you go in with eyes wide open. That way you won’t avoid a mistake—or miss an opportunity.

What to Ask Vendors—Beyond the Demo

It is also important to talk to a vendor to understand what kind of work is required to get value out of the system. Some products work right out of the box. Others require effort on your end. If the latter, the best AI companies will have someone on staff to help you. The very best don’t charge extra for that help.

Thinking back to our starting exercise, understanding the cost and value of the task will help you better understand which jobs and products are even worth investing in.

Security and Privacy Cannot Be an Afterthought

Lastly, it is essential to understand the cybersecurity and privacy posture of any vendor you work with.

Start with certifications and audits. Reputable vendors are evaluated against SOC 2 Type II or ISO 27001 security assurance standards. They should also have zero-day retention (or ZDRs) agreements, with their model providers. Your IT team might also have additional requirements, like up-to-date penetration testing or single sign-on (SSO).

Next, ask how they will customize their product based on your data, if at all. If customization is at the model level, involving training, then your data will leak to other customers. If customization is at the “harness” level, meaning customizing the system without changing the model itself, your data is safe as long as it is sandboxed to your account.

About Edge

At Edge, we have built the world’s first AI trademark product, Certus. Our products are used all over the world by leading enterprises in secure, private environments. Learn how our systems can bring you value in a safe, profitable way by going to withedge.com or stopping by Booth 2012.

Evan Zimmerman is CEO and co-founder of the legal technology company Edge and can be contacted at ejz@withedge.com.