Lead generation and automation work is unusually hard to evaluate before you buy it. The deliverable is a system rather than an artifact, the results arrive later than the invoice, and the vocabulary is shared by people doing very different quality of work.
Three questions do most of the filtering. None of them requires technical knowledge to ask or to assess. We would be glad to be asked all three, which is rather the point of publishing them.
One: three weeks after this goes live, what will you tell me?
The value is in the shape of the answer.
An answer in the form of a status summary — "we'll review performance and optimise from there" — describes a project. Projects conclude when the invoicing does, and the thing built quietly stops being anyone's responsibility.
An answer in the form of a number describes something with an owner. A sentence like "it ran 84 times last month with zero failures" can only be produced if measurement was built before the system was, and if somebody is still watching afterwards.
That is a sentence a buyer should require, not an outcome anyone should be promising in advance. The form is the point. A partner planning to hand you a number in three weeks has to instrument the work first — and instrumenting it changes how everything else gets built.
Two: what happens when it breaks, and who picks up?
Systems that touch other systems break. An integration is deprecated, a form field is renamed during a website refresh, an API changes its authentication. This is normal, and it is not a sign of poor work.
What distinguishes a well-built system is how the failure surfaces. Either a monitor tells you, or a client does. The second is considerably more expensive, and it is the default when nobody planned for the first.
Ask specifically: what is monitored, who receives the alert, and what is the expected time to resolution. If none of that has been considered, maintenance has been assigned to you without discussion.
Three: what does "working" mean, in writing, before anything is built?
Most disagreements about this category of work are not quality disputes. They are definition disputes that surfaced late.
The buyer's definition is usually commercial: booked calls, qualified conversations, revenue. The builder's definition is often technical: the system runs, without errors, as specified. Both are reasonable. They are not the same. When they diverge after delivery, both parties are correct and both are unhappy.
Writing the definition down before the build costs one email. It removes the most common source of failure in the entire engagement.
The AI addendum
If any part of the work involves AI, there is a fourth thing worth establishing — but it starts with your own firm rather than the vendor's.
Research across 1,300+ legal professionals this year found 43% of firms with no AI policy and no plans to write one, and 40% receiving contradictory client instructions about AI use on their matters. Firms in that position are making the decision anyway, one matter at a time, without a standard.
Setting that standard is your responsibility and depends on professional rules a marketing vendor has no business interpreting on your behalf. What a vendor can reasonably be asked for is transparency: a plain statement of what is AI-assisted, what a human reviewed before it went out, and what written record exists of both. A vendor who cannot describe that has answered the question.
What good looks like
A partner worth hiring has these answers prepared and is glad to be asked, because the questions select for the way they already work. The measurement exists because they build measurement. The monitoring exists because they expect failure. The definition exists because they have been on the wrong side of an undefined one before.
The firms that cannot answer will say it is too early to tell. That is also an answer, and it is worth exactly what it costs to hear.
Sources
- 8am, 2026 Legal Industry Report (1,300+ legal professionals). View source