The Client Feedback Gap: Why Satisfied Clients Leave and Lukewarm Ones Stay

My friend showed me the slivers of glass he picked off his pizza. The restaurant would no doubt have welcomed a warning about its deadly garnish, but he didn't want to “make a fuss.”

Decades later, my friend is a partner at a global law firm. When he reads client feedback, I wonder if he remembers telling the server his lethal meal was “fine.”

Calling a night of pizza roulette fine is extreme avoidance. But who hasn't told the server a bad meal was fine to avoid discomfort? So, imagine the appeal of hiding behind “fine” for feedback about someone you depend on.

Asking diners about dinner and clients about lawyers is the only way to learn about their experience, but anything involving humans will be messy.

Holes in the Bucket

Proper client listening takes an unbiased interviewer, believable anonymity, and a discussion guide that digs for risks. Even a well-run program has holes.

The limits start with who you can ask. Your program reaches out to the organizations that chose you and, if you run win-loss debriefs, the ones that shortlisted you and chose someone else. Nobody in the sample is the buyer who never put you on the list.

You also rarely reach everyone who decides. The general counsel, the chief financial officer, the risk officer, and the chair of the board committee judge you against criteria that don’t agree. The same claim can have one judge spinning their chair and the others with their backs turned. In our studies of Australian corporate legal buying, the term “innovative” does exactly that: One person in the decision hears “shaking things up,” while another hears “guinea pigs.” You need to understand the views of everyone who can turn a yes into a no.

Clients don’t give all that up willingly. No general counsel would report, “Another firm was better, but you were the choice I could defend if things went wrong.” On the record, we are all bloodless decision machines. Yet in our studies, career risk keeps surfacing underneath what buyers say about capability and fees.

A debrief has the same ceiling. It gives you the verdict, not all that happened in the jury room. Even if your contact gives you the flavor of the discussion, they know only what everyone was saying, not what everyone was thinking. No one says “right decision for the company but too risky for my bonus” to their colleagues either.

Then there are the answers you can't look for. Who wants to plant ideas in clients with a question like, “How bad would we have to get before you'd choose another firm?”

And you get one pass. The survey closes, the interesting question arrives during the analysis, and it waits until next year.

Even Synthetic Lawyers Have Egos

Synthetic research works those limits from the other side. You build hundreds of AI-generated personas standing in for the buyers in your market, in every role. That includes the people you could never get in front of. You can ask the dangerous questions. And when the answers reveal a thread, you can pull it the same afternoon.

Synthetic research has limits, too. Ask a general counsel whether they would accept lower-quality advice to save money. “Never.” No one wants to be seen settling for second best. Put the same trade-off to a synthetic general counsel and you get the same answer. Even synthetic lawyers have egos.

Survey design can work around that, with humans and synthetic personas alike, but you end up with more questions than a human will sit through. Synthetic clients might have egos, but they are endlessly patient.

Role-play is also the answer to the obvious question: Isn't synthetic research just asking ChatGPT what a general counsel thinks? You can ask ChatGPT, and with its usual confidence, it will spin you its answer: what it thinks the average general counsel would find professionally acceptable.

Ask Claude or Harvey the same question and the answer might be 180 degrees in the other direction. This is why synthetic research involves putting hundreds of personas through carefully designed questions using multiple large language models (LLMs). Synthetic research is not asking a single synthetic respondent, animated by an LLM, to give you the whole decision landscape. That would be like expecting one general counsel to speak for all general counsel.

Measuring Your Tolerance for Guessing

Sort your client listening into three piles.

On the first pile goes what clients told you about the relationship: how the work felt, where service slipped, and what they value in working with you. Only client listening can give you this.

The second pile is anything that asks people to rank, rate, or weigh in the abstract. “Rank these in order of importance: technical excellence, responsiveness, commercial judgment, fees.” Those questions invite respondents to role-play the professional they want to be. If fees came in at the bottom of that list and you are still losing on price, you have a recording of this role-play in your own data.

The third is the doggie-bag data: what partners drop on your desk after lunch with a client. They “like us” and value our “pragmatic advice.” Ask what the client meant by pragmatic, and the partner pauses, then says it again in different words.

Aside from the doggie bags of “data” you pushed off your desk and into the trash, what's missing? The buyer who never put you on the list, the objection not mentioned in the debrief (or even in the decision room), and the questions you couldn't ask, whether for propriety or because the survey got too long. Those gaps between what you know and what you could know are the measure of your tolerance for guessing — at what each decision-maker needs to feel safe choosing you or what ammunition your advocate inside the client needs to press your case effectively.

Five Ways to Stop Guessing

Map the decision, not the relationship. For your top 10 clients, list everyone who can turn a yes into a no: the general counsel, the chief financial officer, the risk officer, and the board committee chair. Mark whom you have heard from directly. The blanks show how much of each decision your listening program never hears.

Swap ranking questions for stories. Instead of “Rank these in order of importance,” ask, “Tell me about the last time you moved work away from a firm. What caused it?” People role-play when they rate. They are more honest when they recount.

Ask the whisper question. “If another firm wanted your work, what could it say that would get your attention?” Satisfied clients can answer it, and the answer is your early warning.

Debrief your advocate, not just the verdict. After a pitch, ask your supporter inside the client which objections they had to answer when you weren’t in the room, and what they wished you’d given them to answer them with.

Keep a question log. Every question that arises from the responses to a survey goes on a list. That list is the brief for your next round of research, whether that’s an interview, a synthetic study, or a partner’s next lunch.

In our own research into legal buying, clients reporting no complaints at all could still name a specific thing that a competitor could whisper in their ear to make themselves attractive. Satisfied and safe are different states, and a satisfaction score cannot tell them apart. The way to stop guessing isn't through a bigger survey or longer client lunches. You stop guessing and fill in the picture when you can ask different questions in a new way.

Learn More About Client Services With the Third Edition of the LMA Body of Knowledge The content in this feature correlates with the Client Services  domain in the LMA Body of Knowledge (BoK). Dive deeper and access the latest edition of the BoK online. The Third Edition of the LMA BoK showcases enhanced expertise across every domain, introduces new competencies in Client Services, Communications and Technology Management, features more advanced skills across all domains and broadens coverage of competitive and business intelligence skills. Plus, it emphasizes a stronger commitment to nurturing diversity, equity, and inclusion across the entirety of the BoK. Learn more.