You cannot speak to your users directly because sales owns every account relationship. How do you run discovery?

What the account team is protecting

The instinct is to read this as territorialism, and occasionally it is, but the more common reasons are ones you would share if the relationship were yours. An account manager carries a renewal number and has watched a product person walk into a client and volunteer a date. Some contracts limit who may contact whom. Some accounts have an executive sponsor who dislikes being routed around, so a researcher emailing a team lead directly creates a problem the account manager then has to absorb. And some users are genuinely expensive to interrupt: a claims handler measured on cases per hour costs their employer money during your interview.

Naming which of these applies changes the whole approach. A commercial fear is answered by a script and a promise about what you will not say. A contractual limit is answered by going through the named contact. A cost-of-interruption problem is answered by shortening the ask to fifteen minutes, or by observing during a naturally quiet period. Candidates who open with "I would escalate to get access" are answering a political question nobody has asked yet, and they lose the round on it more often than they expect.

Change the shape of what you ask for

Access requests fail on their shape more than their substance. "Can I talk to some users?" is open-ended, unbounded, and gives the account manager nothing to take to the client. Ask instead for something specific enough to approve and small enough to be worth approving.

Instead of	Ask for
"Introductions to a few users"	"Twenty minutes with two claims handlers at Northbridge, on how they handle a rejected batch"
"Access to the account"	"A seat on your next quarterly review, listening only"
"Feedback on our roadmap"	"Their three worst days last month, and what happened on each"

The second lever is to make the request useful to the person granting it. An account manager will open a door for research that produces something they can show the client: a written summary, a session that closes an outstanding complaint, evidence that the product team listens. So offer the output before you ask for the input, agree in advance what you will and will not say about dates, and send the summary within two days, so that the second request lands against a track record rather than a hope.

Proxies, and what each one distorts

You will not get a clean sample, so you build a picture out of several dirty ones and read them against each other. What earns marks is not listing the sources but qualifying them.

Support tickets and their transcripts evidence real friction with real specificity, and they over-represent problems severe enough to be worth reporting while missing silent abandonment entirely. Implementation and training staff have watched the work performed dozens of times across clients, which makes them unusually good on process and unusually poor on motivation, because they observe what people do without ever needing to know why. Sales calls tell you what the buyer believes the problem is, which in enterprise software is frequently a different problem from the one the user has, and the gap between the two is itself among the most valuable findings available to you. Internal staff who use the product to serve customers are the most accessible group and the most systematically atypical: trained, motivated, and equipped with workarounds nobody else knows about.

The discipline is to treat convergence across independent sources as weak confirmation and divergence as the interesting signal. When implementation staff describe a step that support has never received a ticket about and the buyer has never mentioned, you have found either a solved problem or an invisible one, and finding out which is worth the two hours of user time you have been rationed.

Behaviour where conversation is unavailable

Instrumentation needs nobody's permission, and it answers a class of question that interviews answer badly. If the hypothesis is that people abandon a form because one field is confusing, the log tells you where they stop, how often they return, and whether they complete it later by another route. If the hypothesis is that a feature is unused, the data settles it outright. What the data cannot tell you is why, or what people did instead, which is exactly the part you were going to spend scarce user time on anyway. So sequence deliberately: use behavioural data to narrow the question until the two hours you eventually get are spent on the one thing logs cannot answer.

One caveat is worth voicing unprompted. Instrumentation in enterprise products is usually incomplete, and an absence of events is ambiguous between nobody doing the thing and nobody having instrumented it. Confirm that a path emits events at all before concluding it is dead.

Quiet overreach is the failure you will not notice

The mode that survives all of the above is unqualified confidence. You gather ten conversations with implementation consultants and two with users, write a clean narrative, and the qualification falls out somewhere between the research summary and the roadmap slide. Six months later a feature ships for a behaviour only trained internal staff exhibit.

The fix is procedural and cheap. Record the role of each source next to every conclusion, and state the decision that conclusion is strong enough to bear. Something like: eleven of twelve sources agree the reconciliation step takes several attempts, which is strong enough to justify investigating it and not strong enough to choose between the two designs, because no source could tell us what people are checking when they re-run it. That sentence is what a senior interviewer is listening for. It shows you know what your evidence is worth, and it separates a research practice from a set of opinions with quotations attached.

The corollary is that blocked access belongs on the risk log with a cost attached, not in a corridor conversation as a grievance. If the honest position is that you cannot choose between two designs without four user sessions, and four sessions need an executive to ask for them, that is a decision somebody senior is entitled to make. Left unsaid, it becomes a guess nobody signed up for.

Access constraints do not excuse you from evidence; they change which evidence you can get. Qualify every conclusion by the sample that produced it, and escalate the missing access as a priced decision rather than a complaint.