1. LandOffer
  2. Blog
  3. All articles
  4. Interview Internal Users Before Building the Workflow They Asked For

Onboarding & Teamwork

Interview Internal Users Before Building the Workflow They Asked For

A fictional translation-reporting discovery study shows how to interview task users, preserve exceptions and workarounds, and map reported evidence to proposed requirements and unanswered validation questions.

10 min readLandOffer team

Before building the workflow an internal team requests, interview people who perform the underlying task. Ask them to reconstruct a recent example: what started the work, what they needed to decide, which exceptions changed the process, and how they finished. Record the requested solution as a hypothesis and investigate the task it is meant to support.

This LandOffer.ai guide concerns product-user discovery with coworkers, not interviews with job candidates. Its reporting-tool guide, simulated notes, and evidence-to-requirement map are original fictional assets. Sources were checked October 25, 2026. No internal users were contacted, no employer records were collected, and no workflow, prototype, or software account was tested or changed for this article.

An interviewer listens to a coworker describing a task with paper examples, while a proposed workflow sketch remains aside rather than leading the conversation.

Translate the requested feature into a research question

“Automatically email our translation backlog every Monday” sounds like a buildable requirement. It still leaves basic questions unanswered. Who prepares the backlog? Who acts on it? What does “unreviewed” mean? Does the email need to trigger a decision, or merely reassure a manager that a report exists?

Keep the original request in the research brief. Then state what you need to learn independently of that solution. In the fictional case below, the question is: “How do documentation staff identify which help-center translations need a language review, distinguish them from items blocked on another approval, and hand that work to someone who can act?” Automatic email is one possible response, not the question itself.

Yale's user-interview guidance distinguishes firsthand users from stakeholder perspectives and encourages including relevant user groups. Apply that distinction by role in the task, not by job title. A manager who also prepares the report may be a user for that activity. A sponsor who only requests the feature is not a substitute for the people assembling and consuming it.

Discuss constraints with the sponsor separately: information access, release authority, existing tools, and the decision the research should inform. Do not promise a delivery date or a particular feature to secure interview participation. The sponsor's preference is evidence about a stakeholder goal; it is not evidence that everyone experiences the task the same way.

Recruit for task differences and prepare neutral prompts

For an internal workflow, include people who create the work, receive it, and cover exceptions. Frequent users may have efficient shortcuts that an occasional backup does not know. A person who stopped using the current report may reveal a different problem from someone who has learned to tolerate it. Record the perspectives you did not reach so the decision does not silently expand beyond the people you heard from.

Keep participation and data handling explicit. Explain the research purpose, whether notes or recordings will be taken, who will receive the findings, and what participants can decline to share. Follow the employer's approved research and data practices. Do not ask someone to open a confidential record merely to make an interview feel concrete. Use permitted examples, sanitized material, or a verbal reconstruction when appropriate.

In the fictional study, researcher Jo speaks separately with Rae, a regular report preparer; Luca, a language reviewer; and Nila, an occasional backup. Sponsor Vik is interviewed separately about the business decision. Three selected users are a deliberately limited first round, not proof of representative coverage or a universal sufficient sample. Another language-review team and an accessibility-support perspective remain outside the round.

The GOV.UK in-depth interview guide recommends a discussion guide, open and neutral questions, real examples, follow-up questions, and informed consent. Jo uses those method principles without adopting a government service's process as an employer mandate. The first conversation also reveals a confusing prompt, which she repairs while retaining the same research question.

Use a full guide and record what the participant actually supplies

Worked asset 1 — completed fictional interview guide and simulated notes. Jo's introduction says: “I am learning how this reporting task works today. I am not evaluating your performance or asking you to endorse an email tool. We will take limited notes, not record the session, and share a task summary with the project team. Please avoid private content and tell me if a question or example is unsuitable.” In the story, each participant accepts that arrangement; no real consent was obtained or implied.

Topic and neutral prompt Follow-up Jo uses Completed simulated note
Recent task: “Tell me about the last time you prepared or used this report.” “What started it, and what did you need to decide?” Rae describes Friday's preparation for the next review cycle. She says her goal was to identify items ready for language review, not every item lacking a final approval. This is her report of the task.
Sequence: “What did you do next?” “Which information told you an item belonged in that list?” Rae says she compared queue labels with comments, then kept a small scratch list of titles. Jo has not watched this happen on a live system.
Exception: “Tell me about an item that did not fit the usual path.” “What was different, and who could resolve it?” Luca describes a translation awaiting product approval rather than language work. He says a generic pending label does not tell him which action is possible.
Workaround: “How did you finish when that information was missing?” “What did the workaround preserve or lose?” Nila reports asking Rae for a screenshot while covering the task. She could read it but could not establish whether its status was still current. No screenshot from an employer is collected.
Responsibility: “Who did you go to for clarification?” “What happened if that person was unavailable?” Rae names the document owner for product-approval questions; Nila says she waited for Rae. These accounts suggest different dependencies, not an agreed ownership policy.
Readback: “I heard that the useful list separates actionable review from blocked items. What have I missed?” “Does that description fit your example, or only part of it?” Luca corrects Jo: blocked items still need visibility, but should not look like work he can complete. Jo records that correction rather than removing blocked items from the proposed scope.

Jo initially asks, “Would an automatic email save you from checking the queue?” She catches the implied benefit and replaces it with the recent-task prompt before continuing. The earlier leading prompt remains a limitation even after Jo changes it. Her notes flag it so the team does not treat the participant's initial agreement as strong evidence for automation.

The round also includes a mock, researcher-created two-row queue artifact. It shows one item awaiting language review and another awaiting product approval, both under a broad pending heading. The participants discuss the example, but Jo labels it as a fictional discussion aid. It is not an observed employer screen, an independently verified workflow, or a usability test of a new design. The complete aid is:

Mock item Broad heading Researcher-added scenario annotation
D-08 Pending Waiting for language review
D-09 Pending Waiting for product approval

The annotations describe invented scenarios, not fields confirmed in the current tool.

Keep notes legible about their status. A participant's recollection is reported experience. A permitted artifact can show what was on that artifact, but does not establish every action around it. An interpretation belongs in a separate column. If you did not measure duration, inspect logs, or watch a task, do not silently convert “this takes ages” into a time-saving baseline.

A research notebook separates reported evidence, a proposed requirement, and a pending validation question, with the requested email sketch kept as one possible option.

Map evidence to requirements and assumptions without losing the limits

NN/g's User Interviews 101, reviewed July 15, 2026, distinguishes interview reports from observed behavior and from usability testing. Memory, missing details, and social desirability can affect accounts. That limitation matters inside an organization, where participants may want to support a sponsor's idea or avoid appearing critical of a colleague.

Worked asset 2 — completed fictional evidence-to-decision map. The evidence IDs below identify simulated records, not real transcripts. Requirements are proposed responses to the task; assumptions and validation steps remain explicit.

Evidence and its status Need or proposed requirement Assumption, next check, and decision owner
E1: Rae reports comparing labels and comments to prepare Friday's list. Preserve the reason an item is waiting and the action needed next. Jo must verify the reported sequence through an authorized task walkthrough. No measured preparation-time claim.
E2: Luca reports that a product-approval block differs from language-review work; he corrects the readback. Show blocked items without presenting them as actionable language-review assignments. Product owner Vik will review this distinction with the document owner; source status definitions are not yet confirmed.
E3: Nila reports relying on a screenshot during coverage and being unable to establish freshness. Give a backup an approved way to identify the current report and its relevant date or state. Access and information freshness need an actual check. Jo will include an occasional backup in the next round.
E4: The mock two-row artifact uses one broad pending heading. Use it to explain the candidate distinction during synthesis, not as proof of production data quality. Actual queue representation remains unknown; the mock cannot establish that a status field already exists.
S1: Sponsor Vik wants a Monday email so review work is visible. Retain an email digest as candidate solution H1; consider a current shared view as H2. Neither option's usefulness, permissions, or usability is tested. Sponsor preference does not resolve delivery design.
Gap G1: The other language team and accessibility-support needs were not included. Avoid committing a shared workflow for all teams based on this round. Jo proposes additional research coverage; scheduling and participation are not established in this article.

The synthesis records this problem statement: “Staff need to distinguish actionable translation review from work waiting on another decision, and backups need an approved current source. The proposed weekly email may help visibility, but its content, timing, and ability to support action remain untested.” That statement preserves the original request without disguising it as a finding.

In the fictional synthesis review, Vik accepts the distinction as a basis for further discovery and asks Jo to prepare a paper-prototype plan. He does not authorize production automation, broaden access, or declare a delivery date. The document owner still needs to confirm the actual status meanings. The task walkthrough and prototype session remain proposed research; no participant scheduling, study approval, or test result is established.

The completed decision note reads: “Do not build the automatic sender from this round alone. Preserve the actionable-versus-blocked distinction and backup-current-source need in the next design brief. Validate source definitions and access, include the missing perspectives, and compare candidate approaches through an authorized study. No efficiency, adoption, or accuracy result is established.”

Close the loop without turning discovery into a promise

Send participants an appropriately scoped summary through the authorized process: what you understood, what remains uncertain, and how the team will use the findings. Invite corrections about the account they supplied. A participant confirming that a summary reflects their experience does not approve the final product design or establish organization-wide consensus.

Keep the interview record and the product decision distinct. A requested capability can enter the backlog with a research reference and an unresolved assumption. A production change still needs the team's actual design, technical, access, and release decisions. Interviews do not remove those responsibilities, and a sponsor's enthusiasm does not supply missing evidence.

When the next question is whether someone can complete a task in a proposed interface, use an appropriate usability study rather than asking whether they like the idea. When the question is frequency or elapsed time, seek an authorized source suited to that measurement. You can combine methods, but label what each one actually establishes.

For internal tools, resist blaming workarounds on careless users. A scratch list, screenshot, or personal reminder may reveal a missing state, unclear ownership, or an access constraint. Investigate what it accomplishes before removing it. The team may ultimately choose the requested automation, a smaller change, a different tool, or no build; the interview's value is the evidence supporting that decision, not a predetermined answer.

Start with one recent task and a neutral question, then preserve the chain from reported experience to requirement, assumption, and next check. More workplace guides are available from LandOffer.ai; the next step is to write the learning question and identify who actually performs the work before committing to the workflow they first requested.

Sources and Further Reading