How do I audit a client backlog?

Audit a client backlog by scoring every story against the same explicit criteria, then reporting the aggregate picture before any individual example. The audit that holds up in front of a client is full-coverage and criteria-based; the one that gets argued with is a sample of stories the reviewer happened to dislike.

This guide is the method end to end: what to ask the client for, why sampling undermines you, what to evaluate in each story, and how to package the findings so they lead to a decision instead of a debate.

What should I ask the client for?

One file: a full export of the active backlog as CSV or Excel. Every major tracker produces one in minutes, and asking for an export instead of tracker access keeps the audit scoped, avoids a security review, and gives you a frozen snapshot to score. Ask for four columns plus one that clients forget:

• Summary, description, acceptance criteria: the text you’re evaluating. Only the summary is strictly required, but criteria coverage is usually where the findings live.
• A stable ID column (the issue key): this is what lets a second scoring round match stories to the first, so you can show improvement instead of a one-off snapshot.
• Epic and author, if available: they unlock quality-by-epic and quality-by-author views. Keep authors anonymized in anything the client sees.

Should I sample the backlog or score all of it?

Score all of it. A sampled audit invites the response that kills the engagement: “you picked the bad ones.” Full coverage turns the conversation from anecdotes into distribution (“38% of the backlog scores below the readiness bar, and it’s concentrated in the checkout epic”), which nobody can dismiss as cherry-picking. Manual review is why auditors sample; a senior reviewer at 10 minutes per story clears maybe 40 stories a day. Automated INVEST scoring removes that constraint: backlogs of 100,000+ stories are in scope, so coverage stops being the bottleneck and your time goes into judgment instead.

What am I looking for in each story?

Six qualities, applied identically to every story: whether it’s independent of hidden cross-team work, negotiable rather than a disguised implementation order, clearly valuable, estimable from the text alone, small enough for a sprint, and testable. In practice, client backlogs fail in recognizable patterns: acceptance criteria that exist but can’t be tested (“works correctly on mobile”), stories that bundle three features under one summary (“Improve the checkout experience”), and dependencies that live in the author’s head instead of an issue link. The finding you deliver is the pattern plus the count, not a complaint about any one story.

How do I turn findings into a deliverable?

Build the report aggregate-first: quality posture up front, the per-dimension diagnosis, the score distribution, and quality by epic where the data supports it. Illustrate with a small, capped list of the lowest-scoring stories and one or two “what good looks like” examples, and attach the complete per-story dataset as a spreadsheet so nothing looks hidden. Write the executive summary yourself; it’s the one page the sponsor will actually read. This structure is also how you stay diplomatic: aggregates critique the backlog, not the people who wrote it.

Vindex for Agencies produces exactly this deliverable: it scores the full export, gives you a worst-first review workspace to exclude junk and add analyst notes, and builds the report white-labeled under your brand, with the full dataset attached. The report builder reference lists every section and branding option.

How do I prove the audit worked?

Re-score after the client acts. The first audit is a snapshot; the second round is evidence. Keep the ID column stable between exports so stories match across rounds, then show the movement: the share of stories above the readiness bar, the dimensions that improved, the epics that still lag. Round-over-round improvement is both the proof the engagement worked and the natural opening for the follow-on: the lagging epic becomes the next statement of work.