Why requests swamp analytics teams
Requests reach an analytics team by every route: a message to whichever analyst replied last time, an email to the team, a comment in a meeting, a slide that needs a number by tomorrow. Each looks small to the person asking. Together they fill the week, and the pattern is familiar.
- Requests describe outputs, not needs. "Can I get conversions by channel for the last six months as a chart?" says what to build and nothing about why. The analyst builds it, the requester looks at it, and the real question, often about whether to move budget, is asked again a week later in a different form.
- Every request is urgent. Without a stated deadline and the reason for it, the loudest and most senior requests go first, whatever their value.
- The same questions are answered repeatedly. Without a record of past answers, the analysis done for one manager in spring is redone for another in autumn, often with a different result because definitions or data changed in between.
- Recurring questions are answered by hand. A weekly figure that should live in a dashboard is assembled manually every week, which is one of the causes the diagnosis of a slow monthly report traces.
- Nobody learns what happened next. The team delivers and moves on, never finding out whether the answer changed anything, so it cannot tell its valuable work from its busy work.
The result is a team that is always busy and rarely influential. The process below changes what is asked, what is answered and what is remembered.
Intake: capture the decision
Intake has one entry point, a short form or ticket in whatever system the team already uses, and every request goes through it, including the ones that arrive in a meeting. The form's job is to turn a request for an output into a statement of a decision. Kaushik's measurement model starts from the business objective and derives goals, measures and targets from it; an intake form that asks for the objective and the decision first applies the same order to individual questions.
| Field | What to ask | Why it matters |
|---|---|---|
| Decision | What decision will this answer inform? | A request with no decision is a candidate for self-serve or for declining |
| Decider and date | Who makes the decision, and when? | Separates a real deadline from a preference |
| What would change | What will you do if the answer is one way, and what if it is the other? | If both answers lead to the same action, the analysis is not needed |
| Question | The question in one sentence | Forces the requester to state it, not the analyst to guess |
| Objective and KPIA metric tied directly to an objective. | Which business objective and measure the decision affects | Links the request to the plan and exposes vanity metricsMetrics that look impressive but do not inform a decision or track a business outcome. |
| Current belief | What do you expect the answer to be, and why? | States the HypothesisA testable statement of the change to make, the effect expected on a named metric and the evidence-based reason for expecting it. so the analysis can test it |
| Audience | Who will see the answer? | Decides the format and how much explanation it needs |
| Recurrence | Will you need this again? On what schedule? | Recurring questions are routed to a dashboard |
The "what would change" field does more work than any other. Asking it often ends the request at intake, because the requester realises the decision is already made, or turns a vague request for "everything about the spring campaign" into a precise one about whether to repeat it. It also tells the analyst what precision the answer needs: a decision that flips only if a figure halves does not need the same care as one that flips on a small difference.
Keep the form short. A field earns its place if its absence has caused rework before; a long form pushes people back to direct messages, which is where the process started.
Triage: four kinds of request
Triage sorts each request into one of four kinds, because each kind has a different route and a different owner. A short triage slot on a fixed schedule works better than a continuous trickle, and the person running it needs the authority to decline.
| Kind | Signs | Route |
|---|---|---|
| Fix | A number looks wrong, a dashboard is blank, a figure disagrees with another system | Straight to the dashboard or data owner; certified dashboards first |
| Recurring report | The same question on a schedule: weekly, monthly, every campaign | Into a governed dashboard, added to the inventory with an owner |
| One-off analysis | A pending decision that needs a bespoke answer | Sized, prioritised and scheduled |
| Self-serve | The answer already exists in a certified dashboard or the findings library | Answered with a link, and a few minutes showing the requester where it lives |
Fixes jump the queue when they affect a certified dashboard, because every decision taken on a wrong number compounds the cost. Recurring requests are the most valuable to catch, since each one routed to a dashboard removes a weekly task for good; the dashboard governance guide covers how the new dashboard is owned and certified, and the metrics layer makes sure its figures match the ones in every other report. Self-serve requests are the cheapest to answer and the best chance to teach; the guide to building data literacy covers how a team learns to find answers itself.
Sizing and priority
One-off analyses are sized and then prioritised, in that order, and the outcome is visible to everyone who asks.
Size the effort in broad bands rather than hours: an answer from existing tables, a new query or join, a new data source or a modelling exercise. Analysts get better at sizing when they compare their estimate with the actual effort after each request, and the team's own record of past requests becomes the guide.
Judge the impact from the intake answers: how large is the decision, how likely is the answer to change it, and when must it be made. A request behind a large budget decision whose outcome is genuinely open ranks above a request for a figure in a slide whose conclusion is already written.
The impact-effort matrix turns the two judgements into a picture the whole team can argue about: quick wins first, major projects scheduled deliberately, fill-ins when there is slack, and the money pits, high in effort and low in impact, declined or simplified. Two rules keep it honest. Keep some capacity free for fixes and genuinely urgent work, sized from how much of that the team handled in recent quarters. And agree who sets priority when requests compete; a RACI matrix with one accountable person, usually the marketing lead the team serves, stops analysts from being the referees between their own stakeholders.
Turnaround agreements belong here too. Agree a target time to acknowledge and to answer each kind of request, based on what the team has actually achieved rather than what sounds good, and publish the queue so that a requester can see where their request sits and what would have to move for it to go faster.
Delivering the answer
The answer is shaped by the decision, not by the analysis. Lead with the answer to the question asked at intake, in a sentence, then the evidence, then the caveats.
| Format | When it fits | What it contains |
|---|---|---|
| One-line answer | A factual question with a clear answer | The figure, its definition and source, and the period |
| Short written note | Most one-off analyses | The question, the answer, the evidence, what is uncertain, and a recommendation |
| Table or small chart | Comparisons the reader needs to see | The figures with counts and the comparison basis stated |
| Dashboard | Recurring questions only | A governed, owned view; never built for a one-off |
| Presentation | Large decisions taken in a meeting | The note's structure, with the answer on the first slide |
A short written note is the default because it can be read without the analyst in the room and filed for reuse. Write the commentary as what happened, what it means and what to do, and be explicit about uncertainty: an answer that rests on a handful of conversions says so, as the guide to reading small numbers explains, and an explanation of a moved metric shows how it was tested, as in explaining a change in a marketing metric. If the analysis found that the question was the wrong one, say that first; it is often the most useful finding. Before an answer that informs a large decision goes out, a second person can check it with the routine for reviewing an analysis before it is shared.
Reuse and self-serve
Every answered request becomes an entry in a findings library: a searchable list, in whatever tool the team already uses, of the questions answered, with the date, the answer, the data and definitions used, the analyst, the requester and the decision it informed. Before any one-off analysis starts, someone searches the library. A recent answer can be reused or refreshed; an old one shows how the question was approached last time and whether the conclusion has since been overtaken.
The library also shows patterns. A question asked three times in a quarter is a recurring question, and belongs in a dashboard. A cluster of questions about one metric usually means its definition is unclear, which is a job for the metrics layer rather than for more analyses. Requests that keep arriving for figures nobody decides anything with are worth raising with the people asking, gently, because they are a sign that the team's reporting is measuring activity rather than outcomes.
Self-serve is the other half of reuse. Certified dashboards with definitions linked from every metric, a findings library people can search themselves, and regular office hours where requesters can bring a question and learn where its answer lives, together remove a large part of the queue. Self-serve works only when the dashboards are trusted; a team that sends people to a dashboard whose numbers disagree with the finance report will get the request back, with less goodwill.
Closing the loop
The process ends when the decision is taken, not when the answer is sent. At the decision date recorded at intake, someone, usually the analyst, asks the requester what was decided and records it in the findings library: changed, confirmed, postponed or not taken. It is a short conversation, and it does three things. It tells the team which kinds of request lead to decisions and which do not. It gives requesters a reason to state the decision honestly at intake. And it builds a record of the team's influence that is far more persuasive at budget time than a count of tickets closed.
Measure the service each quarter with a few figures from the team's own records:
- Requests received, by kind.
- Time from intake to answer, by kind, against the agreed turnaround.
- The share answered by reuse or self-serve.
- The age of the oldest open request.
- The share of delivered analyses that changed or confirmed a decision.
Review them with the marketing lead who sets priority, and change the process where the figures point: more self-serve if the same questions keep arriving, stricter intake if analyses keep changing nothing, a new dashboard if one kind of recurring request dominates. The BI hub places this service alongside the warehouse, definitions and dashboards it depends on.
Spindlewort Energy's first quarter with decision-first intake
Spindlewort Energy is an invented energy supplier with two marketing analysts. In the quarter before the change they received 148 requests, mostly by direct message, of which 61% asked for a specific chart and none recorded a decision. The backlog at quarter end stood at 41 requests, and the median one-off analysis took eight working days.
The team introduced an intake form that leads with the decision, the decider and date, and what would change, a twice-weekly triage slot run by the head of acquisition, and a findings library. In the following quarter it received 112 requests:
| Kind | Requests | Route taken |
|---|---|---|
| Fix | 14 | Sent to dashboard owners; 11 traced to one broken connector |
| Recurring report | 18 | Folded into three certified dashboards |
| Self-serve | 27 | Answered with a link; 16 of them from library entries written that quarter |
| One-off analysis | 53 | 8 declined or merged for lack of a decision; 45 delivered |
Follow-up on the 45 delivered analyses found that 19 changed a decision, 15 confirmed one already planned and 11 had no recorded effect, mostly because the decision was postponed. The median one-off analysis took five working days and the backlog at quarter end was 12. The head of acquisition used the 19 changed decisions, with their budgets, to argue for a third analyst.
Common mistakes
- Taking requests by direct message. Every request goes through intake, including the ones from senior people and the ones made in meetings.
- Asking for the output, not the decision. A form that asks which chart is wanted reproduces the problem it was meant to solve.
- Treating every request as a one-off. Recurring questions answered by hand consume the team; route them to dashboards.
- Building dashboards for one-off questions. They become unowned and outdated; answer in a note.
- No authority to decline. Without it, triage is just a queue with extra steps.
- Prioritising by seniority. Use impact and effort, agreed in advance, and make the queue visible.
- No findings library. The same analysis is redone, and sometimes reaches a different answer.
- Stopping at delivery. Without the follow-up, the team cannot tell influential work from busy work.
- Measuring tickets closed. Speed without relevance is the problem the process exists to fix.
