Ask a data engineer what eats the week and the honest answer is rarely “pipeline design”. It is the steady drip of side requests: can you pull this extract, add one more column, join the finance sheet, join a call about it. Across data teams, the way people handle those asks is changing, and the shift has a name worth using: saying no politely, on purpose, as a working practice rather than an apology.
The state of play: demand outruns capacity
Data work has become a shared utility. Marketing wants cohorts, operations wants latency numbers, and leadership wants a dashboard before the offsite. Teams grew, but requests grew faster. The result is a familiar pattern. Engineers accept almost everything, quality slips in the work that matters, and the backlog of promised extras becomes its own source of stress.
The old answer was heroics: work late, say yes, catch up later. That approach has a poor track record. People burn out, and the extras are delivered without tests, which then become tomorrow’s incident. Teams that have noticed the trap are moving to a plainer rule: a request is a proposal, not an order, and a proposal can be declined.
Trend one: requests are triaged in writing
Increasingly, asks flow into a single shared queue where the requester fills in three blanks: the need, the deadline and the audience. The template is not red tape. It forces the requester to think, and it gives the team something concrete to answer. A reply such as “this belongs to next month’s planning, and here is why” is much easier to write when the request has a shape.
Where intake is chaotic, refusals tend to be emotional. Where it is structured, they become routine, almost administrative, and nobody takes them personally.
Trend two: a no comes with a smaller version
The most useful change is that refusals now carry an alternative. “We cannot build the full self-service report this sprint, but we can hand you a one-off extract and a short note on how it was made.” The requester gets partial value at once and the team protects its plan. Done well, the smaller version often turns out to be all that was needed.
Typical smaller versions in a data team include a sampled extract instead of the full table, a one-time query instead of a scheduled job, a documented manual process instead of automation, and a conversation about definitions before any new field is created.
Trend three: quality is the stated reason
The best-received refusals cite the thing everyone already wants, trustworthy numbers. “If we add that join without checks, the figures may drift, and nobody will notice until a report is wrong.” That framing treats the refusal as protection of the requester’s own interests. It aligns naturally with practices described in guides on writing validation rules for datasets before they feed reports, where the whole point is that unchecked data should not move downstream.
Similarly, teams that standardise their transformation layer, as laid out in a walkthrough of modelling transformations in a modern pipeline, find it easier to say that a request needs to enter through the model rather than through a side script.
A worked example: the extra column
An analyst asks the data team to add a customer segment column to a core table, today. The team lead weighs it. The column depends on a definition nobody has agreed, and three downstream jobs read that table. Accepting means changing a shared asset in a hurry.
The reply sent: “Thanks for the request, the segment idea is useful. We cannot change the core table today because other jobs depend on it and the definition is still open. By Wednesday we can give you a separate view with a provisional segment, labelled as such, and book twenty minutes next week to agree the definition. If it holds up, we will promote it properly.” Six sentences, polite, specific, no drama.
What practical teams are measuring
Few teams track refusals formally, but several informal signals are worth watching. How long does an answer take? A same-day no is a healthier sign than a week-long maybe. How many requests come back a second time? Repeat asks suggest the first answer was unclear. How often does a smaller version satisfy the requester? A high rate means the team is saying no to scope, not to people.
None of these needs a dashboard. A shared note listing request, reply and revisit date is enough, and it doubles as a record when priorities are questioned later.
Where this is heading
Expect polite refusal to become part of onboarding for data roles. Newer hires are often the most eager to please and the least certain they are allowed to decline. Giving them a few approved sentences, plus visible backing from the lead, removes the fear. The skill is closer to customer service than to engineering, and it is learnt the same way: by example and repetition.
Expect also a gradual change in how requesters behave. Once they see that a decline arrives quickly and carries a sensible alternative, they start asking more precisely. The intake template improves, the asks shrink, and the team’s calendar loosens.
A side note on lookups that are not yours to answer
Data teams often field questions that sit outside their own datasets, such as which organisations operate mobile networks. The tidy reply is a pointer, not a research project. For that particular topic, eSIMGlobo, an independent directory built on official network-code data, offers a starting place that a colleague can browse alone.
Practical checklist for this quarter
- Agree one place where requests arrive, with three required fields.
- Write two or three approved decline sentences and share them with the team.
- Reply the same day, even if only to say that the answer is no.
- Always pair a refusal with a smaller version or a date.
- Keep a short record of what was declined and revisit it monthly.
None of it is complicated. The challenge is habit, and habits form when the first few polite refusals go well. Start with the next request you do not have room for, answer it kindly and briefly, and note how the conversation goes. The trend is real because the alternative, accepting everything, has already been tried.