Here’s a common scenario: A manager asks, “What’s our customer retention rate going to be next quarter?”
The analyst could say “94.2%” and move on. The model predicted it. The number is right there in the notebook. Easy answer.
But a better answer would be: “Based on current trends, probably between 92% and 96%, with 94% being most likely. But if that enterprise client we’re worried about churns, it could drop to 89%.”
The second answer is actually more helpful. Now leadership knows what to plan for.
This is the whole game with uncertainty. Stakeholders don’t want fake precision. They want to make good decisions. And good decisions require understanding what you know, what you don’t know, and what could go wrong.
Here’s how to communicate uncertainty without sounding unsure of yourself.
Here’s what happens in most meetings:
Stakeholder: “How many units will we sell next month?”
Analyst: “Well, based on the model, we’re looking at somewhere between 8,000 and 12,000 units, depending on several factors including...”
Stakeholder: “Just give me one number. What’s your best guess?”
Analyst: “Um, 10,000?”
Stakeholder: “Great, I’ll tell the VP we’re doing 10,000.”
Then the company sells 8,200 units and the analyst gets asked why they were off by 18%.
This happens because:
Single numbers are easier to put in slides and spreadsheets
People feel more comfortable with certainty, even false certainty
Nobody taught stakeholders how to think probabilistically
The analyst didn’t push back effectively
The solution isn’t to refuse to give estimates. It’s to give estimates that include the uncertainty in a way stakeholders can actually use.
This is when you’re not sure about the data itself.
Bad way to say it: “The data might have some errors, so I’m not totally confident in these numbers.”
Better: “Our survey had 1,200 responses with a margin of error of ±3 percentage points. So when I say 67% of customers are satisfied, the true number is most likely between 64% and 70%.”
Why it works: You’re specific about what you don’t know and how much it matters. The stakeholder can decide if a 6-point range is good enough for their decision.
Real example: A team was tracking app usage, but their tracking code was broken for two weeks in March. Instead of ignoring it, they showed two versions of the chart: one with March excluded, one with March marked as “incomplete data.” They explained that March numbers could be 20-30% understated. The product team used the chart with the caveat and made their decision knowing the limitation.
Your model makes predictions, but it’s not perfect.
Bad way to say it: “The model is usually pretty accurate, so we should be fine.”
Better: “The model has been correct within 10% about 80% of the time. That means roughly 1 out of 5 times, we could be off by more than that.”
Why it works: You’re giving them odds they can understand. They know what 1 out of 5 means. They can assess if that’s acceptable risk.
Real example: A team built a churn prediction model and told the retention department: “The model flags 500 customers as high-risk. Based on testing, about 300 of them will actually churn, and we’ll miss about 100 who will churn but weren’t flagged. So there will be some false alarms and some misses, but we’ll catch most of them.”
That’s way more useful than “the model is 85% accurate” which doesn’t tell them what to expect operationally.
Nobody can predict the future. Events happen that change everything.
Bad way to say it: “Obviously this assumes nothing major changes in the market.”
Better: “This forecast assumes normal conditions. If a competitor launches a similar product, we could see 20-30% lower sales. If we get that press coverage we’re hoping for, could be 15% higher.”
Why it works: You’re identifying specific risks and opportunities with rough magnitudes. They can plan for scenarios, not just one number.
Real example: During Covid, an analyst was forecasting Q4 revenue. Instead of one number, they gave three scenarios:
Pessimistic: Second lockdown happens, revenue drops 35%
Baseline: Current trends continue, revenue down 12%
Optimistic: Vaccine news boosts consumer confidence, revenue down 5%
Finance used all three to build contingency plans. When the actual result landed at -8%, they weren’t surprised because it was in the range discussed.
Never say “confidence interval” to a non-technical stakeholder. Their eyes will glaze over.
Here’s what to say instead:
“Our best estimate is 10,000 units, but we could realistically see anywhere from 8,500 to 11,500.”
That’s a confidence interval. You just didn’t call it that.
“There’s about a 90% chance we’ll sell between 8,500 and 11,500 units. Most likely around 10,000.”
Same information, framed as probability. Some stakeholders prefer this.
“Here’s what we expect in different scenarios:
If things go poorly: 8,500 units
Most likely: 10,000 units
If things go well: 11,500 units”
This one works great for executives who think in scenarios.
Pick based on your audience. Finance people often like ranges. Executives like scenarios. Operations people like odds because they think about capacity planning.
Here are specific phrases that work well in practice:
“Based on the data, I’m confident this is between X and Y.”
“The data strongly suggests...”
“We’re seeing consistent evidence that...”
“I’d be comfortable betting on this.”
“My best estimate is X, but there’s meaningful uncertainty here.”
“The most likely outcome is X, with a realistic range of Y to Z.”
“I’d say we’re at about 70-80% confidence on this.”
“This is our best guess given what we know today.”
“The data doesn’t give us a clear answer on this.”
“There are several possible explanations, and I can’t distinguish between them with the current data.”
“I can give you my hypothesis, but I wouldn’t act on it without more validation.”
“This is speculative, but here’s what the data suggests might be happening...”
“To answer this with confidence, I’d need [specific data]. Without it, I can only give you rough estimates.”
“The analysis we have points to X, but I’d want to validate it by [specific action] before making a major decision.”
“We could wait another month to get more data, which would narrow the uncertainty from ±20% to ±10%.”
“I understand you need a number for planning, and my best estimate is X. But I need you to know it could reasonably be Y or Z instead.”
“If I had to pick one number, I’d say X. But that’s forced precision - the honest answer is we’re looking at a range.”
“I can give you a single number, but it’ll have a significant margin of error. Is that okay for this decision, or should we build in some buffer?”
That’s not helpful. Be specific about what you’re unsure of and why.
Better: “I’m not sure if the trend will continue because we’ve only seen two months of data. By next quarter, we’ll have a better picture.”
That sounds like you’re making excuses.
Better: “We’re missing data from the mobile app, which is about 30% of our traffic. So these numbers undercount total activity.”
That sounds like you haven’t done your job.
Better: “Based on historical patterns, we’re most likely looking at 8,000-12,000 units. Outside that range would be unusual and would indicate something major changed.”
Models don’t say things. You interpret models.
Better: “Based on the model and my judgment, I expect...”
Most people don’t understand statistical confidence levels.
Better: “I’m very confident, but not certain” or “There’s a small chance I’m wrong, but I’d be surprised.”
Numbers with ranges don’t always stick. Sometimes you need to show it.
Instead of:
Q1 Sales: 100,000 units
Q2 Sales: 105,000 units
Q3 Sales: 110,000 units
Show:
Q1 Sales: 95,000 - 105,000 units (most likely: 100,000)
Q2 Sales: 98,000 - 112,000 units (most likely: 105,000)
Q3 Sales: 101,000 - 119,000 units (most likely: 110,000)
Even better: Show it as a chart with a shaded confidence band around your forecast line. The shading shows the uncertainty visually.
Let’s say you’re comparing two marketing campaigns:
Campaign A: 8.2% conversion rate (±1.1%)
Campaign B: 9.1% conversion rate (±1.5%)
The ranges overlap (7.1-9.3% vs 7.6-10.6%). That means we can’t be certain B is actually better. Show both as bars with error bars, and explain: “B appears better, but the difference might just be random chance. We’d need more data to be sure.”
If you have incomplete or questionable data, mark it clearly on the chart. Don’t hide it. Use different colors, dotted lines, or annotations that say “preliminary data” or “partial month.”
What they say: “I need to know for certain if this will work.”
What you say: “I can’t give you certainty, but I can tell you the odds. Based on similar initiatives, we have about 70% chance of hitting target, 20% chance of exceeding it, 10% chance of falling short. The biggest risk is [specific risk]. If you’re okay with those odds, we should proceed.”
Why it works: You’re honest about uncertainty but still helping them decide. You’re not paralyzing them with doubt.
What happened: You said “between 8,000 and 12,000” and now they’re telling everyone “10,000 guaranteed.”
What you do: Correct it immediately. Email or mention in the next meeting: “Quick clarification on forecasts - I want to make sure we’re all working with the same numbers. My estimate was 8,000-12,000 units, with 10,000 being most likely. The range matters for planning purposes.”
Why it matters: Let this slide once and you’ll get blamed when reality lands at 8,500.
What they say: “Why can’t you be more precise? Don’t we have all the data?”
What you say: “We have good data on [what you know], but [what you don’t know] introduces uncertainty. For example, we don’t know if that competitor will launch next quarter. That alone could swing results by 15-20%. I can give you a precise number, but it won’t be accurate. Or I can give you an honest range that helps you plan for different outcomes.”
Why it works: You’re explaining the trade-off between precision and accuracy. Most people intuitively get this.
What happened: You said 10,000 units (range 8,000-12,000), and actual was 7,200.
What you do: Own it, then explain. “We missed the forecast. We landed at 7,200 vs projected 8,000-12,000. The main driver was [specific thing that changed]. Here’s what was learned and how forecasts will be improved going forward.”
Why it matters: If you’re honest about uncertainty upfront, being wrong occasionally is expected. If you pretended to be certain, now you look incompetent.
What they ask: “What’s causing the drop in engagement?”
What you say: “I’ve looked at the data and I have three hypotheses: [A, B, C]. I can’t determine which is the real cause without [specific additional data or test]. If this is urgent, I recommend we [interim action], and I’ll work on getting clarity on root cause.”
Why it works: You’re not pretending to know when you don’t. But you’re also not leaving them helpless - you’re giving hypotheses and next steps.
Here’s the paradox: being uncertain makes stakeholders trust you more, not less.
When you say “I’m 100% sure” and you’re wrong, you lose credibility forever.
When you say “I think it’s X, but it could be Y, and here’s why” and you’re wrong, they remember that you warned them.
Some analysts try to hide uncertainty because they think it makes them look weak. It does the opposite. Stakeholders work with uncertainty every day - they run businesses, they make bets, they manage risk. They respect people who are straight with them about what they do and don’t know.
The analyst who says “probably around 10,000, could be 8-12” looks more competent than the one who says “exactly 10,247” based on a model that’s never been validated.
Sometimes you’ve communicated all the uncertainty and stakeholders still can’t decide. They want more data, more analysis, more certainty.
Your job then is to help them understand the cost of waiting.
What you say: “I understand you want more certainty. We could wait another month and reduce uncertainty from ±20% to ±15%. But waiting costs us X in potential revenue / time to market / competitive position. Given that trade-off, what do you want to do?”
Frame it as: cost of uncertainty vs cost of delay.
Sometimes the right answer is to decide with incomplete information. Sometimes it’s worth waiting. But make the trade-off explicit.
Before you present any analysis, ask yourself:
About the data:
What data am I missing?
How accurate is my data?
What’s my margin of error?
Are there known issues I should flag?
About the analysis:
How confident am I in this conclusion?
What assumptions did I make?
What could make me wrong?
Have I validated this?
About communication:
Have I quantified my uncertainty?
Can stakeholders use this range to make decisions?
Did I identify the biggest risks?
Did I give them scenarios or alternatives?
About next steps:
What would reduce uncertainty?
Is it worth getting more data?
What’s my recommendation despite uncertainty?
The best data analysts are comfortable saying “I don’t know, but here’s what I do know and here’s how we could find out.”
The worst ones pretend to have answers they don’t have.
Your job isn’t to eliminate uncertainty. Most business decisions happen under uncertainty. Your job is to quantify it, communicate it clearly, and help stakeholders make good decisions anyway.
Next time someone asks for a number, don’t just give them one number. Give them the number with context about how much you trust it and what could change it.
That’s not being uncertain. That’s being honest. And in the long run, that’s what gets you invited to more important conversations.
Because stakeholders don’t want analysts who are always right, they want analysts they can trust. And trust comes from telling the truth, even when the truth is “I’m not entirely sure, but here’s my best assessment.”
Best of luck for everything!
- Sai Bysani, a fellow Hustler!
Keep grinding, keep growing,
The Data Hustle.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.