Roadmap: Ordered Bets, Not a List of Dates

From the field: the January Gantt, the March table

On 8 January I presented a six-month roadmap to management. Nine items, each with a start and end date, and coloured bars across four teams. Nobody objected in the meeting. Two days later, the sales team told a corporate customer: “US stocks are coming in March.” The bar was there. It is not surprising that they read it as a promise.

On 31 March, the four first-quarter items looked like this:

The work in the last row was six days long. On the calendar it took 38 days. We redrew the Gantt chart three times that quarter: 8 January, 12 February, 9 March. Each time the dates moved right, and the bars did not get shorter. The March promise that sales gave the customer was broken. We had a difficult phone call with that customer in April.

But the row that really bothered me was the first one. Price alerts finished on time. By the Gantt chart, it was a success. At the end of March, 1.8% of active customers had set at least one alert. Was that a lot or a little? We did not know. In January, nobody had written down “if this works, what will we see?” I had not written it either.

What a roadmap is for

The job of a roadmap is not to give a calendar. Its job is to put teams and management on the same page about “what first, and why”. The calendar is a side product of that question. A Gantt chart turns the side product into the main product. People discuss the length of the bars, not the reason for the order.

One more thing needs to be separated. The no list in the technical strategy post says what we will not do this year. The roadmap says the order of what we will do. They are not the same page. Something on the no list does not go into the roadmap's Later column either. Later is a waiting room, not a bin.

How it should work: Now, Next, Later

At the start of April we removed the Gantt chart. In its place came one page with three columns. The rules of the columns matter more than the columns themselves:

The range in Now is calculated with the method from the “When will it be done?” post: not one date, but something like “85% likely between 10 and 30 June”. Next has no dates. A date on work whose size you do not know yet gives a weapon to anyone who will read it as a promise. The sales team can talk to customers only about items in Now, and they give the pessimistic end of the range.

The order of Next

Next has no dates, but it has an order, and the order is where the real discussion happens. We ask three questions when we order it. If this bet wins, which number changes, and by how much? How many weeks until the smallest release? If it loses, what will we have learned? The third question felt strange at first. But a bet with high learning value can change the order of the next two items, even when it loses.

The Product Owners set the order, and I challenge it. I tried the opposite too: I wrote the order and the teams challenged it. We lost two weeks on the first draft in April because of that. Most objections were not about the order. They were about not knowing the reason for it. Now every item in Next has a one-sentence note: “why it is in this position”. If you cannot write the note, the item is not ready to be ordered yet.

Every item is a bet

The real change was not the columns. It was how we wrote the items. On the Gantt chart, an item was a feature name: “Price alerts.” On the new roadmap, every item is a bet card. We call it a bet because it can lose, and we want to see early when it is losing.

BET       : regular investment plan (monthly automatic buy)
OWNER     : money-transfers team, PO: Burcu
COLUMN    : Now (2026 Q2)

PROBLEM   : 38% of active customers made no trade in the last 90 days
HYPOTHESIS: if a monthly automatic buy is easy to set up, some of these
            customers will set up a plan and stay
MEASURE   : 3% of active customers set up a plan in the first 6 weeks
EARLY     : 1% by week 3 (small release, only 2 equity funds)
STOP      : below 0.5% in week 3 -> one extra week of trying,
            then stop if still below, move to Later
RANGE     : 85% likely small release between 20-30 April

RESULT    : week 3 at 0.4%. Stopped in week 5. (see below)

The card has five main lines, and each one has a job. The problem is written with a number. Otherwise nobody can disagree with it. The hypothesis follows the pattern “if we do this, that will happen, because”. The measure has a number and a time limit. The early signal is the result of a small release that you can check in two or three weeks. The stop condition is written before the work starts. Once the work starts, nobody wants to give up their own bet. I do not want to either.

The losing bet

The card above lost. We started work on 13 April, and the small release went out on 27 April. At the end of the third week, 0.4% of active customers had set up a plan. The threshold was 0.5%. Burcu asked for one more week, with a notification campaign. We gave it, because this had also been agreed before the work started: one extra try. Week four: 0.45%. At the start of week five we stopped and moved the card to Later.

On the Gantt chart, this work was “April–June, 11 weeks”. With the full version, we would have spent five more weeks. At the end of June we would have celebrated a 0.4% feature as “delivered on time”. The Money Transfers team gave those five weeks to the first item in Next.

Linking the roadmap to the sprint goal

In the Gantt period, we discussed the roadmap at the start of the quarter and did not open it again until the next quarter. Sprints ran on their own. In the sprint goal post I explained that the goal should be one sentence. Now we add one more line next to that sentence: which bet it serves.

In the second quarter, the four teams wrote 26 sprint goals in total. 23 were linked to a Now bet. The other three were one regulatory task and two technical debt sprints, and they were clearly marked “outside the bets”. A goal outside the bets is not forbidden. Being invisible is forbidden. If there are more than three such goals in a quarter, we assume the roadmap is not telling the truth.

How it breaks

We saw four kinds of breakage in three months:

• Later becomes a graveyard. At the start of June there were 41 items in Later. Nobody wanted to say “no”, so every idea went to Later. At the start of the quarter we pruned it to 12. The other 29 were either closed or sent back to their owner with a note: “not now, and here is the condition for later”.
• The measure becomes decoration. On the Reporting and Support team's first card, the measure was “customer satisfaction goes up”. A measure you cannot measure means a bet with no measure. It was rewritten as “weekly support calls from 1,150 to 920”.
• Dates leak into Next. In a management presentation in May, someone said “early Q3” for an item in Next. It was me. Two weeks later I read the same words in a sales email.
• Bets never close. At the end of the quarter, every bet needs a verdict: won, lost, or partly won. A bet with no verdict moves quietly into the next quarter, and Now grows again.

What management said

The first reaction was what I expected: “I want dates.” The CEO was right. Without dates, sales cannot plan and nobody can talk about budget. I offered him three things. A range for every bet in Now. For Next, an order and one sentence on why the order is like that. And at the end of every quarter, the verdict on each bet: which one won, which one we stopped and why, and what we gained by stopping.

The third one convinced him. I wrote about how I set expectations upward in the managing up post. I will add only one thing here: what management really wants is not a date. It is no surprises. A range and an early signal reduce surprises. A single date only delays them.

Results: two quarters side by side

The verdicts on the four Q2 bets: US stocks won (live on 14 May; in 60 days, 7% of active customers traded; the measure was 6%). The regular investment plan lost. The support call bet was partly won: from 1,150 to 980, with a target of 920. Account opening was partly won: conversion from application to open account went from 41% to 52%, with a target of 55%. The team pulled the card application flow from Next at the end of April, so the remaining improvements sit at the top of Next for Q3. One win out of four may look low. But in Q1 we had one out of four “finished on time”, and we did not know if it won.

Checklist

Conclusion

On 31 March the question was: “Are the June dates like this too?” Today I have an answer to the same question: “We have four bets for June. Here are the ranges for two of them, and here is the stop condition for one.” This answer is not as comforting as a date. But it is true.

The Gantt chart gave me nine bars, and none of them asked “if this works, what will we see?” That is why price alerts, finished on time, looked like a success. The bet card asks that question before the work starts.

A roadmap is not a list of promises. It is an ordered list of bets. A promise is kept or broken. A bet is won or lost, and if you see early that it is losing, it closes cheaply.

This is the last part of the series. Across twenty-two parts I have said the same thing in different meetings: a plan is not a promise, it is an estimate, and an estimate only helps if you correct it early. A sprint does this every two weeks. A roadmap should do the same thing every quarter.