Forecast what features will make a release deadline and experiment with scope and start order. Do this early in the planning process to make better decisions about feature scope and start order by showing what will make a target delivery date with a simple tick (yes) or cross (no) for a set of desired features. Often referred to as the cut-line, this spreadsheet shows what will make the cut and what won’t. Often seeing what won’t make a release helps everyone look harder at whether the order matches business need.

Although we have kept this spreadsheet simple on the input side, it performs a Monte Carlo simulation for each feature (1,000 hypothetical deliveries) just like the Single Feature Forecaster spreadsheet. This spreadsheet uses the completion date for each feature as the start date for the next. It then shows which features will be delivered on or before a target date.

Download this spreadsheet from here:  Multiple Feature Cut Line Forecaster: I want to forecast multiple features at once

Multiple feature forecast inputs and results. Tick = yes, Cross = No hope

Multiple feature forecast inputs and results. Tick = yes, Cross = No hope

 

The Forecast Inputs

1. Start date: The calendar date where work will actually begin for the first feature. This shouldn’t be the “planned” start date, it should be the date when you are sure work will start. Look at other project that might be delayed or other reasons key staff might be unavailable to start when deciding what this date should be. We find the start date is often overlooked as an easy estimate to make, its not. Its one of the few that involve thinking about company wide factors rather than just the project or feature work and the people in the room

2. Target Date – This is the calendar date that decides whether to show a tick or cross. If the likelihood of a feature delivery date (set by 3. Likelihood) is on or before this date, that feature is said to make the deliverable and adorned with a green tick-mark. Otherwise it will be given a red-cross mark. This should be the date that code is tested, packaged and ready to enter any final deployment pipeline. Often this date is before the product is released to market, its a hand-off date to operations.

4. Split Rate – We find that a big cause of forecast accuracy is using a delivery rate (throughput or velocity) based on work that has been split into multiple parts when started. It makes the team appear to be delivering faster than it is. Our most commonly seen settings (and suggested if no actual data exists) is a low estimate of 1 (not split) and a high estimate of 3 (a backlog item is split into three parts for delivery). Check if the throughput and velocity estimate or data includes defects. If it does, bump these estimates up to model the most common defect rate rate range so each item embeds its allocation of defects. For example, if every piece of work get one defect, start with a low estimate of 2 (1 for the work + 1 for a defect).

5. Throughput Estimate or Data – The throughput estimate is where you express the delivery pace of the team. This is the rate that  work is delivered in the same units estimated for each feature. Velocity in points or throughput (count of items completed over time) can be used, the only requirement is that the same units be used for item estimates. You also have the choice here of whether to use a low and high bound guess, or to use historical data (where samples are placed in the Throughput worksheet. Our advice is to start with a wide estimate (spread out low and high value). What is important is the actual values of throughput and velocity found out through delivery fall within this estimate, not to have a narrow estimate. When real data starts to flow in from the team, switch over to using it instead of the guess. Normally about 11 samples are needed before i’m comfortable changing to use data. Try both, and see when the forecasts begin to agree, then use the data samples.

7. The Features and Estimates – This is the part of the spreadsheet where you describe the feature (by name) and get a low and high guess of the amount of work for each feature. Similar guidance here to entering the throughput estimate, start wide and narrow as real data and lessons allows. Remember to consider the split rate in input 4. Don’t double count splitting or defect inputs. Choose either to account for scope creep and defects in these estimates or through the splitting estimates – be consistent and explicit! The first column, start order is for convenience. Sometimes when some features are going to miss a release date, experimenting by changing the start order gets the “agreed” features the best chance of making the cut. Its provided here so that immediate results are shown during discussions and meetings (we used to cut and paste rows, but it was slow and error prone, so we added the ability to control order by column A).

8. Month Adjustments – We found some major holiday events, seasons, or company conferences made significant differences to throughput rate. These inputs allow the throughput or velocity estimates to be adjusted depending on which month work falls. The spreadsheet multiplies its random pick for delivery pace by the multiplier you enter for each month if work falls in that month. Its fine tuning mostly. In Europe we find summer pace slows. In the US we find less fluctuation. One key insight we see if a companies vacation policy matters. For companies that have a “use it or lose it” policy for vacation time, the last month (often December) is a fraction of other months for throughput. Ask and look for reasons one month might fluctuate. Re-orgs, team hire and ramp up, conventions and conferences are some of our common reasons for using these adjustments.

The Forecast Results

Screenshot 2016-03-30 11.24.35

The Forecast results are shown to the right of each feature input. Its designed to be simple enough your manager can understand it. We show the start date because its important that any date we show in the table assumes this is the start date; miss that date, then the forecast is incorrect.

Given the start order in column A for each feature, a green tick means on or before the Target date (given the likelihood percentage confidence entered as input 3, 85% in this case). A red cross means it will miss by at least a week (or 2 weeks depending on the time interval entered in the throughput input as part of input 5.). An orange exclamation mark means within one throughput period (on the fence). We encourage our stakeholders to look at what is going to miss and offer to trade something that is going to be delivered. This experimentation is easy by changing the start order in the column A adjacent to each feature. The start orders must be non-duplicated, ascending order from 1 to 10 (errors are shown if you get this wrong).

Conclusion

This tools aims to help early and vigorous discussion about scope and start order of multiple features competing for the same delivery team. Use it to help others see likely outcomes and make hard decisions earlier. Although we have more complex forecasting tools, this is often the right tool for early exploratory discussions about what resources are needed and what set of features will make the cut given teams historical pace of delivery.

Download this spreadsheet from here:  Multiple Feature Cut Line Forecaster: I want to forecast multiple features at once

Love to hear feedback on the type of discussions it causes and suggestions for improvements: troy.magennis@focusedobjective.com

Troy