Professional Documents
Culture Documents
2
01 Forecasting in a digital world 4 Algorithmic forecasting in action 14
06
05 Roadblocks 13
3
Forecasting in a digital world
01
02
Human fascination with the future is part of our evolutionary heritage. Those
03
who can foresee and navigate risks have always been more likely to survive than
those who can’t. That’s just as true in business, where the ability to see ahead
04
continues to separate performance leaders from everyone else. That’s what
forecasting is all about, yet it’s surprisingly difficult (and often expensive) to do. 05
Companies have different motivations for improving their Traditionally, forecasting has been a mostly manual process
predictions. For some, the ability to deliver reliable guidance with people gathering, compiling, and manipulating data, 06
to analysts and markets might drive the decision. For others, often within spreadsheets. With more and more data
being able to trim production waste by predicting consumer available, old-school forecasting has become an unwieldy, 07
demand is what matters most. Still others want to improve time-consuming process that makes discerning what’s
cost management and streamline the forecasting process. important next to impossible. As a result, humans often
resort to their own intuition and judgement, which opens the 08
Sandbagging – the tendency of deliberately door to unconscious biases and conscious sandbagging.
underestimating performance predictions in the hope 09
of exceeding the targets in the future.
Deloitte Analytics
4
There’s another way. Organisations are shifting to forecasting Today, these technologies in the hands of expert forecasting
processes that involve people working symbiotically with talent give companies the ability to discover things they’ve 01
data-fuelled, predictive algorithms. It’s all made possible by always wanted to know − as well as things they didn’t know
new technologies − advanced analytics platforms, in-memory they didn’t know − with more confidence and speed. CFOs
02
computing, and artificial intelligence (AI) tools, including are in a prime position to challenge the way their enterprise
machine learning. These techniques and algorithms enable looks at and consumes data. In the area of forecasting, they
businesses to use more data (internal and external) than was can champion an innovative, data-driven approach that will 03
historically possible. help people project the future of their business. By modelling
the potential impact of important decisions, they can help
04
Similar digital tools are common in our consumer lives. generate smarter insights and stronger business outcomes.
We use a mapping app to predict when we’ll arrive at our
destination. A real-time weather app can tell us precisely when Observations in Africa – in Africa, we have seen larger 05
a rain shower will start or stop. So, it’s reasonable to expect adoption of Machine Learning algorithms in the customer
predictive capabilities when we’re at work. related sectors of business rather than in finance. The
transition of these technologies into the finance area is the 06
For an introduction to the digital capabilities that make data-powered forecasting natural next step and CFOs need to be ready for large scale
possible, read Crunch time 1: Finance in a digital world. adoption in the near future. 07
08
09
5
Algorithmic forecasting
01
02
01
02
03
Human AI applications Modern Data sources Advanced
intelligence computing analytics
capabilities platforms 04
05
06
07
08
09
7
Don’t we have bigger fish to fry?
01
02
CFOs are continually weighing the relative ROI of potential investments that vie for
03
their attention. Here are a few reasons why many leaders are moving algorithmic
forecasting up their priority list:
04
09
8
The ripple effect
01
02
While some aspects of forecasting using algorithmic models are fairly
straightforward, others can be trickier. Changing processes. Building trust and 03
transparency. Creating partnerships between people and machines. These are
tougher challenges. 04
02
Many of the companies we work with have begun their digital finance journeys by investing in cloud, in-memory computing,
and robotic process automation. Others have broadened their ambitions to include advanced analytics, with an emphasis on
03
forecasting. They want to create forecasts that enable faster and more confident decision-making. This is where those digital
investments can begin to pay off. Traditional approaches to forecasting can take far too long, cost far too much, and generate too
little insight about potential future outcomes. 04
06
Top-down planning Bottom-up forecasting Function-specific External reporting
Target setting Product-level forecasting forecasting Market guidance
Integrated financial statement Market- or country-level forecasting Customer retention Earnings estimates
07
forecasting
Direct cash flow forecasting Inventory optimisation
Working capital forecasting
Employee retention and
08
Indirect cash flow forecasting attrition modeling
Demand forecasting
02
The FP&A team for a global consumer product What happened next Looking ahead
manufacturer frequently outperformed their Finance leadership asked Deloitte to help them Leadership sees this as a game changer, and not just
guidance to market analysts. Problem was, they develop an objective, data-driven forecasting for their top-down financial forecasts. The business 03
couldn’t explain the unanticipated growth, and approach. Within 12 weeks, Deloitte’s data scientists segments do, too. Following the successful delivery
holding credibility with the executive team, board, designed a top-down predictive model that of a prototype focused on corporate FP&A, Finance
and industry analysts was a key priority. incorporated the company’s internal historical actuals has added data scientist capabilities to its talent
and external drivers for each global market, including model. Socialising the results with the business has
04
The team suspected sandbagging was the source housing stats, local GDP metrics, commodity prices, created significant demand to dive deeper and extend
of their headache. Individual business unit (BU) and many additional variables. the solution to the business segments and regions.
leaders made their own bottom-up forecasts as
part of the target-setting planning process, which The model enabled the FP&A team to deliver a
Additionally, the client has taken steps to industrialise
the model and to provide business users greater
05
was used for performance incentives. Finance had second-source forecast based on external macro visibility into the driver assumptions and relationships
no objective way to verify or push back on the BU drivers that aligned to market expectations and to the financials.
leaders’ numbers. provided insightful, accurate projections across 06
the P&L, balance sheet, and cash flow statements. The company continues the journey to scale the
Planners also gained the ability to quickly create solution within the organisation on their own.
growth, recession, and other scenarios using desktop The client’s FP&A leader reports, “This is a good-
visualisation software. news story. We made an investment [to prototype 07
algorithmic forecasting], proved the concept, and got
The toolkit the business interest. We created significant demand
The Finance team received the fully functional for a new Finance-developed forecast capability that
predictive model built on an open source platform, will help the organisation improve forecast accuracy 08
which enables their CoE to manage and model and efficiency.”
forecasts on an ongoing basis. Leadership gained
an objective, transparent, and visual conversation
starter for discussions with business units about 09
new opportunities and upcoming challenges for
their markets.
12
Roadblocks
01
02
When you start having Fear and loathing Small thinking Engagement matters Silver bullets
conversations with colleagues People are often scared of the Some people won’t take time to see It’s important to involve end users Algorithmic forecasting often gets
about replacing traditional unknown. Even though Finance is the bigger picture. They’ll go along to help conceive, design, build, and introduced with a lot of hype. Don’t 03
forecasting processes with a discipline grounded in numbers, with algorithmic forecasting if they implement algorithms. After all, they do that. It’s just a tool. That said,
algorithms, you’ll quickly discover some will resist the idea that have to, but they’ll start small and know the sandtraps that need to be when this kind of powerful tool is
that people have preconceptions algorithmic forecasting might stay small, looking for what’s wrong avoided, and their acceptance is key combined with human intellect, the
about what it is, how it works, and enhance their own methods. Help instead of what might be workable − to effective implementation. impact can be transformative. 04
what it means for the organisation. them see that machines can handle all while refusing to give up the old
That’s what happens when change the tedious number-crunching work ways of thinking and acting.
is in the air. while enabling people to spend more
time uncovering valuable insights. 05
06
Black boxes Turf battles Data, data, everywhere
There’s a myth that algorithms are Is algorithmic forecasting a CIO Some companies want to try new
black boxes where magic things thing that needs to be brought into approaches to forecasting, but
happen. While not true, some Finance, or is it a CFO thing that they’re worried their data problems
07
complicated models may seem needs IT’s help? And when it comes will stymie them. Whether those
that way. To help build trust across to bottom-up forecasting, who’s on problems are the result of mergers
the organisation, consider using first − the business or the business and acquisitions, a history of poor
08
easier-to-understand algorithms partner? The reality is that effective data management, disconnected
with transparent inputs in the algorithms require data, computing systems, or all of the above, you’re not
beginning. As comfort grows, so power, and business insight − so alone. The first step toward improving
will the willingness to accept more collaboration is important. Making forecasting for many companies is 09
complex approaches. accountability and ownership getting their data house in order −
decisions up front can improve sometimes even one room at a time.
collaboration and value generation. 13
Algorithmic forecasting in action
01
02
It’s 7:00 a.m., and you’re thinking about the day ahead. Today, your forecasting function is a well-oiled machine, with
By noon, you have to settle on a forecast about how your more than 80 percent of the work happening automatically. 03
business will perform over the next quarter. And at 2:00 p.m., Every piece of financial data you could want is available on
you need to tell that story to a dozen board members on a your tablet. All you have to do is ask − literally. Display the
04
conference call. impact on profits if the cost of steel goes up 20 percent in the
next month. You can drill down, roll up, set aside exceptions,
In the past, your forecasting team would be pulling all-nighters and run a dozen more scenarios before your conference call. 05
for days before your meeting. They’d be grinding through And you can do it all without an army of analysts scrambling
spreadsheets, calculating growth percentages, chasing down to help.
anomalies, and drinking way too much coffee. That was then. 06
07
08
09
14
What changed?
Almost everything.
01
02
03
15
Getting from here to there
Most clients we work with don’t attempt a wholesale change to their forecasting approach from the beginning. Instead, they select a part of their business
or a specific revenue, product, or cost element to use as a pilot or proof of concept. They often run algorithmic forecasting parallel to their human- 01
centric forecast for a period to compare accuracy and effort. Every company will make its own unique journey from its current approach to planning and
forecasting to an improved approach. That said, there are some things you’ll want to consider on your path forward.
02
Prototype
1 2 3 4 and scale 03
A B C D E
Driver analysis 04
Identify the Think about Identify the and data Predictive Dashboards and
Socialisation Ongoing
problem to solve how to proceed help you need cleansing modeling visualisations
•• Define scope •• Decide if you want to •• Assess available talent •• Identify key •• Develop statistical •• Develop dashboard •• Socialise results with •• Align on cycles 05
and ambition build this competency and their abilities revenue and models for P&L views with key metrics key stakeholders all required for refreshing
in-house − or if cost drivers line items based on along the way your models
•• Determine the level of •• Determine if additional •• Develop accompanying
outsourcing to a relevant drivers
•• Collect and •• Prepare detailed •• Assess opportunities
business to work with managed analytics professionals are
needed − financial structure relevant •• Consider using
visualisations
documentation for for machine learning 06
•• Identify targets service makes •• Enable scenario
analysts, data data for analysis models that are easier maintaining and and cognitive
(geographies, more sense analysis functionality
scientists, data for the business to managing the model enhancements
visualisation architects, •• Align on an initial
products, customers, •• Determine if understand, to build •• Elicit feedback from
channels, etc.) algorithmic forecasting or others − and how set of priority
drivers
trust and adoption dashboard users
•• Track results in
parallel − keep score 07
•• Set a realistic is a capability you want you’ll get them
•• Include end users •• Use A/B testing
time horizon to provide as a service •• Collect and •• Train the organisation
•• Determine external in the process of to optimise the
to the enterprise clean required to understand
vendor support needs conceiving, designing, effectiveness of
•• Make sure your
•• Identify which tools you
data inputs building, validating, dashboard displays
the value of this
collaborative approach 08
organisation has the •• Test drivers for and implementing
already have in place
talent and culture to significance forecasting models
embrace this •• Ask IT what other tools •• Test and validate
may be needed the models 09
•• Link P&L forecasts to
the balance sheet and
cash flow 16
Before you go
01
02
A commitment to algorithmic forecasting is both a cultural thing and a statistical thing. Making it happen involves great
people working with elegant technology. Neither is sufficient on its own. Here are some of the lessons we’ve learned while 03
helping companies move forward.
04
People lessons Statistical lessons
•• CFOs should be the ringleaders and champions. Reading at least one book on the •• For mature and stable businesses, predictive models usually require five or more
topic can help ground you in the culture of forecasting. years of monthly data to train properly, with the most effective models often using
10 years of data to identify trends, seasonality, and driver correlation. 05
•• Train your team − and yourself − on the basics of probabilistic thinking and how to
spot and correct for common human biases that can derail effective forecasting. •• Don’t overlook the possibility that past performance may not be relevant to
future performance predictions. Often the data can be used, but may need to be
•• Management and employee engagement becomes easier when models are
transparent. Choosing a less accurate model with more intuitive drivers may
adjusted or filtered by human experts. Experiment with different types of modelling 06
approaches to see which leading and lagging indicators improve the relevancy of
result in better business adoption than a more sophisticated model that people
your results, especially in a changing or disruptive industry environment.
can’t understand.
•• Build an approach where accountability for “the number” remains local − in the
•• Aggregation across business units, divisions, groups, or the whole company
07
generally improves predictability.
hands of those closest to the business.
•• Products with higher sales frequencies yield stronger predictability due to the
•• Assess your workforce needs with the understanding that changing from traditional
greater number of data points. Long-cycle products can require additional data
to algorithmic forecasting may create spare capacity as routine tasks are automated
history for the models to mature., although there are models to account for and 08
and skill gaps where new professionals are needed.
counteract the issue.
•• Don’t underestimate the value of visualisation. It makes forecasting real.
•• Annual forecasts tend to be more accurate than quarterly. Same is true that
quarterly predictions tend to be more accurate than monthly. That’s because 09
variances tend to cancel each other out − up to a point. Over longer time horisons,
predictions can become less accurate as unexpected events become more likely
17
Want to learn more?
Crunch time 6 | Forecasting in a digital world
Start here.
01
02
02
memory.”
05
06
—Marquis of Halifax
English writer and statesman 07
08
09
19
Looking ahead
01
02
07
08
09
20
Acknowledgements
01
02
Authors Contributors
03
Eric Merrill Paul Thomson Ayan Bhattacharya Niklas Bergentoft
Managing Director, Predictive United States United States Advisory
Analytics and PrecisionView™ 04
Offering Leader JoAnna Scullin Dave Kuder Srini Raghunathan
Deloitte Consulting LLP United States United States United Kingdom
Tel: +1 404 631 2141
Email: emerrill@deloitte.com James Guszcza Tim Gross Tim Leung
05
United States United States United Kingdom
Steven Ehrenhalt
Principal, US and Global Finance Max Troitsky Tim Gaus Anna Chroni 06
Transformation Leader United States United States United Kingdom
Deloitte Consulting LLP
Tel: +1 212 618 4200 Jeff Schloemer Laks Pernenkil Alie van Davelaar 07
Email: hehrenhalt@deloitte.com United States United States Netherlands
Deloitte is a leading global provider of audit and assurance, consulting, financial advisory, risk advisory,
tax and related services. Our network of member firms in more than 150 countries and territories serves
four out of five Fortune Global 500® companies. Learn how Deloitte’s approximately 264 000 people
make an impact that matters at www.deloitte.com.
This communication contains general information only, and none of Deloitte Touche Tohmatsu Limited,
its member firms or their related entities (collectively, the “Deloitte network”) is, by means of this
communication, rendering professional advice or services. Before making any decision or taking any
action that may affect your finances or your business, you should consult a qualified professional adviser.
No entity in the Deloitte network shall be responsible for any loss whatsoever sustained by any person
who relies on this communication.