The literature asserts that support from management is necessary for running any type of BI-project, including advanced analytic projects. A survey of executives by Brown and Gottlieb (2016) showed that ensuring involvement of senior management was the most significant reason for analytics success. In addition, respondents from successful analytic projects reported more often that they had received direct sponsorship from the CEO. Senior leaders and CEOs need to be actively involved in advanced analytic projects (Groves, Kayyali, Knott, & Van Kuiken, 2013). For example, Mazzei et al. (2016) suggest that CEOs need to work on four things in order to keep their analytics project on the right track: i) actively manage C-suite dynamics, ii) chose the right analytics leader, iii) challenge existing mental models, and iv) create an environment of rapid innovation. Involvement of senior management is also mandatory for any organization that has the ambition to establish an analytics enterprise-wide initiative, otherwise analytics initiative are likely to remain in siloes within the enterprise (Watson, 2016). Whereas theory suggests that top managers must be involved in enterprise-wide analytics programs, it is much less explicit about start-up experiments, and seldom mentions the involvement of middle management.

Analytics are typically categorized as descriptive, predictive, and prescriptive (Davenport, 2013; Turban et al., 2015). Descriptive analytics are widespread even in smaller organizations and entail the investigation of what happened in the past through structured internal data held in the company’s databases or migrated to a data warehouse solution built for the purpose. However, some degree of maturity in descriptive analytics, even at the level of data mining, does not guarantee success with advanced analytics. Predictive analytics use a variety of computational techniques to infer from historical data what may happen in the future and prescriptive analytics extend these insights to include decision options and their computed consequences. In addition to the algorithm-driven techniques of machine learning, statistical pattern recognition and natural language processing, advanced analytics may incorporate the V’s of big data (Miele & Shockley, 2013): volume (a large scale of data), velocity (streaming data) and variety (external and non-structured data such as text and video). Thus the competences required for advanced analytics are many and varied. Analytical competence development within an organization takes place in two categories: technical-oriented competence, and business-oriented competence (Davenport & Patil, 2012). Technical-oriented competence development involves building up skills that resemble those of a data scientist. Business-oriented competence involves developing the ability to utilize analytics for business value. A recent success factor for wider adoption of BI and analytics (for instance within Tetra Pak, former winner of the Gartner BI Excellence Award) has been the introduction of a BI expert. A BI expert acts as a bridge between data scientists and IT-specialists on one side, and business specialists on the other (Ohlsson, 2017). A similar role, business translators, is suggested by Henke, Bughin, and Chui (2016). Maturity models (Lahrmann, Marx, Winter, & Wortmann, 2011) theorize and assess the level of maturity, and an organization with better existing maturity at the start of its analytics Jumpstart is assumed to be more likely to succeed with it. LaValle, Hopkins, Lesser, Shockley, and Kruschwitz (2010) introduced a maturity model for analytics with three levels: aspirational, experienced, and transformed. For each level six factors are investigated: motive, functional proficiency, business challenges, key obstacles, data management, and analytics in action. The model was introduced to help organizations to assess their analytics maturity. The maturity model does not explicitly focus on progression of analytical competence within an organization. However, progression can to some degree be deduced from investigating the three factors: motive, business challenges, and key obstacles.

Chen, Chiang, and Storey (2012) list five areas where AA are likely to have a big impact: e-commerce and market intelligence, e-government and politics 2.0, science and technology, smart health and well-being, and security and public safety. In practice the business needs of organizations vary considerably, and an early stage in adopting advanced analytics is simply demonstrating their value on a relevant business or organizational problem. Suitable problems explore data, for instance categorizations, predictions, associations, or prescriptions. The organization tends to deal with these problems in a time-consuming manual manner, or regard them as unsolvable. Since AA are normally data driven (Chen et al., 2012), it’s common to select open problems and allow the analytics to automate the search for hidden patterns or trends in the data. Where patterns or trends appear interesting or unusual the next step is to develop interpretations or models that explain them (He, Zha, & Li, 2013). However Rose and Lennerholt (2017) point out that this style of problem solving makes it hard to focus on a defined problem and to solve it by hypothesis testing. As organizations become more familiar with AA techniques, problem choice becomes simpler, speed to insight improves and use becomes more pervasive (Wixon, Yen, & Relic, 2013).

Working with advanced analytics requires that people use analytical tools to investigate and learn about a business problem, in order to improve decision-making. Thus ‘analytical techniques and controlled experiments are tools for thinking’ (Marchand & Peppard, 2013). However it is the people who do the actual thinking and learning, and who make decisions. There are usually at least four categories of people involved in an advanced analytics project: data scientists, staff from the IT department, business people and management. The challenge is to get these diverse skill sets to work together to investigate a business problem. Managers in particular face many problems with advanced analytics projects: analytics force changes on the C-suite (Mazzei et al., 2016), many business people (including managers) are uncomfortable working with data (Marchand & Peppard, 2013), managers don’t understand or trust big data based models (Barton & Court, 2012).

Researchers offer two conflicting models for getting started with AA. The high impact model suggests the start should be substantial, influential and pivotal for the whole organization’s future: ‘the challenge should be big, the model insightful and the business vision complete’ (LaValle et al., 2011). The low cost model focuses on the opposite strategy: small-scale experiments through targeted prototyping for a quick win. According to (Franks, 2012) companies should ‘define a few relatively simple analytics that won’t take much time or data to run……..capture some of the data in a one-off fashion…. turn analytic professionals loose on the data.’ Gust et al. (2017) describe an internal seed project in a Swiss company with some of the same characteristics as a Jumpstart. As for team configuration, emphasis should be put on a small mixed team that work well in a test and learn environment. Having a small group of people with mixed competences offers several advantages in terms of quality, speed and flexibility. The bigger the group is, the longer it may take to complete the project. In an ideal world, a Jumpstart project should mimic a project in a data-driven organization, where business people and data scientists continuously run experiments together in order to learn more about their business (Watson, 2016). This presents challenges both to conventional analytical processes (ask the IT department to prepare a report) and intuition-based decision-making. A method similar to CRISP-DM (Wirth & Hipp, 2000) is often adopted, though some researchers combine this kind of approach with system development methods such as agile development (Larson & Chang, 2016).

Company A operates transport services for freight and passengers between the major destinations of Europe. Advectas were invited to work with Company A’s marketing and IT departments as a part of a wider vision to become data driven - a company which made decisions on the basis of data analysis. Advectas cooperated with the marketing department and operated as a specialized resource for the IT department, but the company’s business intelligence group was not much involved. Four Advectas consultants participated and at least ten people from Company A, including relevant business managers. The project focused on business questions around the passenger loyalty program (what they buy, how often they travel, what the value of the program is, etc.). The project should develop a test environment and some models that should eventually help marketers to conduct much more sophisticated data analysis than their simple data queries and excel sheets currently enabled. Two workgroups with business managers and IT worked on what some eventual dashboards should look like look like and provided a requirements specification. As the project progressed, building the test environment turned out to involve:

porting data from source systems including the company’s aging customer relationship management systems and their website usage analysis software,

The first three objectives were achieved but there were many difficulties to solve along the way. The lack of a clear business (marketing) objective led to many different demands and foci for the project. “In the beginning, it was a bit loose … exactly what data should we use ... [move the data] up in the cloud and then run Power BI ... and some machine learning” (Consultant 1). Later there were rapidly expanding requirements and ambitious expectations. There were many problems with the data, including quality problems, legal problems, and missing data when it came to modelling important problem variables. PowerBI turned out to have limitations in relation to the more advanced models and analytics. There were organizational problems as well, even with simple things like getting hold of necessary software licenses. The marketing department saw the project as an opportunity to bypass the IT department and get the systems they wanted built, but had little expertise in formulating what these systems should look like. At the same time they did not really trust the data or the results they got. Top management was involved in the launch of an enterprise-wide program for data driven decision making, in which the Jumpstart project served as one of the initial steps. However, the overall vision for data driven decision making had uneven commitment from managers and little buy in from staff. Both IT staff and marketers felt threatened by the different approach and new competences required - “[they] were very worried that the machine-learning part would continuously run live analysis and generate the same insights as they [the business people] did” (consultant 1). Advectas understandably focused on the technical development and it was unclear who was leading the necessary organizational change program - “the technical solution, is still rather straightforward and simple, ... but then the other things swelled up” (Consultant 1). The project did not continue to a production system but had important catalyst effects for Company A. They understood a clear need for better data quality and analysis competence and appointed a data quality specialist, a mathematician and a marketing analyst to fill these gaps.

Company B is a Scandinavian retail chain, with more than 200 shops in Sweden, Norway and Finland. The project had a budget of 100 consultant hours and involved two Advectas data analysts and three IT staff from Company B. The IT department was interested to find out what could be achieved with data analytics, but had no serious history with them. They hypothesized that sales might vary with the weather (“many opinions for or against [a potential pattern]” (Consultant 2)), though the business staff were not much involved in the project. However, Company B had no appropriate tools to investigate the hypothesis further on their own. The project focus was set to investigate correlations between historical sales data with location and publically available weather data. The analysts from Advectas settled eventually on a model using the random forest algorithm, and were able to illustrate several interesting correlations, through two deliveries. The final event was a presentation to which senior business managers were invited, which provoked some inspiring discussion and business suggestions. According to consultant 2, “everybody trusted the result ... there was no scepticism ... it was more constructive discussion”. It is not known whether these suggestions were developed or implemented in the business. Advectas’ competences were demonstrated and Company B intend to pursue analytics, but there was no follow up cooperation at the time of writing.

Company C is a recycling company in Sweden. The Jumpstart arose from earlier contacts and projects in sister companies. Five Advectas staff were involved intermittently, and several managers, two in particular, were involved from Company C. The production managers were more enthusiastic for the project than the operators, who were however happy to provide input, and the IT staff were only superficially involved. Company C, wanted an explicit small scale proof of concept which would show that advanced analytics could make the business more effective by optimizing the smelting furnace operation. The furnace operates a continuous cycle which is expensive to run, and where small optimizations save a lot of money and increase production capacity significantly. However there are many parameters (temperature, gas, oxygen, impurities in the raw material, filters) in running the furnace, and various control systems, making it dependent on operator knowledge, skill and experience. Data is provided by 10-15 different sensors in the furnace + visual input. A lot of data was provided, but there were consistency and quality problems, different operation modes, and some encrypted data. According to consultant 3, “it became a jouney of a thousand miles to arrange all that information… we had to put in lots of extra time [to fix data quality problems] .. several months of delay”. The project was in the model development stage at the time of writing, with good potential to codify and automate the operational experience of the operators.

Company D operates a large fleet of tankers carrying cargoes of petrol, chemicals and liquefied gas all over the world. Since fuel oil is a major cost for Company D, they cooperated in a Jumpstart with Advectas designed to help them predict Worldscale rates 14 days in advance. Worldscale rates are used to calculate freight charges and based on several parameters such as bunker prices, port costs, and market rates. In the beginning, expectations were low - “the [business specialist] who worked with it [manual prediction] was very sceptical that a model would be able to guess prices 14 days in advance” (Consultant 1). However, a rather successful small-scale Jumpstart saw the development of a predictive model (random forest model, modified classification) based on historical data (oil spot prices, exchange rates, etc.). The model offered fair (70%) accuracy and good visualization possibilities (e.g. confusion matrix, scatter plot), which were easy to understand and showed promising results. There was a mixed reception to analytics, some positive, others sceptical. The company was keen to continue and implement a production system in machine learning, but a number of problems appeared. Expectations had been raised too high, the model didn’t perform so well with live production data, the historical data had been somewhat manipulated and there were data quality problems which caused the project to drag out over several months. Close cooperation in the project declined as Advectas personnel changed, and Company D staff felt a little threatened as the algorithm seemed to compete with human competences - “we can compare it [the algorithm] with human guesses, but … [the business specialist] did not want to benchmark against the algorithm”. Meanwhile the senior management had become a bit too enthusiastic: the company made a promotion video claiming that they could predict business conditions (e.g. Worldscale rate, oil prices) fourteen days ahead with 95% confidence. Relations between the two companies were turbulent for a while and the eventual production program offered considerably more modest results. However it is still operating and used as an input for managers’ decision-making, and the company understood the potential value of predictive analytics to their business. Company D later hired their own data scientist and is clearly making a serious attempt to be a data driven company.

Company E is a large retail warehouse in Sweden, selling more than 100,000 individual product lines, and attracting customer visits from all over Scandinavia. The company wanted to predict sales for different divisions during a day, which in turn could help optimizing allocation of personnel. The financial managers were involved in the Jumpstart and supplied their own simple sales figure data, but the IT department had their own technologies and solutions and preferred not to be involved. Two of Advectas’ consultants (with assistance from the sales manager) spent about three weeks developing an analysis approach based on neural networks around problems in logistics and personnel allocation. The techniques produced usable results, but were hard to explain to financial managers more accustomed to working with spreadsheets. According to consultant 4, Company E “did not understand how the algorithms worked, … we could explain for them how time-series worked but not neural networks”. Several people at Company E remained sceptical of results generated by the neural networks, and were not willing to compare them with results produced by existing manual techniques.

Although senior management involvement is often assumed to be a pre-requisite for success, there was a varied picture in the project we studied. Company A and Company D had some direct involvement of the top management, where they also had early strategic commitments to more extensive programs. Other projects had little or no top management involvement, for instance the senior management of Company B were invited to the wrap-up event, but were not involved the Jumpstart. Nor was the involvement of top managers always beneficial: the Jumpstart for Company D was a technical success, but the senior management inappropriately raised expectations, contributing to the project running into difficulties. Though some companies showed early indications of a strategic move to enterprise-wide analytics or a data driven culture, this seemed generally beyond the scope of a Jumpstart. However, three out five Jumpstarts had middle managers with relevant expertise actively and beneficially involved: Company A (business managers), Company C (production managers), and Company E (financial managers).

Most of the client companies shared the perception that big data/advanced analytics are here to stay that they need to get started. However the starting competence level for analytics was low; ranging from the use of spreadsheets, simple data queries, and manual prediction (Company A, Company C, Company D, Company E), to no tools at all (Company B). There was little experience with mathematical modelling, machine learning algorithms or decision making based on statistical probability. An additional task for the projects was therefore building up analytics competence amongst the project members. At the kick-off events, Advectas focused on the potential of advanced analytics, in order to bring the project members on board. However there was seldom a baseline for existing analytical understandings (what do we know and how do we know it?), or very clear specifications for what was being analysed and why. The learning curve could be steep - sometimes too steep. Company E (more used to simple formulas in spreadsheet analysis) was offered a predictive model featuring neural networks - however this was effectively a black box that staff weren’t ready to trust. A major obstacle was often the existing information infrastructure - good enough for producing financial reports but not really designed for extracting and integrating data outputs from multiple sources.

A clear, limited and well-focused business problem increased the chances of success. Most of the companies had such a focused business problem requiring predictive analytics as a starting point. Company A worked on a business problem that was less precise and coupled to other marketing goals and the company’s wider analytics ambitions. This encouraged project scope creep and unclear goals, and the involvement of multiple actors (for instance at workshops). It was not so clear that the business problem was solved, or that the projects led to revolutionary business advantage. The analytical models and results provided business insights not previously available, better information for decision-makers and illustrated the potential value of advanced analytics for a real business problem. There was often additional learning about the information and IT infrastructure: data quality, integration, systems, storage, platforms, algorithms etc. The Jumpstart provided a small-scale example of a new way of working where data analytics provide the foundation for decision-making, and emphasized the importance of the information infrastructure that supplies the data.

It was noticeable at some companies that advanced analytics were challenging. The combination of lack of previous experience and a low level of trust in data, IT systems, analytics and consultants may lead to scepticism, lack of confidence, and feelings of being challenged or threatened. Three companies (A, D, E) out of five, encountered some negative reactions to advanced analytics. These companies also experienced issues related to lack of trust in the project’s results. Company B and C had no obvious negative reactions. A potential explanation is that the Jumpstart in Company B and C, did not, in the short term, challenge the existing workforce. In addition, neither Company B or C had any issues with exaggerated expectations on advanced analytics.

The Jumpstarts generally employed the small-scale experiment model, with (not always successful) attempts at time boxing and a core project team. Several projects (A, C, D) experienced serious problems with data quality, data integration, missing data, downloading, cleaning and storing data which were time consuming and costly to fix. Prolonged data preparation also delayed modelling and analytics activities. An advanced analytics experiment is likely to use existing data in new ways, or integrate it in unfamiliar patterns, or incorporate external data. These problems compound widespread inadequacies with many companies’ legacy data infrastructure. The other companies (B, E) had cleaner and simpler data available. Most Jumpstarts used a small and stable core team - typically 2-3 people from Advectas + 2-3 people from the client organization. Company D experienced several changes in the core team (partly because the data scientist skillset was in short supply and was need elsewhere) - this had a negative effect on the project. The Jumpstart at Company A evolved beyond the small-scale experiment - a high level of ambition, expanding requirements, multiple business problems and many people - but this was difficult to handle. Though Advectas’ process model (figure 1) distinguishes between a pilot implementation and a full-scale deployment, this was not a prominent feature the projects we studied, and was not emphasised by Advectas either.

Most of the projects had the following things in common. They adopted the small-scale experiment model rather than the high impact model. In principal, they were designed with a limited budget, constrained project duration, and few core project members. In this respect they were more proof of concept (show us that we can get something useful from analytics), than a revolutionary business advance. They employed advanced analytics involving some form of machine learning or statistical analysis that was well beyond the experience of the company. Middle managers with relevant business and/or IT experience were active and involved. The client company often saw the projects as an investment in competence building and they were normally satisfied when they learned more about their business, or sometimes if they learned more about their IT systems, the data they had available and new analysis possibilities. A clear business problem or hypothesis increased the chances of success. Though a pilot was always developed, a full-scale deployment was more likely to be understood as a follow-up project. The analysis is summarised in table 1.

Table 1

	Company A	Company B	Company C	Company D	Company E
Management involvement	some top management involvement, business (middle) managers involved		production (middle) managers involved	some direct involvement of top management	financial (middle) managers involved
Analytics Experience	spreadsheets, simple data queries	no tools	spreadsheets, simple data queries	spreadsheets, simple data queries	spreadsheets, simple data queries
The Business Problem	less precise business problem	focused business problem, predictive analytics	focused business problem, predictive analytics	focused business problem, predictive analytics	focused business problem, predictive analytics
People and Analytics	some negative reactions	no obvious negative reactions	no obvious negative reactions	some negative reactions	encountered some negative reactions
The Jumpstart Project:	serious data quality problems, many people involved, deployment follow-up	small and stable core team, small-scale experiment	serious data quality problems, small and stable core team, small-scale experiment	serious data quality problems, several changes in the core team, small-scale experiment	small and stable core team, small-scale experiment

As a result of this investigation we understand an AA Jumpstart to be an introductory AA project in a client company inexperienced in analytics, led by a specialist consultancy. The project is designed as a small-scale experiment, with low costs and low risks for the client and with significant interaction between consultants and company staff. Jumpstarts are focused at delivering business value, but also offer various knowledge transfer and catalyst outcomes. Above all they should stimulate further interest in the client company, inspiring new initiatives and functioning as proof of concept - the client learns enough about the value of AA to wish to continue. The consultancy hopes to develop a relationship with the client and win further business.

Figure 3 offers a modified and amplified version of Advectas’ Jumpstart process (figure 1) that is in harmony with what we observed in the case examples. Major phases are specified together with deliverables, such that it could serve as a project management tool for a consultancy managing a Jumpstart.

Figure 3

This section takes the form of normative advice for consultants running Jumpstarts derived from the theoretical and empirical analysis.

Management Involvement

Though management should be involved at various levels, the role of middle management in a Jumpstart is more critical to its success.

Ensure that top management are supportive (they’re not necessarily directly involved). Top management may need to authorize resources for the Jumpstart and show an interest, but the scale of the projects are usually too small for them to be directly involved.

Analytics Experience

Many Jumpstart organizations have a low level of analytical sophistication, which has consequences for the ambition-level and conduct of the Jumpstart.

Understand the company’s analytics maturity and focus on explanation and education (particularly at kick-off and wrap-up events). Low levels of analytics experience mean that most Jumpstarts must focus on knowledge transfer and competence development in order to succeed. An early Jumpstart task is to assess the starting competence level at the client company, and provide the necessary explanation and education at kick-off to bring project members on board. The wrap-up event should focus on lessons learnt, and highlight advances in analytics understanding.

Business Problem

A focused business problem utilizing predictive analytics was advantageous.

Choose a limited business problem focusing on predictive analytics. Descriptive analytics typically require a data warehouse solution (beyond the scope of a Jumpstart), and prescriptive analytics assumes that predictive insights are already available and ready to be automated. A limited business problem involving predictive analytics is the preferred choice.

People Factors

Jumpstarts need to deal with negative reactions as well as enthusiastic ones.

Address expectations and anxiety. The people directly or indirectly involved in a Jumpstart may be inexperienced with advanced analytics. This can lead to two problems: exaggerated expectations or negative reactions. Negative reactions may include scepticism, distrust, and feeling challenged or threatened. Some may even feel that their jobs are threatened. Jumpstart teams should recognize and, as far as possible, address these concerns. Most advanced analytics tools provide better (but not infallible) support for decision makers - they do not replace the decision maker.

Jumpstart Project

The four previous enablers (management involvement, analytics experience, business problem, and people factors) influence, but do not determine the conduct of the Jumpstart project.

Design the project as a small-scale experiment, with a constrained business problem, a small (carefully chosen) core project team, and time boxed execution. Projects are more manageable (and therefore effective) when organized in this way. The experimental form allows the client organization to assess the potential value of analytics; the constrained business problem ensures that the project objectives cannot multiply out of control. A core project team might include from the consultancy side: project manager, data scientist, data engineer; from the client side: business specialist (problem owner) with in-depth knowledge of the problem and data, IT person with knowledge of IT feeder systems, manager with the authority to problem solve and organize resources. The project is more effective if the team remains intact and this is easier to achieve if there is time boxing (1-2 months), which also helps to avoid creep in the project scope, and ensures that there are timely results.