From spreadsheet to AI agent
The technology perspective: from the paper table to artificial intelligence (AI) agents – who plans, how, at what speed and with what precision.
Rolandas Rakštys · Lecturer · Vytautas Magnus University, Agriculture Academy · Faculty of Bioeconomy Development, Department of Business and Rural Development · [email protected] · ORCID 0000-0003-0299-9930
Chapter 2 “The business planning and evaluation process”, section 2.1.4 · English translation of v1.0.11 (2026-10-15) · set v1.0.3 · 12 slides, ≈ 24 min
Rakštys, R. (2026). Chapter 2. The business planning and evaluation process (English translation of v1.0.11, 2026-10-15). Verslo planavimas ir vertinimas [Business planning and evaluation]. Vytautas Magnus University, Agriculture Academy. https://rakstys.eu/studijos/verslo-planavimas-ir-vertinimas/en/skyriai/ch02/ · CC BY-NC-ND 4.0
One assumption – how much time?
The “financial part” of a business plan – a table drawn by hand on paper.
Each technological wave changed not only the tools but the very paradigm of planning – who plans, how, at what speed and with what precision.
Changing one assumption meant recalculating the whole model – for hours or days.
So in practice one – “base” – scenario was built, and sensitivity analysis was a theoretical notion rather than a working tool.
“What if?”
The spreadsheet made interactive modelling possible for the first time: change an assumption and see the effect on the whole model instantly. A change not only of speed but of the paradigm of thinking.
In the words of VisiCalc’s creator Daniel Bricklin, “what if?” analysis became available to all, not just to big corporations with staffs of clerks and MBAs.Bricklin, 2009, p. 21
Planning turned into the continuous testing of assumptions rather than the one-off drawing up of a table.
A single database
Enterprise resource planning
(ERP) – from material requirements planning (MRP) systems, which spread in the 1970s, to SAP R/3 (1992): production, finance and sales in a single process.From the planning book to the team
First – the availability of information
Market data, competitors’ financial statements, macroeconomic forecasts – for small firms and start-ups too. The quality of planning was democratised, but the problem of information overload appeared.Second – speed and iteration
Cloud computing allowed teams to work on the same plan in real time – for example, Google Sheets.Planning from data
More data – not better logic
The problem of spurious precision (see 2.1.3) became sharper: when a model is fed large volumes of data, its results look more “scientific”.
The fundamental assumptions may be just as inaccurate as in a manual analysis.
Large volumes of data do not compensate for poor conceptual logic.
Where AI has the edge
In the M5 competition, machine learning (ML) models significantly outperformed the best classical methods: the advantage of the top five teams over the benchmark method exceeded 20 % (Makridakis et al., 2022).
Time series tasks: demand forecasting, cash flow planning and similar areas.
Regular time series: automated algorithms achieve greater accuracy than traditional linear models.
Where no validated statistical model exists – structural breaks, geopolitical wars, unexpected amendments to legislation – the most robust is a combination of human and algorithm, not full autonomy for either side; beyond the forecasting horizon all that remains is to plan for adaptation and resilience (Tetlock & Gardner, 2015).
From builder to curator
What AI amplifies
Democratisation: small firms that could not afford a team of analysts obtain better forecasts and scenario analyses – relevant to Lithuania, where SMEs dominate.
The digitalisation of the NMA and other support bodies changed the architecture of expectations: a plan must increasingly rest on structured data, not on a textual narrative alone.
Several decimal places – a stronger illusion of being “scientific” than in a simple Excel model, although both rest on the same uncertain assumptions.
Optimism bias
May be reinforced if AI is trained on historical data dominated by successful cases (survivorship bias).Spurious precision
The risk increases further (see 2.1.3).The problem of accountability
When a plan is produced with the help of an AI assistant, who is responsible for the quality of the assumptions – the human or the algorithm?One desk – three shifts
AI is a powerful instrument of planning, not a subject of planning.Ultimate responsibility for the assumptions, the choice of scenarios and taking decisions remains with the human.
A deeper analysis of the relationship between technology and business planning – in the final chapter of the textbook.
References
Bricklin, D. (2009). Bricklin on Technology. Wiley.
Gartner. (2020). 2020 strategic roadmap for cloud financial planning and analysis solutions. Gartner, Inc.
Makridakis, S., Spiliotis, E., & Assimakopoulos, V. (2022). M5 accuracy competition: Results, findings, and conclusions. International Journal of Forecasting, 38(4), 1346–1364. https://doi.org/10.1016/j.ijforecast.2021.11.013
Ries, E. (2011). The Lean Startup: How Today's Entrepreneurs Use Continuous Innovation to Create Radically Successful Businesses. Crown Business.
Tetlock, P. E., & Gardner, D. (2015). Superforecasting: The Art and Science of Prediction. Crown.
Rakštys, R. (2026). Chapter 2. The business planning and evaluation process (English translation of v1.0.11, 2026-10-15). Verslo planavimas ir vertinimas [Business planning and evaluation]. Vytautas Magnus University, Agriculture Academy. https://rakstys.eu/studijos/verslo-planavimas-ir-vertinimas/en/skyriai/ch02/ · CC BY-NC-ND 4.0
These slides are based on the textbook Verslo planavimas ir vertinimas [Business Planning and Evaluation]. The text was translated, structured and visually laid out with the help of artificial intelligence tools (Claude), following the protocol for working with AI and the verification procedures set by the author.