Investments in advanced automation including Gen AI are the top priority for enterprises
Gen Ai proudly sits at the Peak of Inflated Expectations in the Gartner hype curve for Emerging Technologies, 2023, projected to reach transformational benefit within two to five years.
Following its release in November 2022, ChatGPT reached over 1 million users in five days and 100 million users within two months. This made it the fastest-growing consumer app ever.
Market size for Artificial Intelligence is projected to reach $184.00bn in 2024. Market size is expected to show an annual growth rate (CAGR 2024-2030) of 28.46%, resulting in a market volume of $827bn by 2030.
Two-thirds of jobs could be partially automated by AI. But many of these jobs will be complemented by AI, not substituted by it. (Goldman Sachs, 2023)
So, wherever you turn, AI is taking a seat at the top table with ‘C’ level executives. We are seeing a spectrum of approaches ranging from cautious à to the more pioneering that are experimenting with AI.
There is no right or wrong approach. With regulations being introduced, threat of fines, damage to brand and poor customer sat – many are taking time to align to overall strategy, enhancing their operating model and undertaking risk assessments. Others have several AI applications aligned to specific use cases and are testing rapidly
We can safely say that it’s a top priority for most organisations across all sectors.
While enterprises are confident in their ability to manage data the big obstacle for gen AI adoption seems to be the lack of quality data for training.
History is often a good predictor for the future. If you look at organisations attempts to exploit Bus Intelligence and Knowledge management over the last decade – many have failed due to lack of available data, poor quality data, poor data governance and poor change management.
AI algorithm training requires historical data to test and validate hypotheses, this then needs to be supplemented with ongoing live data for continuous learning. Low volumes of data limiting AI learning potential leading to sub-optimal results.
These bring with it a series of challenges for every organisation:
- Low data management maturity across the entire enterprise
- Lack of sufficient data management capabilities to enable AI solutions
- Missing architecture that defines the AI data needs (including volumes, velocity, variety, and accuracy)
- Low quality data and lack of data integration which slows down AI deployments
- AI solutions will create technical debt over time, as well as new challenges and risks that need to be addressed.
What are the main challenges to getting started on the AI journey?
All organisations want to understand why they are investing and what benefits it will deliver.
Here we have the main benefit drivers that that would be assessing your use cases against:
- supporting growth – new sales, upselling, and cross selling
- reduce costs – removing bottlenecks, process inefficiencies, maximising automation opportunities, staff time for high value tasks
- enhancing the experiences of employees and customers
- Risk and compliance controls – introduction of new regulations.
There are many data concerns around:
- bias – unconscious or inherent prejudices in the data,
- hallucinations – AI is programmed to give you an answer so it makes conclusions based on the data available and will make up answers,
- pollution – when integrating internal data with external data,
- capture and use of personal data (including biometrics),
- digital cloning leading to fraud etc – an application called Deep Voice from Baidu, only needed 30 mins of audio to clone a person’s voice
So, its paramount you start with well understood and well-defined use cases that help solve business problems in a controlled manner.
Every business needs to understand/define the uses cases that will be relevant across industries and some that will be specific to an industry. These are by no means exhausted, it’s a starting point to explore. A few examples:
- Auto call summarisation – saves agents time by automating post call summaries
- Conversational analytics – using NLP to analyse thousands of pieces of data and find useful insights present within this data in seconds (on calls, chats, emails – across all interactions)
- Fraud detection pattern spotting in customer transactions
- Queue management optimisation predicting volumes and dynamically adjusting staffing levels to queues
- FNOL (First Notification of Loss) in insurance – capturing all associated data to expedite claim processing.
The common pitfalls organisations are making in the data area and what they could do to be better prepared
We have mentioned several pitfalls earlier but to summarise:
- Not being clear what AI means for their business (threats and opportunities)
- AI architecture is needed that defines and aligns business, application and technology components
- Lack of skills and knowledge around the new concepts and modules needed: data lakes, machine learning, data governance and cloud solutions
- Not fully defining the ‘sphere’ of data needed for each use case
- Not evaluating the inherent bias in existing data that will influence the LLM learning models
- AI projects get slowed due to not planning in data mapping, cleaning and integration needs
- Not appreciating that Machine learning requires large volumes of data from across the enterprise.
Some practical steps for organisations to consider
It’s really focused on addressing the pitfalls:
- Focus on the data strategy, governance and data needs for each AI use case
- Define your guiding principles for AI at an enterprise level
- Engage the right stakeholders early, from business owners, data engineers, Infosec and IT (your AI squad)
- Assessing the impact on jobs both changes to existing and creation of new roles
- Recognising the cyber threats – bad actors can now launch more sophisticated and effective attacks faster and at larger scale – SecOps involvement in critical
- ESG (Environmental Social and Governance) is becoming more of a focus, like bitcoin mining, the requirement for extensive computational processing power is highlighting needs for renewable energy resources.
Remember AI is not a technology solution, it is impacting ways of working for customers and employees and will need adequate change management investment.
Getting started in a controlled way
Some of the practical things that need to be done, many are the 101 steps for business projects:
- Be clear on why you are wanting to implement AI – well understood pain points and associated use cases
- Understand the risks, and develop mitigation plans (regulation, skills, data, customer and employee impacts)
- Spend time modelling the benefits, run multiple scenarios, factor in the investment costs (technology, people and change)
- Review and enhance your operating model
- Document the process changes and impacts
- Define the KPIs to measure success
- Don’t be afraid to stop, learn and iterate.
Davies – Consulting Division are experts in the AI and Automation domain, if you need help with evaluating AI in your organisation reach out to David Ilett, CX Technology Director.
