AI Gave Me Back 3–5 Days Per Project - Here’s How
Field notes on productivity, rigour and ethics from a boutique consultancy
If Medium is your go-to reading spot, you can catch this post there too! Just click here.
AI has crept into most consulting workflows, whether we acknowledge it or not. And that matters, because in this line of business trust is everything - it takes years to build, and just one weak claim to lose.
As a result, AI is a (very sharp!) double-edged sword for consultants: it can clear hours of drudge work, but it can also smuggle in errors you can’t defend.
When used well, AI can accelerate the unglamorous parts of research - transcription, first‑pass coding, basic pattern discovery - so teams can spend more time with judgment and clients. Used badly, however, it invents facts and obscures provenance.
This article is a practical, consultant‑centred way to capture the upside while defending rigour and ethics. And, most importantly, I share my estimates of how much time AI is actually saving me and how. No hype here!
Why this matters now
Two shifts are colliding in consultancy projects. First, clients expect consultants to be faster and cheaper without lowering the evidential bar. Second, quality assurance and legal teams are tightening scrutiny on data protection, reproducibility and fairness.
AI can help square the circle - but only if we design our research workflows to make every claim traceable and every risk owned. That is the bar: speed with evidence, not speed at all costs.
In everyday terms, AI is strongest at mechanical cognition (turning messy text into structured signals) and narrative scaffolding (drafting outlines, testing counterfactuals). It is weakest at extracting meaning when context is ambiguous and where things are being said between the lines.
So - I will talk about an operating model that keeps AI in its lane and puts human judgment where it belongs: on interpretation, trade‑offs and client interactions.
Before you dig deeper - have you had a chance to read the article below? It will tell you more about what consultants do and how they think!
The Non‑Negotiables of Rigour (in Plain English)
These are the things that you really can't afford to get wrong as a consultant using AI. If you mess them up, you are very likely to lose trust and credibility - which, in practice, means losing business!
So, let’s preface any of the use cases and time savings with some good practices that you should keep in mind as you embed AI into your ways of working.
Provenance: If you can’t show where a claim came from, you don’t own it. Make sure you can map claims to the source data.
Reproducibility: A future you must be able to check and validate today’s analysis. Keep prompts, chat transcripts, model names and parameters. Note that this is very easy to achieve - all you need to do in most cases is not deleting past chats!
Privacy and lawful basis: Know what personal data is processed, where and why - and communicate this clearly to participants and clients (based on your local legal framework, like the GDPR). In some cases, clients might have information that they do not want to put through AI, so make sure you consider that separately.
Fairness and equity: Check outputs across meaningful slices (e.g. roles, region) and investigate disparities. You are responsible for avoiding bias in your findings and recommendations, and I can guarantee that AI will introduce some. But AI can also help you test for this and repair any issues before the findings go to the client.
Transparency. Adopt an AI Disclosure Box in proposals and reports. Cover where AI was used, which model(s) and what tasks. Hidden automation is an ethical and commercial risk. Ideally, publish an AI use policy on your website for maximum transparency.
Now, let’s dive into my daily practices!
How AI Actually Changed My Research Workflow
Here’s how AI has reshaped my actual workflows. What follows is not hypotheticals scenarios: they are real hours saved and real risks I have experienced first-hand.
Scoping and Proposal Writing
Modern language models can parse request for proposals and invitations to tender (also known as RFPs and ITQs) and clearly outline the requirements, timeline and key areas of focus.
Language models can also summarise prior reports written by others - which can help in cases where the client wants a brief literature review - and by yourself - which can help you foreground the right and most relevant prior experience in the team or background section. On top of that, it can also suggest project methodologies based on how your previous work might map to the new opportunity’s requirements.
Time saved: 3-5 hours on the exploratory parts of proposal writing.
Key risks: You might miss elements of the client brief if you rely solely on AI; Your proposals can become flat and impersonal if the methodology is wholly AI generated and does not show clear tailoring to the client context.
Takeaway: Use AI for first drafts, never final proposals. Clients can smell generic work instantly.
Stakeholder Engagement
Consultants most often have to engage others - whether employees at a firm, experts or other stakeholders. To do so, you need to carry out two key activities: identifying the right people and preparing questions to discuss with them.
In many cases, either you or your clients will have a (relatively) clear idea about who to speak to. However, in other circumstances, you will have to identify people beyond your usual networks or the usual suspects. These days, AI can help a lot with this: if you specify the type of person you need, chatbots have become pretty good at providing suggestions. Of course, these will only be as good as your guidance and biased by what the AI ‘knows’ (i.e. usually publicly available information). Even when suggestions miss the mark, however, they are a useful starting point.
In terms of preparing questions for interviews, workshops or other forms of engagement, AI is fantastic. If you provide a clear brief regarding the scope of work and some guidance on the topics you’d like to cover, it can prepare an interview guide for you with minimal effort. The challenge here is that only experienced individuals can tell if the result is good enough. As a result, this is an amazing accelerator for senior staff, but a waste of time (and, actually, a missed opportunity to learn) for more junior colleagues.
Time saved: 1-2 hours on stakeholder identification; 2-4 hours on preparing questions (depending on the complexity of the engagement).
Key risks: Junior staff are unlikely to benefit from this too much. In the case of stakeholder identification, they often won’t know exactly if the recommendations made are sensible; in the case of question generation, you can only judge the AI results and iterate if you know ‘what good looks like’ (which requires experience!).
Takeaway: A powerful accelerator for experienced consultants, but a poor substitute for the hard-won learning juniors need.
Consulting Admin
There are lots of ‘boring bits’ in consulting projects - repetitive tasks that are necessary but not very interesting. The most annoying for me has always been generating transcripts: this used to require getting a recording from whatever software you were using, uploading it to a platform for transcription (or sending it to a transcriber) and then getting the transcript and storing it somewhere. On top of this, I also had to create a mini summary so I could remember what a single participant had said in a consultation with at least 20 more.
Now, tools like Microsoft Copilot do it all. The meeting is usually scheduled through my Bookings page via Teams, then Copilot takes care of recording, transcription and summarisation without a single click (beyond pressing the record button!). The system even manages the recording file for me, based on my organisation’s data retention policy.
Each consultancy will have its own boring bits. In these cases, AI can shine without much human input at all. This is also where pretty much anyone in the company can benefit, regardless of seniority.
Time saved: At least 15 minutes per recording - for 20 interviews, we’re talking about 5 hours!
Key risks: None that I’m aware of, this is an absolute no brainer. The only thing you need to make sure of is that you have permission from the participants to record, but this applied pre-AI as well.
Takeaway: Automate the admin ruthlessly, but always ask permission to manage and retain other people’s personal data.
Analysis and Synthesis
AI for Thematic Coding
In a typical project, I (or a colleague) used to analyse findings through in-depth thematic coding, done manually in tools like ATLAS.ti or NVivo. Today, AI can take the first pass and dramatically speed up the process. AI is excellent at identifying and organising candidate themes from transcripts or notes. This gives you a head start before you go deeper into classification and interpretation.
AI can also stress-test your analysis. For example, you can ask: “If this were false, what else would we see in the data?” This helps uncover blind spots or weak assumptions.
Note that I’ve not covered how AI can help with the analysis of quantitative data (like numbers and costs) - this is simply not something that I do personally as my work is mainly qualitative in nature. I know that AI can help with this, and I’ve experimented with quantitative analysis using Claude especially. I’m sure the opportunity is significant, though I can’t quantify it!
AI for Synthesis (Use with Caution)
Where AI struggles is synthesis, meaning turning the information you have analysed and thematically coded into insights for clients. Generative models don’t “think”: they apply patterns, and their patterns can bias your own logic. I don’t recommend using AI to generate synthesis, as it risks pulling you away from what the data actually says.
Instead, once you’ve developed your own argument, use AI as a critical friend: validate your reasoning, check for gaps and surface ancillary evidence.
Time saved: This is difficult to assess because it really depends on projects and the complexity of findings. However, I would say about 1-2 days saved on the mechanistic part of thematic coding and 1 additional day in the validation and testing part of the synthesis process.
Key risks: Enormous. This is something I recommend only people with enough knowledge of the subject matter do, to avoid errors and biased synthesis. This is the part that has to be human-mediated in all cases, with very high oversight. That said, if you know what you’re doing it’s amazing help, and saved the day a number of times!
Takeaway: Let AI clear the undergrowth, but keep your own hands on the steering wheel of meaning.
Recommendations and Reporting
I still write 98% of my reports, as AI-generated text tends to be too vague and not sufficiently incisive - it’s simply not acceptable. The remaining 2% is things like acknowledgements, scope of work, methodology, so things that AI can extract from the proposal and I can then tweak manually to fully match the final approach taken. That’s super helpful and makes the least interesting parts of report writing smoother.
AI also helps with identifying narrative devices that you can use in your reports. This means storytelling templates that help your insights and recommendations land. AI can run your project evidence and key findings against a lot of potential candidates and recommend a few that seem to map to it most effectively. Note that, to do this properly, you have to feed the key findings into the chatbot - not just ask randomly based on the data. The latter approach is almost guaranteed to get you a biased story that is not built around the most significant evidence.
For recommendations, the approach is similar to what we just covered regarding storytelling. You need to identify recommendations based on your own judgement and knowledge of the project and client context. Then, you can feed them into AI alongside the project data and ask it to:
✅ Validate them: Are they correct and evidence based?
🤔 Critique them: Is any recommendation misinterpreting the project data?
🔎 Expand them: Did you miss any key opportunities for recommendations that could be inferred from the data?
🥇 Prioritise them: Is there more evidence supporting certain recommendations?
Time saved: 1-2 hours saved on storytelling strategies, 1 day saved on writing and about 3-4 hours on prioritising and validating recommendations.
Risks: The core risk here is taking anything coming out of AI at face value. It’s the exact same as for analysis and synthesis - in the hands of an experienced colleague, this is gold; if left with someone with limited experience of the subject, it’s quite risky. Either way, it' is typical to leave the development of recommendations with more senior colleagues anyway, so these risks are more in principle when it comes to reporting and recommendations (whereas analysis is more often the job of junior colleagues).
Takeaway: AI can strengthen your recommendations, but should not create them - judgment must stay your own.
Quality Assurance
Let me foreground this: in my experience, today’s language models are terrible for quality assurance. Simply unreliable because of the inherent randomness of the technology. BUT (!) I still make most of my documents go through language models to check them before sharing them externally.
Here’s my process:
Write a document
Review it in detail
Ask AI for feedback
Ask AI to look for typos, incomplete sentences and grammatical errors
Fix them (Then, I resubmit each draft at least two more times)
Ask a colleague for feedback and review
Make appropriate revisions
Share externally
Does it look like I’m going overboard asking AI to review my drafts a minimum of three times? Well, let me reassure you that I am not - this is intentional. Generative AI tends to spot different things each time I resubmit a document - so I just do it a few times, till I feel what it comes up with is not actual errors but just AI trying to please me by ‘finding something.’ As a side note, I would do such a detailed review only for final reports, whereas interim documents that are shorter and more manageable manually I would just do a single round of quality checks with AI.
Time saved: 1-2 hours saved on spotting mistakes. Tons of frustrations saved to colleagues who would otherwise spend time fixing grammar, spelling and punctuation as opposed to substantive elements of the document.
Risks: Not many - as long as you take all recommendations made by AI with a grain of salt and think about them before implementing corrections.
Takeaway: Treat AI like an extra proofreader, not a final reviewer.
Governance Without Bureaucracy
To manage the use of AI you need a minimum system that fits your firm size. Boutique consultancies can publish a two‑page AI use policy, maintain a prompt library with examples and add the disclosure box and evidence index to their standard templates.
Larger firms should map these controls into their quality management system, set up an AI review gate and maintain a vendor register that documents storage locations and training opt‑outs.
And then there’s everything in between!
Takeaway: You don’t want to create a new religion for AI… Just a practical set of checks and balances that scale with your organisation.
Keeping the Human Edge
AI is no longer optional in consulting - this much I can say confidently. It has already reshaped the way we scope, analyse and deliver. The numbers alone speak clearly: dozens of hours reclaimed across proposals, stakeholder prep, transcription, coding and reporting.
The lesson from my own practice is simple: AI belongs in the mechanical layers of our work, while the interpretive, strategic and client-facing layers remain firmly human. Clients pay consultants for judgment, not for auto-generated slides. They trust us to see what the data doesn’t say, to surface trade-offs and to hold the mirror up to their organisation.
This means that the non-negotiables of rigour - provenance, reproducibility, privacy, fairness, transparency - must act as guardrails for every use case. Without them, speed becomes fragility. With them, speed becomes confidence.
In my view, AI isn’t diluting the craft of consulting. It is sharpening it. It is stripping out the repetitive friction and letting me focus more energy on what clients truly value: insight, clarity and action.
That is the real promise here: not working faster for its own sake, but working smarter in ways that elevate both quality and trust.
I'm Andrea, a management consultant with over a decade of experience across industry and academia. I work with commercial, non-profit, academic and government organisations worldwide, helping them capture meaningful insights through mixed methods research.
I write about practical frameworks to help you discover what others miss. My main goal is to translate complex concepts into techniques that readers can use immediately.