How Khatabook Integrates AI Into Creative Workflows

Every creative professional these days is experimenting with AI tools—ChatGPT and Claude for text, Midjourney for images, Runway for video, etc. While individual success stories are aplenty, integrating AI into the workflow of an entire creative team is a different challenge. 

That’s why this conversation with Mini Gautam, Creative & Brand Director at Khatabook, is so interesting. In the last six months, the creative team at Khatabook has systematically integrated AI tools into their everyday workflows and seen promising results.

Outcomes from an internal impact assessment study at Khatabook

Better variety and quality of graphic visuals with 2-3X CTR

20-30% improvement in the speed & efficiency of content creation

Better team morale because of upskilling + Time freed up to do more ideation work

Sharpening the Brief

The creative team at Khatabook supports three business pods: the Khatabook app, Khatabook Lending, and Bizanalyst, their Tally on mobile app. The team of five works on around 200 push notifications and 80 Whatsapp messages per pod per month. “At such volumes, I didn’t want the work to start feeling tedious or the team to focus on just churning out creatives,” Gautam says. “I began to experiment with AI tools to see if they could help us work smarter.”

The team started by using ChatGPT as the sixth person in the room during their brainstorming sessions. At this stage, their focus was not the output from the tool. They were more interested in understanding how GPT could influence their creative ideation and what new perspectives it brought up. 

What they discovered was that the process of framing questions and prompts for ChatGPT nudged the team to think more sharply about what they wanted to achieve and articulate these clearly. 

They also realised that GPT’s real superpower was to ingest, analyse and learn from inputs. “We already have a wealth of data from past campaigns in terms of what kind of messaging or creatives work for different pods and customer segments. I began to use these insights to produce more refined output from ChatGPT.” says Gautam.  

Institutionalising Via Templates 

For a writer, nothing is more intimidating than a blinking cursor on an empty screen, especially when there’s a deadline looming. “My writers’ strength lies in how they think and choose the right words. Why waste their talent on grunt work? I convinced them to think of AI tools as an intern they’re training. The tool gives them a basic draft or version that they can then edit and refine. This way, they never have to start from a blank page and get many BAU (Business-as-usual) tasks done quickly.” says Gautam.

1. Writing from scratch: When the team wants to create something original, they use ChatGPT as a brainstorming partner. They begin by asking it to assume a role (‘I want you to think like an expert copywriter specialising in original, creative marketing messaging’). They share inputs about the use case, audience, medium, content goal, etc. and prompt ChatGPT to come up with multiple ways of writing the message. This output is then further refined through prompts like ‘Give me options that create a sense of urgency’ or ‘How can I say this with a touch of respectful humour?’ 
2. Writing based on templates: The team analysed their best-performing campaigns to create a set of prompts customised for every content type and business pod. These briefs are pretty detailed, covering the role of the AI, main message or theme, target audience, purpose of the asset, channel/medium,  tone, inspiration, format, examples, and overall guidelines. With these prompts, ChatGPT is able to give them a version of the copy that’s well-structured and 80% in place, which is further refined by the writers.

Customising for Individual Strengths 

What I found interesting was that while the team has a template library, everything is customised to play to the strengths of individual team members. This is a great way to blend human and AI capabilities.

Say, a writer is strong at creative ideation. For a theme like ‘Grow your business’, their strength lies in coming up with different metaphors for growth. (e.g. watering a sapling, ladder to success, gears in motion, etc.) They use GPT to translate these ideas into copy, create vernacular versions, and variations for A/B testing. Contrast this with a writer who is strong in business understanding (knowledge of customer emotions, purchase journey etc.). They share their insights with GPT and use it as a brainstorming partner to generate possible creative directions. 

Everything is validated through extensive A/B testing to see what messages perform well in different contexts. The learning is fed back into the AI tools, thus closing the loop. 

Ideating Without Limitations

As a brand catering mostly to a ‘New Bharat’ audience, the team has always found it difficult to source images that represent people and scenes from India beyond the urban/metro cities. The same few stock images available are used by multiple brands catering to a similar audience. However, doing photoshoots for every use case is impossible. 

This limited the kind of visuals the team could use and they relied primarily on a library of brand graphics and illustrations.

“The prompting process for visuals is quite different,” Gautam explains. “We spend a lot of time ideating on the theme. Take for example, a caption like ‘Now serving instant growth’. First, we think about what to focus on: growth, instant, or serving? We felt that ‘growth’ is an idea we’ve done before. To try something new, we focused on ‘now serving’. The phrase instantly conjures up the idea of food. And that becomes our prompt. Something like ‘Create a realistic image of a man dressed in Indian attire, serving money on a plate’. We have to go through a few iterations of course but the upside in design is huge.”  

Gautam notes that the output created by Midjourney is 80% usable and faster than if the team had to rely only on Photoshop. They use Krea to retouch and retexturize the images and stop them from looking ‘plasticky’. 

Khatabook’s AI-aided visuals are delivering 2-3X the clickthrough rates compared to their pre-AI visuals, all constraints such as bandwidth and timelines being the same.

Persuading the Team to Embrace AI

All creative professionals wonder if AI is going to replace them someday. When she first brought up the idea of using AI at work, Gautam heard the same concerns from her team. “We undeniably live in an era of AI. It’s better for creatives to upskill and be well-versed with the latest AI tools than miss the boat. My focus was to show the team that AI is here to support them, not replace them.” 

Says Bhogi Ram, Design Lead at Khatabook, who now regularly uses AI tools in his workflows, “What AI has done is to significantly simplify the process of turning our creative ideas into reality. We are now able to bring our ideas to life more efficiently and effectively.” 

Gautam summarises, “With AI, everyone can produce mediocre work. In non-artistic creative work, what makes a designer or writer stand out is their intentionality and thinking. Use AI to reduce grunt work, so you can free up your time to focus on more impactful creative tasks. Thinkers will be rewarded more in an AI-enabled world—it’s high time for that shift.”

Challenge faced	How it was tackled
Team was hesitant to adopt AI tools due to unfamiliarity and the fear of replacement	Had 1:1 conversations and informal sessions to demonstrate the benefits and supportive possibilities of AI tools
Scaling the AI solutions from the pilot stage actually increased turnaround time	Individually tested prompts before scaling. Went deeper into one pod before scaling horizontally for other pods.
AI output was inconsistent at first.	Got separate access and personalised GPT for the creative team. Developed highly tailored prompt templates for each pod. Refining prompts regularly based on learning.

Plans for the Future

The team has more AI-aided experiments lined up, including:

1. Content Hyper-Personalisation: Customise creatives and copy to the last detail, based on customers’ geographical region, language, and psychographic traits.
2. Unique Visuals: develop a library of visual assets for Khatabook with a unique language and personalised themes, thus minimising dependence on stock images.
3. Predictive AI: leverage past data to create a high-performing asset library at a faster rate.
4. Automate QC: use AI to ensure quality control and formatting for high-volume content such as WhatsApp and push notifications, to reduce grunt work.
5. Text-to-video: use AI to simplify video production

A Tip from Mini Gautam for Creative Team Leads

Approach this as a system design challenge than a creative challenge. Don’t get stuck at the “What prompt should I use?” stage. Instead, think about how you can implement AI-aided workflows at scale so that your team can spend more time thinking versus executing stuff.