2 C
New York
Sunday, November 26, 2023

What I realized in utilizing AI for planning and prioritization: Content material technique is likely to be protected from automation

by Tom Johnson on Oct 6, 2023



I experimented with utilizing AI instruments to assist with planning and prioritizing my documentation work. Nonetheless, I discovered that the AI instruments weren’t very useful for this advanced, analytical activity. This implies that content material technique roles, which require higher-level considering like strategic evaluation and decision-making, may very well be a promising space for technical writers to specialise in as AI begins automating extra routine writing.

Planning and prioritization duties

Though not typically seen as a core tech writing activity, I spend most likely 10% or extra of my time attempting to determine what I must be engaged on amongst a bunch of attainable duties and priorities. Can AI instruments assist with planning and prioritization? That is the query I discover on this put up.

The brief reply is principally no, AI instruments aren’t that useful with this activity. That is truly a great end result, because it factors to an space that tech writers can deal with with out the worry of it being automated by AI.

Be aware: Because of the restricted availability of instruments for this activity, I wasn’t in a position to do the total testing I needed to carry out. In consequence, this put up is extra theoretical than experiential.

Inputs to contemplate when doing doc planning

When doing planning and prioritization, you could have a number of inputs to contemplate:

  • Goals at completely different ranges: These aims embrace company-wide aims, department-wide aims, group-wide aims, team-wide aims, and even private aims. Recognized by many various names, these aims are sometimes known as “OKRs,” for aims and key outcomes. Theoretically, the OKRs ought to stack, with every degree supporting the aims above it, finally rolling as much as the company-wide aims. This stacking supplies coherence throughout the group. Nonetheless, in actuality, the aims not often stack.
  • Analytics: Analytics can determine the preferred content material, content material with potential usability issues, content material with findability issues, least-visited content material, and extra. These analytics can inform priorities ultimately. On the very least, a listing of your high ten hottest pages ought to seemingly obtain extra consideration than pages nobody visits (until the one customer is a multi-million greenback companion).
  • Launch calendar: The discharge calendar identifies upcoming product releases (that require supporting documentation) for all of the groups you assist. Gathering launch data is vital to planning, particularly as a result of it identifies a few of the lengthier documentation efforts that may take a number of weeks to finish. Minor doc duties don’t often seem on the discharge calendar, however these main product releases that usually require weeks of documentation work do.
  • Backlog of bugs: Nearly each tech writing crew I’ve labored on has 100+ bugs within the backlog, with completely different creation dates, severity, precedence, and problem for every bug. The record of bugs is often a miscellaneous record in random order. Some bugs have been filed by tech writers, others by engineering groups. Some bugs is likely to be previous and now not related, whereas others are nonetheless related however previous and other people appear to be getting by simply fantastic as is.

Appropriately figuring out priorities

Past these inputs, you must navigate misplaced priorities resulting from extra-vocal requesters. Simply because a crew makes a doc request and is vocal about fixing the doc concern, it doesn’t at all times imply it is best to prioritize it. There is likely to be extra urgent points impacting customers which might be outdoors the requesting crew’s priorities.

For instance, suppose Problem A is leading to many frustrations and failures for customers, however the Alpha crew isn’t liable for Problem A. As an alternative, crew Alpha has an enormous upcoming launch for a characteristic that, in all actuality, isn’t even one thing customers need. Workforce Alpha is driving you to create documentation for his or her upcoming SuperDuper Widget, and so they’re invested in having you dedicate countless hours to creating the right docs. All of the whereas, Problem A continues to discourage customers and lose enterprise income; customers don’t have a transparent strategy to relay the issues with Problem A to you.

It takes numerous acumen and willpower to backburner Workforce Alpha’s request and deal with Problem A. Now multiply this sort of prioritization throughout 5 completely different groups, and also you’ve received an actual problem. Suppose Workforce Alpha occurs to be co-located with you, or is accustomed to the way to file docs, and even simply has doc champions. That doesn’t imply Workforce Alpha’s tasks are extra essential than Workforce Omega’s tasks, or Workforce Zeta’s, or Workforce Beta’s, and so forth. Briefly, you want an goal manner of figuring out precedence other than the one that occurs to be shouting the loudest.

Biking in important-but-not-pressing duties

There are additionally doc duties which might be essential however which don’t map to any OKRs, aren’t issues persons are asking you about, and aren’t even user-facing bugs. These bugs may very well be scripts to automate doc era and publishing, inner docs that outline processes to observe, bug templates that require requesters so as to add crucial particulars when filling bugs, and so forth. You need to cycle this work into the opposite doc work.

From all of those inputs, you must resolve what to work on. Ideally, you do that planning throughout dash planning intervals, which scrum-following groups are inclined to do each 2-3 weeks. Nonetheless, in my expertise, work rhythms are often too dynamic to do that in a neat style. Incoming requests ceaselessly disrupt and shift dash plans. Moreover, it’s uncommon that I dedicate a complete day to auditing, labeling, and prioritizing bugs in a manner that might facilitate correct planning. Even gathering analytics information, reviewing targets, and updating the discharge calendar are time-consuming duties in themselves.

Right here’s what often occurs: Each few days, I take out a clear sheet of paper and write down all of the urgent duties I’ve for the day or week. I draw a vertical line down the web page and put all of the work duties beneath a title known as work, and on the opposite facet title it house and put all my private duties. About each 2 to three days, I repeat this course of with a brand new piece of paper. It feels ridiculous and embarrassing to clarify that that is my group course of, nevertheless it’s what I’ve been doing for 15 years.

Can we use AI to assist with the planning and prioritization of doc work? Let’s see.

The experiment

Right here’s my preliminary experiment to make use of AI for planning and prioritization.

Step 1: Make sure that data is updated

To experiment, I first wanted to ensure all the knowledge was populated on-line someplace. If the duty was in my head solely, there’s no manner AI might assist with planning and prioritization. To verify all data was famous, I did the next:

  • Make sure that bugs exist for the work. I appeared via my electronic mail to ensure all doc requests have been represented by bugs within the ticketing system. Plenty of occasions, conversations begin right here however haven’t materialized into bugs but.
  • Audit all of the bugs within the backlog to ensure they’re present. For every bug, I tagged them with the product title and the precedence. I additionally made certain the titles mirrored the work.
  • Make sure that my crew targets are updated. I made certain our crew targets have been present. We make quarterly targets throughout a planning course of, however they aren’t set in stone. They will additionally go old-fashioned, get postponed, or be duties which might be already accomplished. If the duties have been accomplished, I marked them as accomplished so the AI device didn’t issue them into the plans.
  • Create a launch calendar. Though many groups have launch calendars to trace product releases, sadly I wasn’t this organized and didn’t have one. To place collectively this launch calendar, I would want to navigate our launch system for product launches; the issue is that not all groups use this technique, so it’s not dependable to trace all characteristic releases.
  • Export the newest analytics. I ought to have exported our analytics, however I didn’t do that as a result of I haven’t been monitoring my doc analytics currently.

Step 2: Create a listing of planning and prioritization guidelines

On this step, I created a listing of planning and prioritization guidelines for the AI to observe. How can the AI resolve what’s a precedence until I’ve outlined what a precedence is? Listed here are my guidelines for precedence:

  • Prioritize bugs which might be marked as P0 or P1 over P2, P3, and P4 bugs.
  • If the duty is massive (that means it would take greater than 3-4 hours to finish), counsel breaking it down into subtasks.
  • Hold time-sensitive duties with due dates in thoughts. They may must be prioritized greater than duties with no due dates.
  • Group related duties to benefit from related momentum and context.
  • Prioritize bugs which might be associated to targets (OKRs).
  • Don’t prioritize bugs which might be blocked resulting from incomplete prerequisite duties.

Step 3: Run the evaluation

After making certain my data sources have been updated, I exported all the knowledge right into a single Google Doc and plugged it into Pocket book LM.


The outcomes have been fairly poor. It appeared to select duties at random. I made numerous queries right here, attempting completely different approaches, nevertheless it was clear that the AI wasn’t good at this. The responses weren’t wherever shut.

As I famous in the beginning, I’m restricted within the instruments I can use for this. I didn’t use Bard, ChatGPT, or Claude, that are my favourite AI instruments.

Additionally, I don’t assume my record of exported bugs had all the knowledge the AI device wanted. For instance, the product tags won’t have come via.

I additionally tried to use AI to assist with planning and prioritization of my private duties, and for these experiments, I used extra mainstream AI instruments. However once more, the AI responses upset me. I believed I might use AI to extra intelligently plan my weekend duties and deal with them effectively and productively. However probably not. The advisable breakdown of duties was form of dumb and unhelpful.


This can be a subject I’ll have to revisit when extra liberal use of AI instruments is allowed inside the company, however I’m not optimistic. The issue is that planning and prioritization contain numerous context and judgment. Not all of this context is specific within the data the AI has to course of.

If each bug have been specific about when it must be revealed, the way it connects with the bigger group’s targets, who the reviewers are, what their timelines and crew are like, and extra, then maybe AI would do a greater job with planning and prioritization. However actually the work of planning and prioritization is gathering this data. It takes many hours to cross-check bugs in opposition to upcoming releases and different assist ticket priorities. There may very well be a future the place this data is extra available and may help planning, however my preliminary experiments with AI right here weren’t useful.

Be aware that AI instruments aren’t solely absent on this space. There’s a brand new AI device for aim planning known as Summit: AI Life Coach. It’s extra for private targets, although, and I feel it’s supposed to supply teaching and suggestions about your progress. I performed round with it briefly however didn’t really feel impressed to strive utilizing it extra intensively. I’m guessing that describing the progress towards my targets in a diary-like manner and getting suggestions may very well be invaluable. The mere act of holding a journal about my targets could be an important finest follow that’s seemingly a great system and strategy for goalkeeping, as it could preserve me centered and reflective. However that’s not AI, that’s only a aim journal.

The positives

Though this put up would possibly seem to be a downer, it truly isn’t. The failure of AI to completely assist with documentation planning and prioritization is definitely a great factor, because it supplies a possible space that may’t simply get replaced by AI. Maybe content material technique is an space that opens up for technical writers to emphasise, in a extra foolproof future manner.

This can be a subject I’ll have to discover in additional depth. And clearly, my planning experiments have been half-baked. However I feel content material technique is an effective space to increase on.

Expanding into content strategy

Rapidly surveying analysis

A fast survey of analysis means that analytical considering and sophisticated decision-making received’t simply be automated by AI, particularly in comparison with duties reminiscent of writing. Listed here are just a few sources that assist this concept.

Kai-Fu Lee, creator of AI Superpowers: China, Silicon Valley, and the New World Order, says “AI can not create, conceptualize, or handle advanced strategic planning” (10 jobs which might be protected in an AI world).

Within the Financial Occasions, Satyam Sharma says “Jobs involving excessive ranges of human interplay, strategic interpretation, important choice making, area of interest abilities or material experience received’t get replaced by automation anytime quickly” (Scared that ChatGPT will take your job? Expertise or jobs that received’t get replaced by AI in future).

Transferring on to extra substantial sources, The Way forward for Jobs Report (Might 2023), which highlights abilities that can be in demand, notes:

Analytical considering and inventive considering stay a very powerful abilities for employees in 2023. Analytical considering is taken into account a core ability by extra corporations than another ability and constitutes, on common, 9% of the core abilities reported by corporations. Inventive considering, one other cognitive ability, ranks second, forward of three self-efficacy abilities — resilience, flexibility and agility; motivation and self-awareness; and curiosity and lifelong studying — in recognition of the significance of employees capacity to adapt to disrupted workplaces.

Right here’s a screenshot from the report displaying the high-demand abilities (p.39):

High-demand skills

In AI Isn’t Able to Make Unsupervised Choices (Harvard Enterprise Evaluation), Joe McKendrick and Andy Thurai clarify that AI lacks human qualities like empathy, ethics, and morality which might be essential for real-world choice making. The article makes use of the “trolley downside” thought experiment for example this — a human would make a extra nuanced choice than an AI system centered solely on details and information. The authors write:

Synthetic intelligence is designed to help with decision-making when the info, parameters, and variables concerned are past human comprehension. For essentially the most half, AI programs make the fitting selections given the constraints. Nonetheless, AI notoriously fails in capturing or responding to intangible human elements that go into real-life decision-making — the moral, ethical, and different human issues that information the course of enterprise, life, and society at massive.

Extra to the purpose, in addition they word that whereas AI can create content material, it won’t know what content material people need to eat, and it may need biases in choosing what kind of content material to create. In different phrases, AI can’t be left unsupervised from human route and steering. The authors clarify:

Notably, the expertise can now create unique textual content that reads as if written by people. Developments over the previous couple of years, particularly with Google’s BERT, Open AI/Microsoft’s GPT-3, and AI21 Labs’ Jurassic-1, are language transformer fashions that have been skilled utilizing large quantities of textual content discovered on the web together with large units of information, and are outfitted to provide unique textual content — sentences, weblog posts, articles, brief tales, information stories, poems, and songs — with little or no enter from people. These will be very helpful in enterprise duties reminiscent of conversational AI, chatbot response, language translations, advertising and marketing, and gross sales responses to potential clients at an enormous scale. The query is, can these AI instruments make the fitting selections about the kind of content material folks search to eat, extra importantly produce unbiased high quality content material as unique as people can, and is there a danger in machines choosing and producing what we’ll learn or view?

For instance, with documentation, tech writers would possibly pursue extra proactive, decision-directed documentation reasonably than adopting reactive conduct by which they reply to doc requests solely.

Even wanting again a bit additional, in a 2017 McKinsey report titled Jobs misplaced, jobs gained: Workforce transitions in a time of automation, the researchers say strategic decision-making jobs are among the many harder-to-automate occupations that may proceed to be wanted sooner or later, whilst automation impacts a considerable share of labor globally:

Automation may have a lesser impact on jobs that contain managing folks, making use of experience, and people involving social interactions, the place machines are unable to match human efficiency for now.


Primarily based on this fast survey of analysis, it appears clear that analytical considering and sophisticated decision-making are areas that may battle to be automated. As technical writers, if we wish to survive the AI wave and keep away from layoffs resulting from automated bots that may abruptly write all of the docs (for instance, see CEO roasts human employees he fired and changed with ChatGPT), then we would shift extra into content material technique roles — the form of technique that entails analytical considering and decision-making, not simply content material manufacturing.

This would possibly imply specializing in documentation technique greater than documentation bugs, on assembly with key stakeholders to assemble insights and evaluation reasonably than simply fulfilling requests to put in writing docs, and synthesizing the fitting doc plans primarily based on analytics, consumer suggestions, and product roadmaps reasonably than simply responding to product crew requests to put in writing docs for his or her subsequent launch.

About Tom Johnson

Tom Johnson

I am an API technical author primarily based within the Seattle space. On this weblog, I write about subjects associated to technical writing and communication — reminiscent of software program documentation, API documentation, AI, data structure, content material technique, writing processes, plain language, tech comm careers, and extra. Try my API documentation course if you happen to’re on the lookout for extra data about documenting APIs. Or see my posts on AI and AI course part for extra on the newest in AI and tech comm.

In case you’re a technical author and wish to carry on high of the newest traits within the tech comm, you’ll want to subscribe to electronic mail updates under. You too can study extra about me or contact me. Lastly, word that the opinions I categorical on my weblog are my very own factors of view, not that of my employer.

Supply hyperlink

Related Articles


Please enter your comment!
Please enter your name here

Latest Articles