TenderAgents
Guide

Map your bid process before you add AI agents

A practical guide to bid process mapping: rebuild one real tender, find the waits and rework, set baselines, and decide where AI agents should help.

· 16 min read · For: UK in-house bid managers, proposals leads and BD heads considering AI agents

Most bid teams that try AI start with a tool and a sample question. The output looks impressive. Then the harder questions arrive: which documents are allowed in, who checks what comes out, and whether the time saved on drafting just moves to review. Bid process mapping answers those questions before you commit to a tool. You map one real tender from start to finish, see where the time and rework actually went, and only then decide where an AI agent can help.

This guide is for UK organisations with an in-house bid or tender function that are thinking about AI agents. It sets out a practical, seven-step method you can run with your own team in a few working sessions: pick a completed tender, rebuild its timeline, find the waits and loops, classify every step, write data rules, set baselines, and turn it all into a short pilot brief. It doubles as a straightforward tender process improvement exercise. Plenty of what you find will need a better checklist or a clear owner, not AI.

You will get copyable templates, a four-way classification for deciding what stays human, baseline measures, a worked example and a checklist. Adapt it to your own controls, buyers and procurement regimes.

Why bid process mapping comes before AI tools

Starting with the tool gets the order wrong, and the bid profession is saying so. APMP UK's Bid Writing Focus Group published ten questions bid professionals should ask about AI. Among them: where AI is already being used across the bid lifecycle, and which tools are right for your organisation. Its advice is to define your AI strategy first, so tools support real value rather than experimentation. Its own survey found that just over half of respondents said AI tools had improved the efficiency of their bid writing, while 81% saw accuracy as the key concern.

That concern about accuracy is the reason to map first. A process map shows you:

  • where time goes, split into hours worked and days spent waiting
  • where work loops back, and why
  • which steps carry legal or commercial risk, where a mistake can get a tender disregarded or commit you to something you cannot deliver
  • which steps are repetitive and checkable, so an agent can help under human review
  • what the baseline is, so any later claim about time saved rests on something you measured

Without a map, you tend to automate the most visible step, usually drafting, which is often not where the elapsed time goes.

Step 1: Choose one completed tender for your bid process review

Map a real bid, not an idealised process diagram. The gaps between the written process and the one your team actually runs are where the useful findings sit.

Choose a tender that is:

  • recent, so the people involved still remember it
  • typical of the work you want help with, not your largest or strangest bid
  • complete, with the buyer pack, addenda, clarification log, drafts, review comments and final submission still available
  • safe to use, with names, prices and anything you are not permitted to reuse redacted

One well-mapped bid beats a whiteboard session covering every bid type. Map a second type later and compare.

Who should be in the room

Keep the group small. Local Partnerships' business process mapping toolkit, written for council procurement and contract management, recommends a core team of around four to five people who know the end-to-end process, with a senior manager as sponsor who endorses the work and helps get people to attend. It also suggests giving participants specific roles, including a sequence keeper who makes sure activities are recorded in the right order, a document recorder who logs the important documents, and a decision curator who confirms which decisions are made at which point and by whom.

A bid team can run a lighter version:

Role Who usually takes it What they do in the session
Sponsor Head of bids or BD director Agrees the purpose, protects the time, accepts the findings
Facilitator Bid manager or someone neutral Keeps pace, asks the obvious questions, stops blame
Sequence keeper Bid coordinator Keeps the timeline in order and dated
Decision curator Bid lead Records every sign-off, who gave it and when
Contributors Writer, a regular SME, someone from pricing Fill in what happened in their part of the bid

Step 2: Reconstruct the tender timeline

Walk through the bid from the moment the opportunity appeared to the debrief. Record each step as a row. A shared spreadsheet works fine.

Field What to capture
Step What happened, in plain words ("extracted mandatory requirements from the ITT")
Trigger What started it (notice published, go decision, SME reply)
Owner The named role who did it
Inputs Documents, library items and people it depended on
Output What it produced and where that was stored
Touch time Hours actually worked (label as estimate if not logged)
Wait time Elapsed time spent waiting for someone or something
Rework Whether it was redone, and why
Decision or approval Any sign-off, and who gave it

Use the documents to correct memory. Email timestamps, version histories and the clarification log usually show waits nobody remembers as waits. If a touch time is a guess, write "est." beside it.

Typical stages in a bid workflow

If you are stuck for where to start, use these stages as prompts: opportunity logged, bid/no-bid, pack triage, requirement extraction, compliance matrix, kick-off, clarifications, evidence gathering, drafting, pricing, quality review, final compliance check, submission and debrief.

Capture addenda and clarification answers as their own rows. They often trigger rework across several earlier steps, and that ripple is easy to miss if you record them as a footnote.

Step 3: Find the waits, loops and handoffs

Once the timeline exists, look for patterns rather than individual mistakes. Common ones in bid work:

  • SME input requested late, or briefed so loosely it came back unusable
  • evidence hunted for during drafting because nobody knew where the current certificate or case study was
  • addenda or clarification answers reaching some writers and not others
  • the same answer rewritten from scratch because the last good version could not be found
  • review squeezed into the final day, so reviewers polished prose instead of checking substance
  • pricing and the quality narrative drifting apart until a late joint check

Mark each finding on the map, then total wait time against touch time. That tells you whether faster drafting would change the end date at all.

The map will surface more problems than you can fix at once. Bidsmithery's guide to bid process improvement suggests choosing one or two changes, giving each an owner, a completion date and a measure, then testing it across the next few suitable bids and comparing like-for-like measures where possible. Apply the same discipline here. Write the full list down, then pick.

Step 4: Classify every step before you integrate AI into the bid process

This is the core of discovery and the most direct answer to how to integrate AI into a bid process: you do not integrate it into the whole process, you place it in specific steps with a named human check. Sort every step into one of four groups.

Group What it means Typical examples
Human only Judgement, commitment or legal accountability Bid/no-bid, win themes, pricing, accepting terms, certifying, sending clarifications, submitting
Agent-assisted, human checked Repetitive work that can be checked against a source Pack triage, first-pass requirement extraction, matrix scaffolding, pulling candidate library passages, first drafts from approved material, consistency checks
Checklist or admin Mechanical steps that need a list more than intelligence File naming, word counts, portal field lists, upload checks
Fix first Steps where the process itself is broken No owner for evidence, no approved library, review never booked

Two rules help when a step is borderline.

If a mistake in the step can get the tender thrown out, keep a human confirmation on it. Cabinet Office guidance on assessing competitive tenders under the Procurement Act 2023 explains that section 19 requires a tender to be disregarded where the supplier does not satisfy the conditions of participation. A tender may also be disregarded where the supplier has breached a procedural requirement in the tender notice or tender documents. An agent can help find those conditions and requirements. A person should confirm each one. The competition documents for each tender set the actual rules, and regimes differ (Scotland has its own procurement rules, and some older procurements still run under the previous regulations), so check the pack every time rather than relying on a general rule.

If the step depends on content you do not yet trust, it belongs in "fix first". An agent retrieving from an ungoverned shared drive mostly speeds up confident errors. Fix the library before you automate retrieval from it. Our guide to building an approved bid knowledge library your team can trust covers owners, statuses and review dates.

Questions to ask of each step

When the group disagrees about a step, ask:

  • Can the output be checked against a source document in minutes?
  • Does it involve a commitment to the buyer, such as price, terms or a certification?
  • Does it need information you cannot share with an AI tool under the buyer's terms or your own policy?
  • Is it slow because the work is hard, or because an input is missing? If an input is missing, it is "fix first".

Where the tested agent roles fit

The agent-assisted group usually clusters around a few roles. In the workflow we use, a Tender Scout finds and logs opportunities, a Bid Director agent supports qualification triage, a Compliance agent produces a requirements and gaps matrix, and a Bid Writer agent drafts only from approved sources and leaves gaps open rather than filling them. Human review follows each one. You can see how these steps connect in the example agent workflow. The point of the map is to check which of those roles your own process actually needs, and in what order.

Step 5: Record data and permission rules

For each agent-assisted step, note what information it touches:

  • buyer documents, and any terms in the pack about how they may be used
  • your own approved content, and who owns it
  • personal data, such as CVs and named staff
  • commercially sensitive material, such as pricing and margins

Public buyers are paying attention to this. PPN 017, which applies to central government departments, their executive agencies and non-departmental public bodies (other contracting authorities may adopt the approach), makes clear that suppliers' use of AI is not prohibited during the commercial process. It does suggest buyers may ask suppliers to disclose their use of AI in creating a tender, and may put proportionate controls in place so suppliers do not use confidential contracting authority information, such as confidential tender documents, to train AI. It also warns that AI tools can increase the risk of misleading statements through "hallucination". Your map should show where buyer documents go, which tools touch them, and how you would answer a disclosure question honestly.

Where personal data is involved, start with the ICO's guidance on AI and data protection. The ICO states that, because of changes made by the Data (Use and Access) Act, the guidance is under review and may change, so check the current version before you set rules.

A simple data rules table for each assisted step keeps this concrete:

Assisted step Data it touches Allowed source Not allowed Human check Owner
Requirement extraction Buyer pack Current pack and addenda only Previous buyers' packs Bid lead confirms every mandatory row Bid manager
Library retrieval Approved answers Items marked approved and in date Drafts, expired case studies Writer checks source and date Library owner

If you are using a platform such as Grok Bot, treat things like workspace permissions, retrieval limits and logging as controls to validate in your own environment during the pilot. Do not assume them from a feature list.

Step 6: Set baseline measures for tender process improvement

Pick a few measures you can collect from the mapped bid and the next two or three similar bids.

Measure How to collect it Why it matters
Hours from pack receipt to a first complete requirements list Timesheet or diary, labelled estimate if needed Shows triage and extraction effort
Reviewer and SME hours per submission Ask each person after the bid, record honestly SME time is usually the scarcest resource
Number of rework loops, and their causes Count from version history and review comments Shows where quality breaks down
Defects found at final review Log missed mandatory items, wrong templates, out-of-date evidence Shows what the compliance check is catching
Share of answers built from approved library material Tag each answer at final review Shows whether reuse is governed or ad hoc

APMP UK suggests a simple scorecard to test whether AI is actually helping, covering time saved, quality gains, subject matter expert time released and compliance issues caught.

Keep win rate out of your baseline for AI decisions. Win rates move with price, competition and contract mix, and a handful of bids cannot separate those effects from a process change. Track it, but do not judge a pilot by it.

Any target you set should come from your own baseline, not someone else's headline figure. If you want to test how changes in hours per bid would affect how many tenders your team can take on, the bid capacity calculator lets you run scenarios with your own numbers. For the wider question of adding capacity without losing control, see increase bid capacity without handing over judgement.

Step 7: Turn the bid process map into a pilot brief

Discovery should end with a short document the team agrees with, not a slide deck. It should cover:

  1. the bid type in scope
  2. the two to four steps where agents will assist, with the human check for each
  3. the steps that stay human, named explicitly
  4. the "fix first" items, who owns them and by when
  5. data rules for each assisted step
  6. baselines and how they were collected
  7. what would make you stop

Stop criteria matter more than people expect. Examples: a mandatory requirement missed by the agent and not caught by the human check, or reviewer hours rising across two bids. Writing these down first removes the pressure to keep going on momentum.

That brief becomes the charter for a controlled AI pilot inside your bid team, with training, acceptance gates and handover.

Worked example (hypothetical)

A nine-person facilities management bid team maps a recent framework mini-competition.

The timeline shows writing took less time than anyone expected. Most elapsed time went on two waits: four days for SMEs to return method statement input, and two days of version confusion after an addendum changed a word limit. Evidence for a mandatory insurance requirement turned up on the last afternoon.

The team classifies its steps:

Step Classification Human check or owner
Requirement extraction Agent-assisted, human checked Bid lead confirms every mandatory row
Compliance matrix scaffolding Agent-assisted, human checked Bid manager checks against pack and addenda
SME briefing Fix first Bid manager owns a one-page brief template with a deadline
Insurance and accreditation evidence Fix first Named owners, expiry dates logged
Addendum handling Checklist Coordinator circulates and confirms receipt by every writer
Method statement drafting Deferred Revisit once the library is cleaned
Pricing, sign-off, submission Human only Commercial lead and bid director

The pilot brief covers two assisted steps, two fix-first items and one new checklist. Nothing about the outcome is guaranteed. The map simply tells the team where to start, and that drafting was not the first problem.

Common mistakes in bid process mapping

  • Mapping the official process. Map what happened on a real bid. The written procedure will tell you what should have happened.
  • Recording only touch time. Waits are usually where the elapsed time goes. Record them separately.
  • Turning it into a blame session. Look for patterns, not individual errors, or people stop telling you what happened.
  • Automating a broken step. If the step is slow because an input is missing or a library is untrusted, an agent will not fix it.
  • Skipping the baseline. Without it, any later claim about time saved is opinion.

Checklist

  • One completed, typical tender chosen and redacted
  • Small core team, sponsor and roles agreed
  • Timeline built from documents, not memory alone
  • Touch time and wait time recorded separately, with estimates labelled
  • Waits, loops and handoffs marked, with one or two changes chosen
  • Every step classified: human only, agent-assisted, checklist, fix first
  • Conditions of participation and mandatory items kept under human confirmation
  • Data and permission rules written for each assisted step
  • Baselines collected before any pilot starts
  • Pilot brief agreed by the bid lead, with stop criteria

Frequently asked questions

What is bid process mapping?

Bid process mapping is reconstructing how a tender actually moved through your team, step by step, from the opportunity appearing to the debrief. For each step you record the owner, inputs, outputs, hours worked, time spent waiting and any rework. The result shows where time and quality are lost and gives you a baseline for any change.

How long does a bid process review take?

It depends on the bid, but rebuilding one tender's timeline is usually a working session rather than a project, plus time beforehand gathering documents. Classification and the pilot brief usually need another short session. Allow longer if the bid had many addenda or contributors.

How do you integrate AI into a bid process safely?

Place AI in specific, checkable steps rather than across the whole process, and name a human check for each one. Keep judgement, commitments, pricing, certification and submission with people. Write down what data each step may use, collect a baseline first, and agree stop criteria before the pilot starts.

Which bid tasks should stay human?

Bid/no-bid decisions, win themes, pricing, accepting terms, certifying statements, sending clarification questions and submitting the tender should stay with people. Anything where a mistake could get the tender disregarded, such as confirming conditions of participation, should keep a human confirmation even if an agent helps find the information.

What should you measure before adding AI to a bid team?

Measure hours from pack receipt to a complete requirements list, reviewer and SME hours per bid, rework loops and their causes, defects found at final review, and the share of answers built from approved library material. Collect these from the mapped bid and the next two or three similar bids. Leave win rate out of the decision, because too many other factors move it.

Do buyers allow suppliers to use AI in tenders?

PPN 017 says suppliers' use of AI is not prohibited during the commercial process for central government bodies, though buyers may ask you to disclose AI use and set controls on how their confidential information is used. Each tender's documents can set their own terms, so check the pack before using any tool on it.

How TenderAgents helps

TenderAgents is a UK AI consultancy that works with in-house bid teams. Workflow discovery like this is where we start: mapping a real tender with your team, then setting up AI agents inside the process you already run, in a workspace you own and control, and training your people to use them. If that sounds useful, read more about implementing AI agents inside your tender team.

We do not sell a software licence, write your bids for you or promise wins. Your team keeps final review and submission, and agents never log in to portals or submit tenders. If you want a structured starting point for the qualification step, the bid/no-bid worksheet is free to use.

Sources

  1. APMP UK Bid Writing Focus Group — Making AI a Competitive Advantage: 10 Questions Bid Professionals Should Ask (17 July 2026)
  2. Local Partnerships — Business Process Mapping Toolkit (13 March 2024)
  3. Bidsmithery — Bid process improvement: controls that earn their place
  4. Cabinet Office — Guidance: Assessing Competitive Tenders, Procurement Act 2023 (updated 17 August 2026)
  5. Cabinet Office — PPN 017: Improving transparency of AI use in procurement (17 February 2025)
  6. ICO — Guidance on AI and data protection (under review following the Data (Use and Access) Act)

Talk to us about your bid process

A discovery discussion with your bid manager or proposals lead. We ask how a tender moves through your team today, where approved knowledge lives and who signs things off, then tell you honestly whether in-house implementation fits.